Introduction to article:
It is very difficult to divorce cutting edge tech from the corresponding terms used in patent statute. Without understanding one, you cannot understand the other.
The Court of Appeal, in this appeal #CA2004000036 #EmotionalPerception vs Comptroller of Patents, is therefore tasked with acquiring an understanding of both. To some extent it’s a chicken and egg situation.
On Day 1 of the hearing, it took counsel for the UK Patent Office 1.5 hours to introduce the concept of ANN technology and the invention. This potentially reflects the difficulties faced with deciding on exclusions for new technologies, and the complexity of AI. Today, Emotional Perception hopefully settled some of the muddied waters to support an objective assessment based on the merits of the case and the potential deficiencies in either existing case law [and how it is to be applied] or the confusion that may have arisen from the Patent Office’s line of argument.
We are perhaps likely to see a decision that confirms the nature of “programs for computers” and whether and to what extent ANN technology is patentable in the U.K.
Objective open-mindedness is the order of the day for both sides and the judiciary. Such an approach, leaving any preconceived ideas at home, allows for clarity and longevity in any decision. We expect the decision on this likely important #AI #ANN #patent case to take some time since judgement was reserved, but we would not be surprised if the decision was with us by the end of June.
Skeletons and aspects of the oral submissions have today been made available on our website.
Download Emotional-Perception-AI-CoA-Skeleton.pdf (PDF)
Download IPO-Emotional-Perception-Skeleton-8-3-24.pdf (PDF)
Download CoA-Subs-v-4.1-For-Bruce-BCD-Comments.pdf (PDF)
Skeleton Argument for the Appellant (Court of Appeal)
Page 1 of 29
Appeal No: CA-2024-000036
(CH-2022-000144)
IN THE COURT OF APPEAL
ON APPEAL FROM THE BUSINESS AND PROPERTY COURTS
OF ENGLAND AND WALES
INTELLECTUAL PROPERTY LIST
PATENTS COURT (SIR ANTHONY MANN)
ON APPEAL FROM THE DECISION OF THE COMPTROLLER GENERAL OF PATENTS DATED 22 JUNE 2022 (BLO/542/22)
IN THE MATTER OF THE PATENTS ACT 1977
and
IN THE MATTER OF UK PATENT APPLICATION GB1904713.3 IN THE NAME OF EMOTIONAL PERCEPTION AI LTD
______________________________________________________________
THE APPLICANT/RESPONDENT’S SKELETON ARGUMENT
ON APPEAL TO THE COURT OF APPEAL
______________________________________________________________
MARK CHACKSFIELD KC EDMUND EUSTACE
8 New Square Lincoln’s Inn
Instructed by Bruce Dearling of Hepworth Browne
For and on behalf of the Applicant/Respondent
Page 2 of 29
A. I
NTRODUCTION
1. The global music streaming industry is projected to grow from $25 billion in 2021 to $103 billion by 20291. Spotify, a world leading music streaming service, alone had an R&D spend in 2021 of €912
23 million. The global video streaming industry is larger still.
2.The Application GB1904713.3 (‘the Application’) relates to AI, or, more specifically, to adaptive neural networks (ANNs), their training and application in technology that gives rise to particular practical benefits to these industries, amongst others. The Judgment is of substantial commercial importance to the Applicant (‘EPAI’) and the industry more broadly, and has attracted enormous interest.
3. By the Judge’s decision he considered the nature of the inventions claimed in the claims of the Application, and having done so concluded that an ANN is not a ‘program for a computer’ at all4. He held that the invention is cast at a level of generality above the specific manner of its implementation,
56 whether in hardware or as a software emulation. The exclusion was therefore not engaged. For good measure the Judge also held that, were he to be wrong about that, then in any event the invention as claimed demonstrated a ‘technical effect’ in accordance with the established case law and was not for a ‘program for a computer … as such’7.
4. By its appeal, the UKIPO seeks to argue that the Judge was wrong so to conclude. The UKIPO’s approach is, however, wrong in law and incorrect on the facts. The UKIPO’s approach is to strip out from the claimed invention the inventive concept which underlies it, and then to say that it
8 is only that idea which is to be compared to the statutory exclusions. But that is to take the ‘contribution approach’ which was ruled to be wrong by this Court in Genentech, Merril Lynch and Aerotel. The UKIPO’s approach also asks whether the invention is ‘computer implemented’, an ill-defined
1 https://www.grandviewresearch.com/industry-analysis/music-streaming-market. 2 https://www.statista.com/statistics/813751/spotify-research-development-costs/. 3 https://www.grandviewresearch.com/industry-analysis/video-streaming-market#:~:text=The%20global%20video%20 streaming%20market,21.3%25%20from%202022%20to%202030. 4 Judgment [58], [61]. 5 Judgment [56]. 6 Judgment [61]. 7 Judgment [76] and [78]. 8 See in particular at [32] – [34].
1.
Page 3 of 29
phrase which is not used in the Act, and positively criticises the Judge for looking for a ‘computer program’9; despite that being exactly what the statute requires10.
5. The UKIPO also seeks to withdraw (‘interpret’ it says11) its concession in front of the Judge that in a hardware ANN at least there is no computer program at all. EPAI does not object to this withdrawal on formal grounds, since it does not wish it to be said that the decision from this Court is based on that later-disputed concession, but EPAI says that the UKIPO’s original concession was in fact rightly made, and its reversed position on this appeal is artificial and wrong.
6. The UKIPO further incorrectly understands how the ANN of the claims operates. The UKIPO states repeatedly that the ANN is making subjective or emotional judgments12, but that is wrong. In use the ANNs of the invention make purely mechanical assessments of the physical properties of tracks. That the results are better inter alia because of the improved track recommendation is a benefit of the invention, not how the ANN functions.
7.EPAI submits that the correct approach in law to a case such as this is:
a.First to ask the Aerotel questions. These serve correctly to orientate oneself for the statutory question, but do not themselves answer that statutory question;
b. Second to consider whether the invention is within the exclusion, which itself has two aspects:
i) Does the invention involve a “program for a computer” at all?
ii) If it does, then to consider whether the invention is a “program for a computer … as such”.
8.This latter step (b.(ii) above) requires a consideration of the invention as defined in the claim, and asking whether that invention is just (“as such”) a computer program. It is well established in the cases that an invention which is properly characterised as a method or system for doing something (itself not excluded under the Act), such as designing a drill bit, canning soup, sending an internet alert message or analysing an image file, is not excluded just because it may be implemented using a computer program. The cases also show that even where a claim contains some integers which might be said to be a computer program, if it also contains elements which are external to that computer program it
9 UKIPO’s skeleton §27. 10 “… as such”. 11 UKIPO’s skeleton [63]. 12 E.g. UKIPO skeleton §§8, 40, 41 and 56.
2.
Page 4 of 29
will also normally satisfy the statute. This is because in each case the invention claimed is not to a computer program “as such”: in the first case the invention is at a level of generality above the manner of its implementation, and in the second the invention contains elements in addition to those which might be described as a computer program. This is often referred to in the cases as the search for a ‘technical effect’, but that language can be rationalised as just a – possibly slightly distracting – paraphrase of the statutory test.
9. Once the proper questions in law have been identified, it quickly becomes clear that the Judge was right for the reasons he gave. It also becomes equally clear that the UKIPO’s case, based as it is on the long-rejected ‘contribution approach’, is wrong.
10. In this skeleton, we start with a fairly brief introduction to the technology and the Application, then we examine the English law in some detail, followed by a brief summary of the approach in the EPO and other major EPC jurisdictions. Finally we draw this all together and explain why the Judge was right, and the UKIPO’s appeal is flawed and wrong.
B.TBEE
ECHNICAL ACKGROUND AND THE XPERT VIDENCE
11. The technology in this case is at times complicated. The Judge outlined the nature of ANNs in [14] – [18]. They are further described in Annex 1 to our High Court Skeleton (which the Court is invited to read) [SUPP/1/pp.42-46], and further detail is found in Prof. Pardoe’s report at §§19 – 43 [SUPP/8/pp.212-217].
12. An ANN consists of a collection of nodes (‘neurons’), arranged in layers and coupled to the
13 neurons in the layer above and below by connections (‘links’). Each neuron receives inputs from neurons in the layer above, and passes on its output to neurons in the layer below. Often each neuron in one layer will have connections from every neuron in the layer above, and to every one in the layer below, but this is not always the case.
13. The links have ‘weights’ and the neurons ‘biases’. Some neurons will feed inputs onto others strongly (by their link having a high weighting factor), whilst others may do so more weakly, or even negatively. Each neuron typically also has a bias associated with it (i.e. a positive or negative offset).
14. In a trained ANN this complicated internal structure takes inputs which cascade through the layers of neurons to provide useful outputs: to identify similar images in response to a user request, or to identify drug candidate molecules, for example. The ‘training’ of an ANN is an initial iterative
13 Other than in the top and bottom layers, for obvious reasons.
3.
Page 5 of 29
process by which its internal structure is amended and refined using test data to produce the working product, which the Judge describes at [16] – [17].
15. A point which is important to appreciate is that ANNs, comprising this structure of nodes, links, weights and biases, can exist in a hardware form, or be emulated using software. As the Judge puts it in [14] and [18], the former is a physical box with electronics in it, whilst in the latter a conventional computer runs a piece of software which enables the computer to emulate the ANN as if it were a
14 hardware ANN. Hardware solutions for training and constructing ANNs are commercially available, and there is no practical differences between ANNs implemented in software or in hardware15. Software and hardware implementations are thus the same in terms of the architecture, weights and biases, and the outputs produced16. It is just a question of which is a more convenient or efficient way to implement the ANN in a particular scenario17. This is important as it is informative of what an ANN actually is, and also of the level at which the inventions of the claims are cast (which refer to ANNs themselves, but above or agnostic of the manner of their implementation).
16.Further, even where emulated on a computer, an ANN is both in form and nature very different from a computer program. A computer program takes the form of serial logical ‘if-then’ type statements defined by the human programmer and which define exactly what it is that the programmed computer does. A software supported ANN, however, does not rely upon a series of instructions of that type, but instead is an emulated machine supported on a software platform.
17. The core utility of ANNs (including the ANNs of the Application) lies in their ability to address problems which would be intractable to computer programming. To write a computer program requires the programmer to understand the problem at hand and the manner of its solution, from which to formulate a series of logical commands for the computer to follow. Where the problem is itself intractable, then a computer programmer (and computer program) cannot help – how, rhetorically, can a programmer write a program when the solution to the problem is not even understood? By contrast an ANN is a machine-based system which is able through iterative training on a (usually very extensive) data set to create for itself an internal structure which solves the otherwise intractable problem. The solution to the problem is embedded into the structure of the ANN, i.e., in its links, nodes, weights and biases; that structure being the result of iterative changes made during the training process. In many cases, certainly including those of the Application, there may be enormous
14 Judgment [18]. 15 Judgment [18]. 16 Judgment [53]. 17 Pardoe [43].
4.
Page 6 of 29
insight in how the training objectives and the training data are used to cause the ANN in question to evolve during training; but even then the computer scientist does not know in advance what structure the ANN will adopt nor what patterns and relationships the ANN will ultimately pick up in the data, and indeed it is normally impossible even once the ANN is trained to understand how it is approaching the problem in question to produce the answers given.
18.It is for those reasons that the Judge was right to say that a claim to an ANN is not a claim to a program for a computer, and did not engage the exclusion at all (indeed, was a technical thing in itself18). At the level of the claim, an ANN is a machine the structure of which provides its function. That that thing can be implemented as a physical ANN or emulated on a software platform does not change the nature of the claimed invention, and does not matter. The ANN of this Application, and likely ANNs in general, are outside of the exclusion.
19.In addition the claims of this Application contained wider technical features, and the Judge was right to conclude that they would also have been allowable on this alternative basis.
C.TA
HE PPLICATION
20. The Application is technically dense and complicated. It can be found at [SUPP/6]. The Judge summarised it at [4] – [13]. We provided a more detailed explanation at Annex 2 to our High Court skeleton [SUPP/1/pp.47-57], which the Court is invited to read, and is it further addressed by Prof Pardoe at [44] – [61] [SUPP/6/pp.217-221].
21. In briefest outline, the position is as follows. As the Judge records at [4], the prior approach to making music recommendations tended to be limited to recommendations within fixed types of music, based on categorisation by human tagging or investigating human playlists. The result was that recommendations tend to be limited by genre (hard rock playlists beget hard rock songs, even if there might be a musically very similar song in a different genre – Wagner perhaps). By the innovative pairwise training described by the Judge at [8] – [12], the trained ANN of the invention obtains a structure which can be used to analyse the physical properties of tracks (rhythm, tonality, timbre etc) to identify other tracks which will in fact have similar musical or emotional feel19. Thus an ANN of the invention is able to make recommendations based on only physical attributes but which can cross genres – with the invention if you put in hard rock you may indeed have a musically similar Wagner track returned. As the Hearing Officer held, and the Judge recorded, the invention results in better,
18 Judgment [78]. 19 This is referred to as the ‘semantic’ feel in the Application, due to it being derived from a written description.
5.
Page 7 of 29
more accurate recommendations20, provided also by a system which is faster and simpler than the prior art21.
22. Note, however, that what the ANN is doing in use is purely objective – it performs a physical analysis of the physical properties of the track, and outputs a recommendation based also on the physical analysis of the physical properties of the recommended track. Whilst the result is a (subjectively) better recommendation, that (subjective) improvement is only a benefit of the invention – it is not how the ANN of the invention actually works. The ANN of the invention does not make value judgments, nor does it have subjective views on musical taste. Nor is it unusual in patent law that the benefit of an invention is a happier user – to the contrary that is true of many, many patented inventions: improved massage machines, better car suspension, clearer audio equipment, improved video or audio codecs in telecommunications, etc., all have a (subjectively) happier user at their heart.
23.Note also that the claims are to a system (claim 1) or a method using that system (claim 4), in each case containing inter alia an ANN, a database, a communications network and a user device to which
22 the recommendation message is sent. The claims are not merely to the idea of the pairwise approach to the training objectives, even if that is a significant (and technically very clever) aspect of the inventive step underlying the claimed inventions.
24. Finally, whilst to date it has been EPAI’s position that it was unlikely that claims 1 and 4 needed to be considered separately, it has always reserved its rights in this regards. Given the UKIPO’s approach on this appeal it is possible, albeit we still think unlikely, that a distinction may arise23.
D.TL
HE AW
25.The law in this area is not, it is probably fair to say, always of complete simplicity. Nonetheless, as we have explained above, the caselaw can be synthesised into a fair and logical structure, which properly has regard to the wording and principles of the Act. The key to understanding all of these cases, as it ever is in patent law, is properly to focus on the identification of the claimed invention itself.
20 Decision [48], [53] – [54] and [69]. 21 Decision [49] and Judgment [13]. 22 At the hearing in front of the Judge there was a discussion as to whether the recommendation message contained the recommended music file, or whether a link would be enough. It was common ground, however, that this did not affect the decision and can be left open. 23 See in particular §87 below.
6.
Page 8 of 29
26. What is excluded under s1(1)(d) and 1(2)(c) of the Act24 are inventions which are ‘programs for computers … as such’. Under the English approach (the ‘technical effect approach’25 (with ‘rider’)) the question asked is whether the invention as defined in the claim makes a ‘technical contribution’26 or has a ‘technical effect’. Within the EPO a different approach is taken, and the focus is upon the presence of any piece of physical hardware (sometimes called the ‘any hardware approach’). These in a sense may come full circle when, under the consideration of obviousness under Art. 56 EPC, the EPO will exclude features which do not contribute to the technical character from its consideration of Arts 54 and 56. But nonetheless the approaches are quite different.
27. Furthermore, as Lewison LJ commented in HTC v Apple27, our Courts (and the Technical Boards of Appeal) focus not on the words of the statute, but instead upon a search for a technical contribution or effect. As Lewison LJ says: instead of arguing about what the legislation means, we argue about what the gloss means. But it is important to understand what the gloss is really about, which is the proper identification of the claimed invention and whether, properly characterised, it is (or contains aspects which are) other than a ‘program for a computer … as such’. There might be something to be said for moving on from that specific language to something more focussed on the wording of the Act and the invention, since the ‘technical contribution’ language does not always add clarity, but that question does not affect the substance of the law.
28. Rather than seeking clarity in this case, however, the UKIPO seeks to add yet further glosses to the statutory test, by recasting the exclusion as appertaining to ‘computer implemented inventions’, without precision as to what that means and why that different language should be used when the Act refers to a ‘program for a computer … as such’. The reason the UKIPO does this is to try to gloss over its problem that an ANN is not a computer program at all. We therefore resist that invitation, and focus on the statutory language herein.
29. EPAI’s position is that the Judge correctly summarised the law, and reached the right conclusion. Although cognisant of the divergence between the English caselaw and that in the EPO, and bearing in mind also Actavis c Merck28, for the purposes of this hearing our primary case (subject
24 Patents Act 1977 (as amended). 25 With the ‘rider’ that a novel or non-obvious improvement in an excluded category does not count as a technical improvement. 26 Given that Aerotel rejects the ‘contribution’ approach, the choice of language here is unfortunate. 27 [2013] EWCA Civ 451, at [143] – [147]. 28 [2008] EWCA Civ 444, [2008] RPC 26.
7.
Page 9 of 29
to one tweak, which we think proper but which likely does not matter for the Court’s decision29) is that the English law is correct and should be applied in this case. The UKIPO agrees30. Towards the end of this skeleton, we nonetheless also briefly summarise the approach of the EPO, in case it is of interest to the Court, and formally we reserve our position on the law were this case to go further.
The Statute and EPC
30.The Court will be familiar with subs. (1) and (2) of s.1 of the Act, which are among the statutory
31 provisions that are specifically mentioned in s.130(7) of the Act and so should be construed consistently with the corresponding provisions of the EPC. In its current form (EPC 2000) those provisions are found in Art.52.
The Nature of the Exclusions
31. Whilst the exclusions constitute what Jacob LJ described in Aerotel32 as a ‘disparate group’, both here and in the EPO some degree of consensus is emerging about them, perhaps bolstered by the addition to Art. 52 EPC 2000 of the words ‘in all fields of technology’ as we discuss below. The
33 modern understanding is set out by Birss J (as he was then) in Halliburton at [26] – [27], where he stated in particular that “at the heart of the law is the consistent principle that an inventor must make a contribution to the art (that is to say the invention must be new and not obvious) and that contribution must be technical in nature (susceptible of industrial application and not within one of the areas excluded by Art.52(2)). … In my judgment the observation remains true today taking full account of Aerotel and Symbian as well as the other cases decided at the Patents Court level since.”
32.The EPO takes much the same view. In G1/19 the EBA commented on the limited nature of the exclusions under Art 52(2)(c). It noted in [24] and [25] that the words ‘in all fields of technology’ were inserted into Art. 52 EPC 2000, to bring it into line with Art. 27 TRIPS, and expressed the intent of TRIPS not generally to exclude programs for computers from patentability. Accordingly, the TBA concluded that this was a bar to a broad interpretation of the exclusions listed in Art. 52(2) EPC.
29 Concerning the effect of the amendments in 2008 to bring to Art. 52 EPC into line with the protection afforded to the “all fields of technology” requirement of Art. 27 TRIPS, see paragraph 33 below. 30 UKIPO skeleton §79. 31 Which states “130(7) … it is hereby declared that the following provisions of this Act, that is to say, sections 1(1) to (4) … are so framed as to have, as nearly as practicable, the same effects in the United Kingdom as the corresponding provisions of the European Patent Convention … have in the territories to which those Conventions apply.” 32 Aerotel Ltd v Telco Holdings & Ors [2006] EWCA Civ 1371, [2007] RPC 7. 33 Halliburton Energy Services Inc’s Patent Application [2011] EWHC 2508 (Pat), [2012] RPC 12.
8.
Page 10 of 29
33. In Aerotel at [9] Jacob LJ had taken the view that there was no indication as to whether the exclusions should be read narrowly or widely. Aerotel was, however, a case decided under the EPC 1973, without the wording ‘in all areas of technology’ which was added by the amendments to create EPC 2000 and referred to by the Board in G1/19. Whilst we think that this observation unlikely
34 matters to the decision in this case, in terms of a wider understanding of the law in this area we submit it would now be correct to read the Act with the amendments to EPC 2000 and the statements
35 in G1/19 in mind, and that the exclusions should be viewed as narrow.
E. T”?
HE MEANING OF A PROGRAM FOR A COMPUTER
34. This is, as far as we are aware, the first case in which there has been a close focus on the meaning of a ‘program for a computer’ under the Act. Before the Judge, as it did before the Hearing Officer, EPAI relied upon a set of dictionary definitions set out by the Judge at [36]: for example “a set of instructions for a computer to perform some task” and “a set of instructions that makes a computer do a particular thing”. As the Judge noted at [36], the UKIPO accepted these definitions – this not being something the UKIPO seeks to resile from on this appeal. This was also supported by Prof. Pardoe’s description of a computer program in his report at [22], cited by the Judge at [37]. We further note Jacob LJ’s description of a program for a computer in Aerotel at [31] as “a set of instructions” and a “series of instructions”, and also Gale36 at p.324 “a series of instructions capable of being followed by a cpu to produce a 40 – 50 desired result”. This is also consistent with the policy which Jacob LJ identified as underlying the program for a computer exclusion, namely overlap with protection under the law of copyright; which posits that the computer program comprise a human-created set of written instructions, i.e. at the level of idea/expression protected by copyright.
F.TEL
HE NGLISH AW
Aerotel, Merrill Lynch and Genentech
35. Aerotel is well known and well cited for its eponymous test. It is informative, however, to start with the analysis which underlay the final decision, which the Court included in an Appendix from [78] onwards of that decision.
34 Although insofar as necessary EPAI would rely upon it. 35 For completeness we refer the Court to the first instance decision in Symbian [2008] EWHC 518 (Pat) at [46] – [48] where Patten J rejected a much more extreme argument that the changes to Art. 52 EPC in the light of TRIPs allowed the High Court to reject Aerotel wholesale and adopt the approach of the EPO ‘trio’ of cases – Pension Benefit, Hitachi and Microsoft. See also Halliburton at [79]. That is not our case here – only that the exclusions should now all be viewed as narrow. 36 Gale’s Application [1991] RPC 305.
9.
Page 11 of 29
36. The Court’s analysis starts at [78] with the then-leading decision in Merril Lynch37. Here the Court recorded that in his first instance decision in that case38 Falconer J had rejected the application on the basis of the so called ‘contribution’ approach; namely to ask whether the inventive step resides
39 only in the contribution of excluded matter. Given the approach of the UKIPO in its skeleton, it is important to set out Falconer J’s reasoning here (emphasis added):
“In my judgment, where an invention for which a patent is sought involves any of the matters specified in paragraphs (a), (b), (c) and (d) (“an excluded matter”), on its proper construction the qualification in section 1(2) does require of the Patent Office an initial enquiry and assessment as to whether the inventive step resides in the contribution of that excluded matter alone – if only that contribution of the excluded matter is the inventive step, the invention is not patentable by virtue of section 1(2). I endorse, therefore, the view of the principal examiner that in the determination of whether or not an application relates to an excluded thing it is necessary to take into account whether the non-excluded features are already known or obvious.” 37.Thus the approach of Falconer J was to strip out from the claim all that was known (or obvious), and ask if that which remained – that which provided the inventive step – fell within the exclusions.
38. In [80] of Aerotel the Court considered the decision of the Court of Appeal in Genentech40, which was heard after the first instance decision in Merril Lynch but before its appeal. As the Court set out in [80] – [81], the Court of Appeal in Genentech noted a difference between the approach of Falconer J and an important decision from the TBA41 in Vicom42 (discussed below) which made it clear that the mere fact that a computer is involved is in itself no objection – what is decisive is the technical contribution made by the invention as defined in the claim when considered as a whole.
39.The Court of Appeal in Genentech preferred the approach in Vicom, rejecting the ‘contribution’ approach of Falconer J in Merril Lynch. The Court in Aerotel (at [82]) cited and approved sections of Dillon LJ’s and Whitford J’s judgments in Genentech in which they pointed out that the contribution approach gave illogical and unworkable results. See also [34] of Aerotel where the Court summarised the point: “suppose the “discovery” of the genetic (nucleotide) sequence which encodes for a particular valuable protein and a claim to a novel cloning vector incorporating that sequence. If you “strip out” the discovery all you have is a known sort of cloning vector. So all that has been added is the discovery – since that is unpatentable the claim is unpatentable
37 Merril Lynch’s Appn (1989) [1989] RPC 561. 38 Merril Lynch’s Appn (1989) [1988] RPC 1. 39 As summarised by the Court in Aerotel at [26]. 40 Genentech’s Patent [1989] RPC 147. 41 Technical Board of Appeal of the EPO. 42 Vicom (1986) T208/84, [1987] OJ 14.
10.
Page 12 of 29
too. That cannot be right – it would exclude many valuable inventions. Hence the contribution approach is wrong.”. That has much in common also with the analysis of Nicholls LJ in Gale at 323 – 324. 2941
40. Although in [35] of Aerotel the Court questioned the correctness of this analysis they (a) stated that they were bound by it, and (b) their proposed solution to this in Aerotel would have been to reveal the fallacy in the ‘contribution approach’ in a different way, not to suggest that Genentech was wrong to reject the contribution approach at all.
41. This is important, as the contribution approach rejected in Genentech should be compared with the approach taken by the UKIPO on this appeal, where they repeatedly state that certain features of the claims can be ignored because they were known in the art, or not part of the inventive step, and say that the inventive remainder is all that can be focussed upon for the purpose of the exclusions43. That is to apply the contribution approach, but that approach is wrong in law and has been rejected.
42.Returning to Aerotel, at [83] the Court summarised the decision on appeal in Merrill Lynch, which applied the reasoning in Genentech and Vicom. The Court then proceeded to consider a number of
44 other cases, including from [121] the decision of Peter Prescott QC in CFPH’s Application. The Court picked up on the subject of the identification of a ‘technical contribution’ in [125], which we set out below along with [104] of CFPH of the citation which followed it (emphasis added):
125. Mr Prescott held that the policy against the patenting of computer programs as such must extend to precluding patenting a computer when running under the instructions of a new program. Otherwise the exclusion is pointless. The test was whether the claimed subject-matter was new and non-obvious merely because there was a computer program. If on the other hand the computer program was merely a tool for doing something in a better way it was not caught by the exclusion unless that “something” was itself excluded, e.g. as being a business method. … [104] But the mere fact that a claimed artefact includes a computer program, or that a claimed process uses a computer program, does not establish, in and of itself, that the patent would foreclose the use of a computer program. There are many artefacts that operate under computer control (e.g. the automatic pilot of an aircraft) and there are many industrial processes that operate under computer control (e.g. making canned soup). A better way of doing those things ought, in principle, to be patentable. The question to ask should be: is it (the artefact or process) new and non-obvious merely because there is a computer program? Or would it still be new and non- obvious in principle even if the same decisions and commands could somehow be taken and issued by a little man at a control panel, operating under the same rules? For if the answer to the latter question is ‘Yes’ it becomes apparent that the computer program is merely a tool, and the invention is not about computer programming at all. It is about better rules for governing an automatic pilot or better rules for conducting the manufacture of canned soup.
43 UKIPO’s skeleton at e.g. §§11, 32 – 34 and 54. 44 CFPH’s Appns [2005] EWHC 1589 (Pat), [2006] RPC 359.
11.
Page 13 of 29
43. That analysis is pithy and compelling (and plainly Jacob LJ would have said so had he thought otherwise). At its core is the proper identification of the claimed invention. If the invention is merely the computerisation of something that is known, then that is within the exclusion, but if – analysed correctly and at the right level of generality – the invention is for a product or process which is merely implemented on a computer, then the invention is that product or process (and not ‘a computer program … as such’) and is patentable. We will see below that the EPO said the same in Vicom at [15] – [16].
44. Returning to the start of Aerotel, in [38] the Court recorded that it was bound by inter alia Merril Lynch and Genentech and it set out the well known test at [40]: “The approach is in 4 steps: (1) properly construe the claim, (2) identify the actual contribution; (3) ask whether it falls solely within the excluded subject matter; (4) check whether the actual or alleged contribution is actually technical in nature”.
45. The first step was further described by the Court in Aerotel at [42] “42. No-one could quarrel with the first step – construction. You first have to decide what the monopoly is before going on the question of whether it is excluded. Any test must involve this first step.”
46.As to the second “43. The second step – identify the contribution – is said to be more problematical. How do you assess the contribution? Mr Birss submits the test is workable – it is an exercise in judgment probably involving the problem said to be solved, how the invention works, what its advantages are. What has the inventor really added to human knowledge perhaps best sums up the exercise. The formulation involves looking at substance not form – which is surely what the legislator intended.” Thus one must look at the substance not the form, but the question is of the substance of the claimed invention, not some kind of extracted summary of its
45 inventive step, as the UKIPO approach it in its skeleton. That is what the discussion of Merril Lynch and Genentech was about.
46 47.The third step was addressed at [45], and as Lord Neuberger’s commented in Symbian at [8] & [12], it is really just to ask the statutory question47:
45. The third step – is the contribution solely of excluded matter? – is merely an expression of the “as such” qualification of Art.52(3). During the course of argument Mr Birss accepted a re-formulation of the third step: Ask whether the contribution thus identified consists of excluded subject matter as such? We think either formulation will do – they mean the same thing.
48. Finally in Aerotel at [46] the Court commented that the fourth step should not be necessary because the third step should have covered that. It was included for consistency with Merrill Lynch, and although we are now 16 years further down the line we are not aware of any case where step 4
45 n.43 supra. 46 Symbian v Comptroller-General of Patents [2008] EWCA Civ 1066, [2009] RPC 1. 47 It is in that sense rather like the Pozolli approach to obviousness which, as Lord Hodge noted in Actavis v. ICOSat [60], serves to orientate the Court for the statutory question rather than answer the statutory question itself.
12.
Page 14 of 29
has mattered. There might be something to be said for simplifying matters and putting that one to bed.
49. Summarising, it is important to appreciate what Aerotel is really about. First, it contains a rejection of the ‘contribution’ approach and an affirmation of the ‘technical effect’ approach. Secondly, it provides a structure for approaching the statutory question. The first two Aerotel questions serve correctly to orientate the Court for the statutory question, which is addressed in question 3 (and in a sense question 4); but Aerotel does not really help answer the statutory test.
Vicom
50. Vicom is an important case from the EPO, approved and applied in many English decisions including Genentech, Merrill Lynch, Aerotel, and numerous following cases. It involved an application which concerned a method of digitally filtering image files using a novel mathematical algorithm. As such it was paradigmatically both computer-based and mathematical. It was granted nonetheless.
51.The Board considered and rejected the mathematical method exclusion in [4] – [9], which we touch upon in that context below. From [10] the Board moved on to the exclusion for computer programs. In [11] the Board noted that the invention in that case could be implemented on new hardware specific to the invention, but also upon an entirely conventional computer programmed for that purpose. Paragraph [12] then reads as follows:
12. The Board is of the opinion that a claim directed to a technical process which process is carried out under the control of a program (be this implemented in hardware or in software), cannot be regarded as relating to a computer program as such within the meaning of Article 52(3) EPC, as it is the application of the program for determining the sequence of steps in the process for which in effect protection is sought. Consequently, such a claim is allowable under Article 52(2)(c) and (3) EPC.
52. Unpicking that slightly, the point the Board is making is that a claim to a process controlled by a computer is not to be considered a claim to computer program as such, but rather the computer is just the mechanism by which the invention (the ‘sequence of steps in the process’) is implemented or put into effect. This aligns with the soup canning process in CFPH supra. See also AT&T48 at [18].
53.Vicom [13] and [14] consider and reject an allegation of lack of novelty made on the basis that the claims cover a conventional computer operating according to new instructions. That then takes us to [15] and [16] dealing with the computer program exclusion (cited in Genentech, Merril Lynch, Aerotel and, as we will see, many other English cases), emphasis added:
15. In view of certain considerations by the Examining Division which appear to apply to the apparatus claims as well (cf. paragraph 10 above) it remains to be examined if the present apparatus Claim 8 would
48 AT&T Knowledge Ventures LP’s Patent Application [2009] EWHC 343 (Pat), [2009] FSR 19.
13.
Page 15 of 29
be objectionable under Article 52(2)(c) as qualified by (3) EPC. For reasons analogous to those given in paragraph 12 above, the Board holds that this is not the case and the same applies to the other apparatus Claims 9-11. Generally claims which can be considered as being directed to a computer set up to operate in accordance with a specified program (whether by means of hardware or software) for controlling or carrying out a technical process cannot be regarded as relating to a computer program as such and thus are not objectionable under Article 52(2)(c) and (3) EPC.
16. In arriving at this conclusion the Board has additionally considered that making a distinction between embodiments of the same invention carried out in hardware or in software is inappropriate as it can fairly be said that the choice between these two possibilities is not of an essential nature but is based on technical and economical considerations which bear no relationship to the inventive concept as such. Generally speaking, an invention which would be patentable in accordance with conventional patentability criteria should not be excluded from protection by the mere fact that for its implementation modern technical means in the form of a computer program are used. Decisive is what technical contribution the invention as defined in the claim when considered as a whole makes to the known art. Finally, it would seem illogical to grant protection for a technical process controlled by a suitably programmed computer but not for the computer itself when set up to execute the control.
54. So, it is no objection to patentability that the claimed invention is performed on a conventional computer operating in accordance with a computer program. What is ‘decisive’ is what technical contribution the invention makes to the known art. If the invention is at a higher level of generality than the computer/computer program used to implement it, then it will generally be outside the exclusion since the claim is to the invention implemented by the program, not to the program itself
49 (or ‘as such’). This recalls the soup canning machine example of CFPH. That is why the TBA states that, generally speaking, an invention which would be patentable in accordance with conventional patentability criteria should not be excluded from protection by the mere fact that for its implementation modern technical means in the form of a computer program are used. Likewise where the same invention can be implemented in hardware or in software, then it does not matter which is chosen as it can fairly be said that that is just an implementation choice and bears no relationship to the invention properly identified at all.
Symbian
55.Symbian concerned a DLL, or dynamic link library, which was a piece of software used to make programs run more efficiently on a computer. It was therefore what is sometimes called a ‘better computer’ case. The Court reviewed the cases from [37], citing extensively from Vicom, and in [49] and [51] followed the Board’s analysis in Vicom, two other TBA decisions relating to IBM50, and what
49 This is consistent also with Jacob LJ’s description of the technical contribution in Vicom itself being a new method and apparatus for manipulating images (see Aerotel at [81]). See also per Lewison J in AT&T Knowleght Ventures LP’s Patent Application [2009] EWHC 343 (Pat); [2009] FSR 19 at [20]. 50 IBM CORP/Data processor network T6/83 [1990] O.J. E.P.O. 5, [1990] E.P.O.R. 91 and T115/85 IBM CORP/Computer- related invention [1990] E.P.O.R. 107.
14.
Page 16 of 29
this Court said in Merrill Lynch and Gale. At [54] the Court concluded that the application was patentable, not only because it would produce a better computer, but also because it would benefit devices external to the computer, such as cameras, and furthermore recorded that “the fact that the improvement may be to software programmed into the computer rather than hardware forming part of the computer cannot make a difference – see Vicom; indeed the point was also made by Fox L.J. in Merrill Lynch”. See also the end of [58] of the decision.
AT&T and the Signposts
56. In AT&T, Lewison J applied inter alia Vicom, the IBM cases, Merril Lynch and Gale, summarising the law saying “what the Board [in Vicom] are saying … is … that you assess the patentability of the claimed invention ignoring the fact that it operates through a computer program. If, ignoring the computer program, it would be patentable, then the fact that a computer drives the invention does not deprive it of patentability”. Lewison J also sought to
51 formulate some ‘signposts’ to assist in the identification of a ‘technical effect’ at [40]. The Judge set
52 them out in the Judgment at [21]. As Birss J (as he then was) pointed out in Lenovo at [23] the signposts are really focussed on what might be called a ‘better computer’ case, and are not of universal application anyway53. Outside of ‘better computers’ cases, that leaves signpost (i), which just begs the statutory question. The signposts thus have little role to play in this case.
Halliburton
57. This case concerned a claim to a method of design of a drill bit, although the actual step of manufacturing the designed drill bit was not claimed54. Birss J (as he then was) provided a summary of the modern law in [32] – [33] and [38], which the Judge cited in the present case at [67]. Note in particular Birss J’s statements at [32] ” … The question is decided by considering what task it is that the program (or the programmed computer) actually performs. A computer programmed to perform a task which makes a contribution to the art which is technical in nature, is a patentable invention and may be claimed as such. …” and [33] “… What if the task performed by the program represents something specific and external to the computer and does not fall within one of the excluded areas? Although it is clear that that is not the end of the enquiry, in my judgment that circumstance is likely to indicate that the invention is patentable …”. This is all at one with Vicom and the cases citing it.
51 Which the judge subsequently revised in HTC v Apple at [148] – [150]. 52 [2020] EWHC 1076 (Pat). 53 As indeed Lewison LJ himself noted in HTC v Apple at [149]; see also per Kitchin LJ at [51]. 54 See at [17] – [18].
15.
Page 17 of 29
Gemstar and Protecting Kids
58. These two cases are of interest as concrete examples of the demonstration of another way in which a technical contribution may be demonstrated, namely where there are integers of the claim which are external to any computer/computer program. In such a case, the claimed invention as a whole is not to a computer program ‘as such’.
59.In Gemstar the first two patents in suit were held to fall within the exclusions as being programs for a computer, but the third was not. That patent, referred to as the ‘transfer patent’ in the judgment, concerned a claim in which an electronic programming guide enabled a user to select a function which resulted in a TV program being saved to an external hard drive. The Judge held at [234] that that movement of data from the computer to the hard drive was a physical effect which took the claim outside of the exception (note that there was nothing clever in the transfer process itself).
60.Protecting Kids was a decision which concerned a system for monitoring electronic communications, so that users (e.g., children) were not exposed to inappropriate content or language. See at [4]. The system involved sending to parents an alarm message when content judged as potentially inappropriate was accessed. The ‘clever’ bit was in the manner of the monitoring of the electronic communications, which as a result was said to be faster and more reliable than the prior art (see [5]). As a result the alarm messages were sent out more rapidly and reliably as well (see [6]), but they were otherwise entirely conventional alarms (see [25], [31]). There was therefore an “improved alarm” (see [31], [32]), even though the improvement only came about as a result of the more efficient monitoring process performed by the program on the computer.
61.At [34] Floyd J held that the relevant claim was allowable, since the generation of that improved alarm message sent to a remote device gave it technical character. Note here that the Judge based his decision on two aspects of the alert in Protecting Kids:
62. First, he drew attention to the fact that the sending of an alert message over a communications network was a physical process, rather than an abstract one – an effect external to the computer, just like the file transfer in the third patent in Gemstar;
63. Second, he relied on the fact that the monitoring was an improved monitoring (even though that monitoring was a process occurring entirely within the program on the computer). The inventive step lay within the program performed on the computer, but the claim was allowable because the invention as a whole could be characterised as a better way of monitoring internet access.
G. C
ONCLUSION ON THE LAW
64. From these cases we submit that the right approach can by synthesised as follows:
16.
Page 18 of 29
65. First – to start by asking the Aerotel questions, which principally serve correctly to orientate the Court as to the statutory test, but which they do not really help in answering55.
66. Second – then to consider the statutory test. As Parker LJ noted in Gale at 328 – 329, that 503 itself has two strands: does the claimed invention engage any of the exceptions, and if so does it do so to the extent that it is a claim to excluded subject matter ‘as such’.
a.One therefore has to ask whether the invention involves a ‘program for a computer’ at all?
b. If it does, is the claimed invention a ‘program for a computer … as such’.
67. That final step is often phrased as a search for a ‘technical contribution’, and the caselaw contains three clear strands (albeit they overlap in some cases):
a. First, are those cases where the invention is properly characterised as a method or system for doing something, and is above (or agnostic of) their specific mode of implementation. That they could be implemented on a standard computer using a computer program does not matter; a fortiori where there are other alternative implementations available. This is exactly the point made in for example Vicom at [15] – [16], AT&T at [18] & [20], the improved monitoring of internet access in Protecting Kids, and of which the soup canning process of CFPH cited in Aerotel is a vivid illustration.
b. Second, are those cases where the invention includes elements external to any computer/computer program, such that the invention as a whole does not relate to a ‘computer program … as such’. Examples here include Halliburton at [38], the data transfer in Gemstar and the remote alert message in Protecting Kids. Note that in each of these cases, the ‘clever bit’ lay in the program operating on the computer, but the claims were allowable nonetheless because they resulted in an effect external to the computer.
c. Third, are the so-called ‘better computer’ cases, such as the two IBM decisions. These are, however, likely just a subclass of the first class – a case in which the invention is properly characterised as a better computer, albeit that the improvement may be implemented through a computer program.
68.We do not suggest that there is a closed class of technical effect, and each case has to be considered on its facts, but the above form established threads running through the caselaw. Further
55 There is something to be said for concluding that the time has come to hold the fourth step is not in fact needed, as it is wrapped up in the third, although this is not a necessary point for this case.
17.
Page 19 of 29
whilst phrased as a search for a ‘technical effect’, in the first (and third) classes the invention is not to a ‘program for a computer’ at all, but is an invention at a higher level of generality – even if it can be implemented using a computer program. In the second class, even if some elements of the claim might properly be classified as a program for a computer, others elements exist outside of that, as so therefore the claim as a whole is not to a program for a computer ‘as such’.
69.Finally we note for completeness that where a computer program leads to such a technical
56 effect, then the program itself can be claimed – see Astron Clinica at [51].
H. THEEPO
70. The EPO’s approach to the exclusions is different, and very simple. The EPO adopts the ‘any hardware’ approach, and as such software-focussed claims which recite any hardware (a computer, a storage medium, a server etc) are allowable under Art 52 (T258/03 Auction Method/Hitachi). That approach, however, should also be understood in conjunction with the EPO’s approach to obviousness pursuant to the so-called Comvik approach by which certain features relating to non- technical fields are deemed to be known to the skilled person (or worked into the problem, in the problem-solution approach) for the purposes of the obviousness analysis. This is set out in the UKIPO’s skeleton at §76, and is discussed in some more detail in G1/19 at [31] – [33]. A non- technical feature is only ignored if it does not contribute to the solution of a technical problem, and an invention directed to a problem in a non-technical field can regardless be claimed if it provides a technical solution to that problem – see T1769/10 IGT in particular at [4.2]. We note that the UKIPO does not suggest that the EPO approach should be adopted here (UKIPO skeleton §79), and nor do we on this appeal.
I.OEPCJ
THER URISDICTIONS
Germany
71. The Courts of Germany follow an approach which is similar to that of the English Courts, and do not apply the EPO’s ‘any hardware’ approach. The leading case is understood to be Dynamische Dokumentengenerierung (Dynamic Document Generation)57, which sets out in section 4(c) that what is necessary to avoid the computer program exclusion is that a technical problem is solved. Although there was hardware in the claims (which referred to ‘at least one microcontroller’ and a ‘host computer’ for example), the Court did not just hold that that was sufficient (as the ‘any hardware’ approach
56 Astron Clinica Ltd v Comptroller General [2008] EHWC 85 (Pat). 57 Xa ZB 20/08 (BPatG).
18.
Page 20 of 29
would dictate), but instead went on to consider whether or not the claimed solution as a whole was ‘technical’.
France
72.The French courts, too, do not apply the EPO’s ‘any hardware’ approach. In what we
58 understand to be the leading case on computer implemented inventions, Sesame, the Cour d’Appel de Paris considered the nature of the invention in the round, and did not apply the simple any ‘hardware approach’. Indeed, the Court rejected the application even though it contained hardware elements59.
The Netherlands
73. The general approach of the Dutch Courts is to follow the law as applied by the EPO – but noting that the Dutch courts are not obliged to do so and do not always do so. The Netherlands has no national patent examination procedure, but a registration system, and one therefore does not find decisions of the Dutch Patent Office on the exclusions. In the only decision we have found, Orope
60 Germany v Nokia, the Hague District Court applied the law essentially in accordance with the EPO’s approach (see at [4.14]).
J. TJ’D
HE UDGES ECISION
74. After a few introductory paragraphs, the Judgment starts with a discussion of the Application, noting in [11] & [12] that in use the analysis performed by the ANN is purely of the physical properties of the tracks to make improved recommendations61. In [13], the Judge records the additional benefits of the speed and simplicity of the ANN of the invention.
75.From [14] – [18], Mann J discusses ANNs, rightly noting that ANNs consist of structures of nodes and layers (with links and biases), and that in a computer emulation the computer runs a piece of software which enables the computer to emulate the hardware ANN as if it were a hardware ANN
58 Cour d’Appel de Paris, 26/06/2016, case 15/01962. 59 There have been two relatively recent decisions of the Cour de Cassation in the field of excluded subject matter, but they add little. In the first, Thales (Cour de Cassation, Jan 11 2023, case no 075/2019.) the appeal was allowed on the basis that the decision of the Cour d’Appel de Paris was insufficiently reasoned, and remitted. In the second, Bull59, the Cour de Cassation refused to hear an appeal from the Cour d’Appel de Paris allowing the application to proceed. The case concerned the exclusion relating to the performance of mental acts, and the Cour d’Appel’s decision involved an investigation of the technical nature of the invention as a whole (not the simple application of the ‘any hardware’ approach). 60 Orope Germany GMBH vs Nokia Solutions and Networks OYECLI:NL:RBDHA:2022:9193. 61 See also the Hearing Officer at [48] and [54].
19.
Page 21 of 29
– software and hardware ANNs are the same, and have the same effect, save for the manner of their implementation.
76. After a review of the Hearing Officer’s decision, the substance of the Judgment starts at [32], where the Judge addressed the related questions of where are the computer and the computer program which the UKIPO relies upon to found its objection. In [33], the Judge records the UKIPO’s concession that a hardware ANN does not involve a computer program at all.
77. From [36] the Judge addresses whether a hardware ANN was itself a computer. EPAI contended that a computer was a ‘device for executing computer programs’; which the Judge rejected but which is something the UKIPO has now conceded (see its skeleton §62)62.
78. At [43] the Judge turned to consider the question of whether there was a computer program at all in a hardware ANN (which is an important point, since it is indicative of the nature of an ANN itself). That there was no such program was conceded by the UKIPO before the Judge, but this now forms the subject of its ‘Ground 3’ of its appeal, which it is convenient to address now.
79.After the oral hearing in the High Court the Judge sent a Note to Counsel [SUPP/3] which stated that he struggled to understand the UKIPO’s case on where the program for a computer lay in this case (reflecting also EPAI’s oral submissions to the same effect), and he requested clarification of certain questions. By its response [SUPP/3] the UKIPO stated in clear terms at §5 that “Insofar as the invention is implemented in hardware form, there is no relevant computer program to which the exclusion applies. In the premises, had the invention been limited to a ‘hardware-only’ implementation, it is unlikely that the computer program exclusion would have been engaged.” There was no uncertainty about that.
80.On this appeal, the UKIPO seeks to withdraw that concession, and now says that the hardwired structure of a hardware ANN itself is a ‘computer program’ – i.e., that the ANN itself is the program. The reason the UKIPO seeks to do this is because it has realised that if a hardware ANN is not or does not contain a computer program, then that shows that ANNs themselves are not computer programs – and given the clear law that an invention should not be excluded merely because it is implemented using modern technical means in the form of a computer program (Vicom, AT&T and many others) then it must follow that the Judge’s decision is right.
63 81.The UKIPO is therefore forced to argue that the ‘box with electronics in it’ is itself a ‘program for a computer’, or that “it is the set of weights and biases which therefore form the ‘program’ of this unconventional computer.” The UKIPO says that this is so because the electronics form the nodes,
62 And which is the subject of EPAI’s Respondent’s Notice. 63 Judgment [14].
20.
Page 22 of 29
links, weights and biases which enable the physical ANN to perform its function64. But so what? The electronics in any electronic device give it its function, but that does not make electronic devices into computer programs, and there is no principled reason why an ANN should be treated differently. Those electronics are not a ‘set of instructions’, but rather are hardware which achieve a function. There is no ‘computer program’ in a hardware ANN. Indeed on the UKIPO’s case, there is no ‘program for a computer’ but rather the program is the computer. The whole logic of the UKIPO’s position is confused and wrong.
82. The UKIPO also seeks to make comparisons with computer programs stored in ROM, but that is a different kind of thing. That is just storing a series of instructions for a computer in a permanent rather than temporary (RAM) sort of memory. It does not change the nature of the instructions. See also Astron Clinica at [41] – [42].
83.The UKIPO also says that it is enough to constitute a computer program that the links, nodes, weights and biases “start with human input and configuration and use an intermediate computer system to generate
65 the final product”. But that does not make the trained ANN a computer program, and the statute does not exclude the product of even a computer assisted process as Halliburton shows (there the drill bit was not even claimed).
84. It is submitted, therefore, that the UKIPO’s original concession was correctly made, and was right. A hardware ANN is not, nor contains, a program for a computer. It is a box of electronics.
85. As a consequence, since the UKIPO agrees on this appeal that “computers are devices for executing a computer program” (UKIPO skeleton, §62), then EPAI must also be correct on the point in its Respondent’s notice, namely that a hardware ANN is not a computer at all.
86.Turning to a software emulated ANN (which the Judge addressed in particular from [48]), the UKIPO refers to two potential computer programs: that used in training and that which provides the emulation platform in use.
87. EPAI accepts that a computer program is used in aspects of the training stage. Programming at that stage is something which enables the training objectives and approach to backpropagation to be fed through to the then untrained ANN, so that the training process can be set in motion66. It is just the method by which the training is performed, and we cannot see that it matters to this debate at all. As the Judge noted in [59] it does not serve to transform the whole system into a program for
64 UKIPO skeleton §63(2). 65 UKIPO skeleton §66. 66 See T2/239 – 243.
21.
Page 23 of 29
a computer, and one needs to be more analytical than that. The argument is completely irrelevant to the system claim 1, since that claims a trained ANN, but not the training process itself67. As regards claim 4, the point still does not matter. The exclusion bites where the whole claim comprises excluded subject matter. Even were (arguendo, which we do not agree) the training step to be objectionable, the Court would still need to resolve whether the other steps comprise excluded subject matter, including therefore the previous focus of the debate: the trained ANN itself. Furthermore the Judge also accepted that the trained ANNs themselves were to be regarded as technical effects (or technical things) – something the UKIPO accepted68 and has not appealed. That is also correct; just as the design of the drill bit was a technical effect in Halliburton – see at [71]. This is why we have submitted above that although it is possible that a distinction could arise between the claims, we do not think it likely that it does (or insofar as there is a distinction, the claims are both allowable anyway).
88.As regards the underlying software platform in a computer emulated ANN, for the reasons we have set out, that is an irrelevance. If an ANN itself is not a computer program, then its implementation by emulation on software is just an implementation decision, which is not relevant having regards to the law which we have set out. What matters is that which is claimed at the level of generality at which it is claimed. Here the claims are to (amongst other integers) the ANNs, agnostic of the manner of their implementation.
89. The Judge was therefore correct to conclude that the claimed invention did not claim a program for a computer at all, and that the exclusion was not invoked at all. See [50], [58] and [60] – [61].
90.The Judge nonetheless went on to consider from [63] whether, were there to have been a relevant program for a computer, the claimed invention was nonetheless outside of the exclusion for not being a claim to a program for a computer ‘as such’. See §§66.b and 67 above.
91. The technical contribution of the claims of the Application was set out by Hearing Officer at [53]69:
53. … [T]he invention of the Application is an ANN-based system for providing improved file recommendations. The invention may be hardware or software implemented. The fundamental insight is in the training of the ANN which analyses the physical properties of the file by pairwise comparisons of training files. In these pairwise comparisons the distance in property space between the output (property) vectors of the ANN is converged to reflect the differences in semantic space between the semantic vectors of each pair of files. The result is that in the trained ANN, files clustered close together in property space will in fact have similar semantic characteristics, and those far apart in property space will have dissimilar semantic characteristics. Once trained the trained ANN can then
67 Compare also Halliburton where the designed thing, the drill bit, was not claimed at all. 68 See Judgment at [78]. 69 See also Judgment [24], cross referred to in [65]. Minor typo in the final sentence corrected.
22.
Page 24 of 29
be used to identify, swiftly and accurately, files from a database which correspond semantically to a target file, and to provide – again swiftly and accurately – file recommendations to a user device (over a communication network).
92. The Judge, having summarised the law at [64] – [72] including Aerotel, Halliburton, Vicom, Gemstar and Protecting Kids, held at [76] that the claimed inventions, were they to have involved a program for a computer at all, would have been allowable on this alternative basis too. In particular, the Judge noted that the claims of the Application contain specific integers directed to the sending of the recommendation message and file to a remote user device over a communications network, and observed that that was the result of a technical analysis by the ANN based on technical criteria (i.e., the physical attributes of the tracks involved). He accepted that that is an external transfer of data (as per Gemstar), and is on all fours with claim 33 in Protecting Kids. There, the provision of an improved alert message, sent over a communications network to a terminal, provided a relevant technical effect, even though the ‘clever bit’ lay within the process for monitoring internet access occurring with the computer itself. Further, as in Protecting Kids, the message sent is better in that it contains a better
70 recommendation, and the system is faster and simpler. The Judge accordingly held that the claims of the Application were not to a program for a computer as such.
93. Further, the Judge held at [78] that a trained ANN itself was a technical advance, something which the UKIPO appeared to accept before him and which they do not challenge in this appeal. The Judge accordingly held that that also was a technical effect sufficient to render the claims outside of the exclusion, and it is not clear what (if any) answer the UKIPO has to this point.
94.Finally, even were the ANNs themselves to be considered to be computer programs, the claims could also be allowed on the basis that, properly characterised, the claimed inventions are
71 systems/methods for providing an improved file recommendation and recommendation message. That is an invention at a level above the manner of implementation of the system/method (i.e. a hardware or software ANN), and in accordance with the established law (summarised at 67.a above) the invention is not for a computer program as such. The claims should have been allowed on this basis too.72
70 And also is new in the sense that a prior art approach would send a different message. 71 Or expanding this, improved systems/methods for providing an file recommendation and recommendation message based on an analysis of the physical properties of the respective tracks. 72 This is the subject of the Applicant’s respondent’s notice, since the Judge recited the argument in law but did not need to address it on the facts.
23.
Page 25 of 29
K. TUKIPO’
HE S APPEAL
Ground 1
95. By its first ground of appeal, the UKIPO says that the Judge wrongly identified the inventive
73 contribution, which – the UKIPO says – was merely in the idea of “the training program”. It is not particularly clear what the UKIPO really means by that, but in any event throughout the UKIPO’s skeleton it muddles up the concepts of the invention claimed in an application, and the inventive
74 concept or idea underlying the claim: see §4 above. We set out some instances in the footnote below, but this is particularly acutely set out in the UKIPO’s skeleton at §§32 – 34. It sets out in §32 that claim 4 covers inter alia, a training program, an ANN, a database, the relevant data, a target device and a network, but then in §33 states that much of that was conventional at the priority date, and therefore in §34 concludes that “all that is left is the training program. That is, in substance, the contribution of claim 4”.
75 It describes this as ‘deconstructing’ the claim, but, as we have said, that kind of reductionist argument is exactly the ‘contribution approach’ which was rejected in Merrill Lynch, Genentech and Aerotel. See §39 above. The UKIPO’s approach is also contrary to what is said in Vicom, that what must be considered are the claims as a whole76. The UKIPO’s approach also is contrary to the Hearing Officer’s summary of the technical contribution of the claims of the Application, as we set out in §91 above, and which the UKIPO does not appeal77.
96. The UKIPO similarly argues that the analysis must be one of ‘substance not form’, citing Aerotel at [43] and [44]78. We do not disagree with this, but the UKIPO errs by taking this proposition far too far. It uses it as support for its ‘deconstructive’ contribution analysis which strips away everything save the inventive idea underlying the claim. That that is wrong is shown by Aerotel itself and the rejection in it of the contribution approach promulgated in Merrill Lynch at first instance. Indeed, were the UKIPO to be correct, many of the previous cases would have been decided wrongly: Vicom, Gemstar, Protecting Kids, Halliburton, Genentech etc.
73 UKIPO skeleton §34. 74 §11 – EPAI does say that the features listed are part of the claimed invention, but merely accepts that they are not individually new or inventive. §31 – the substance of the claimed invention does not reside in the training of the ANN, although that is a core part of the inventive step (a ‘fundamental insight’). 75 UKIPO skeleton §32. 76 §53 above. 77 Indeed accepted in front of Mann J, and appears to do so in front of this Court – see the UKIPO’s skeleton at §12. 78 e.g. UKIPO skeleton [19], [20].
24.
Page 26 of 29
97. The UKIPO’s skeleton also misunderstands the statement in Vicom (and following cases) that what is decisive is the substance of the claimed invention, not whether it is implemented in hardware or by software. In §§25 – 26 the UKIPO takes this to support a position that even hardware ANNs would be within the exemption. But that is entirely the opposite of what Vicom says – that case is making the point that if your invention is properly characterised as a method or system for doing ‘X’, then it will be outside of the exclusion irrespective of how it is implemented. See §§52 – 54 above.
Ground 4
98. This ground only arises if the UKIPO were to be correct on ground 1. By this part of its argument, the UKIPO argues that the Judge was wrong to say that the sending of the recommendation message over a communications network to the user device was not such as to constitute a technical contribution taking the claim as a whole outside of the exclusion. The UKIPO therefore also seeks to distinguish the decisions in Protecting Kids and Gemstar.
99.The UKIPO’s case is wrong for at least four reasons:
100.First, the UKIPO repeats its incorrect approach of seeking to ignore features of the claim which are a functional part of the invention. Specifically the UKIPO refers to the fact that the network is conventional and the message standard in form (although its content is new) – see for example UKIPO’s skeleton at §§39, 41, 55. But that was also true in Protecting Kids, and UKIPO’s argument misunderstands or ignores the reason why there is a consideration of such technical features – as we set out in §68 above, it is the presence of those elements of the claim which means that the claimed invention as a whole is not to a ‘program for a computer … as such’.
101.Second, the UKIPO repeatedly asserts that the selection of the recommended file is ‘subjective’ (see for example the UKIPO’s skeleton §§8, 56), but that is to misunderstand the technology completely. Once trained, the ANN of the Application makes decisions based solely on the objective, physical characteristics of the tracks – features like tonality, rhythm, timbre, based for example on a Mel spectrogram and discussed in the Application at col 23 – 31, col 36 and Fig 4. Although it 27616 – 27 is not known exactly how any particular trained ANN analyses the features of the tracks to make decisions, since the workings of an ANN are generally not human understandable, it is known that this is based on the physical analysis of the tracks – after all that is the only input data into the ANN in use. This is what the Judge sets out in [11] & [12], and is why he states in [76] that although the criteria used by the ANN cannot be described in purely technical terms (since they are unknown to any human); the ANN has gone about its analysis in a technical way, identifying the recommended file by the application of technical criteria. As we have said before, the UKIPO is confusing the nature of the analysis (based on purely physical features) with the benefits of the invention. The invention is
25.
Page 27 of 29
beneficial because the recommendations are better, more aligned with the musicality of the tracks, as well as for other reasons such as being a simpler and faster system (Judgment [13]), but that is not how the ANN of the invention operates.
102.The UKIPO refers at §42 of its skeleton to T0306/10 Yahoo!, but misunderstands that case. That decision concerns obviousness, and is somewhat compressed (likely due to the fact that the applicant did not attend oral proceedings – see at V – VII in the Summary). That was a case much like the two patents which were rejected in Gemstar, in which there was no feature external to the computer – the recommendation was just data on a screen. Further, the analysis was of business decisions, not physical data as in the present case. In the passage the UKIPO refers to it appears that the applicant in T0306/10 sought to justify the existence of a technical effect by reference to saving time because the human user of the service may make fewer clicks on screen. It is the user which plays more or fewer songs, or issues more or fewer search queries – see also T1741/08 at 2.1.6, cited in the paragraph of T0306/10 the UKIPO refers to. The case says nothing about a truly external effect such as the external transfers of data as occur in the present case, as well as in Protecting Kids and Gemstar, and indeed the T1741/08 decision cited in T0306/10 is clear that an external transfer of data is a technical effect – see 2.2.1. As such, the pair of decisions support EPAI’s position, not the UKIPO’s.
103. The UKIPO also refers to T1567/05, but this is a case concerning the exclusion for the presentation of information. The case itself concerned, in effect, colour coding a known stress analysis for building designs. The point being considered in the cited paragraphs is that that exclusion applies to all presentations of information (see 3.5) and that there is no technical effect in choosing a particular presentation to suit a particular class of person. It has nothing to do with this case at all.
104.Third, the UKIPO seeks to say that the Judge was wrong to hold that the external recommendation message and file was an external technical effect, and to distinguish Gemstar and Protecting Kids79.
105. The point with Gemstar was that the first two patents merely presented the information differently on a screen. Mann J considered that that could not be considered a technical effect, since most computer programs have this as a feature so would restrict the scope of the exclusion
80 unnecessarily, although those features were also excluded as being the presentation of information (see [52] and [148]). That has nothing to do with subjectivity, as the UKIPO puts it, and indeed in
79 Note that the UKIPO is wrong to say that ‘the only part of the invention relies upon by the Applicant as being relevant to any question of technical contribution was “the sending of a message over a communications network”‘, if they mean anything other than that is the focus of the Applicant’s case for this part of the argument. 80 It is not clear that he is correct as to this, although it probably matters little: see the summary of Vicom in Aerotel at [81].
26.
Page 28 of 29
the third patent the information which was transferred was selected by the user of the EPG – and therefore a truly subjective selection (contrast the Application). As regards Protecting Kids, the UKIPO simply misreads the case in its skeleton at §51. The point which Floyd J was making in the section they cite was that whilst an alert in general may just be information presented to a user, the remote alert sent to the user terminal in that case “is … qualitatively different” because the sending of the alert is “undoubtedly a physical one rather than an abstract one”, and as such was akin to the third Gemstar patent, just discussed. Those cases both make the same point – namely that an external transfer of data will usually establish a relevant technical effect external to the computer. That is unsurprising – a physical transfer of data from one device to another is not ‘a computer program … as such’ (or at all), but is a real-world event. Both Gemstar and Protecting Kids support EPAI’s case.
106. Fourth, the UKIPO does not deal with the fact that it appeared to accept before the Judge (and has not appealed) that a trained ANN is a technical thing in itself – see the Judgment at [78]. That decision is unsurprising and right; and follows from the nature of ANNs themselves as machines
81 which perform a useful function. That plainly establishes the allowability of the claims of the Application, and as we say in §93, it is not clear what the UKIPO’s answer to this is.
107. Finally the UKIPO also refers to ‘an alternative approach’ in their §§52 – 57, by which “a technical solution which is not itself intrinsically objective or technical satisfies the technical contribution threshold if the contribution as a whole addresses a technical problem”. Whilst we completely accept the premise, this is the point we make in §§67.a and 94 above – the CFPH soup canning machine type of case. The UKIPO’s arguments here are wrong for the reasons set out there.
Ground 2 – the UKIPO’s concession
108.We have addressed the substance of this above, and here only deal with two tail end points. First the UKIPO also refers to the decision in T0702/20 Mitsubishi82. This decision concerned a method of training ANNs which was divorced from any real-world purpose. It was completely abstract and untethered and, for example, would have covered processes such as forecasting stock market evolution (i.e. a business method). It was not considered excluded as a computer program as such, but was arbitrary, non-technical and therefore held to be obvious83. It does not support a proposition that ANNs are generally unpatentable, however, and quite to the contrary the Board was careful to state at [20]: “[t]he Board stresses that there can be no reasonable doubt that neural networks can provide
81 So long as the function is not in itself excluded for being a business method or the like. 82 UKIPO skeleton §78. 83 See at [19.1].
27.
Page 29 of 29
technical tools useful for automating human tasks or solving technical problems” and therefore be patentable. This case principally turns on its unusual facts, but the Board’s statement of principle at [20] supports EPAI not the UKIPO.
84 109.Second the UKIPO says that there may be other computer programs engaged in other parts of the system (e.g., the database). But that is irrelevant. It is no objection that parts of a claimed system are implemented on a computer, the claim as a whole has to be within the exclusion.
Ground 3 – Mathematical methods
110. This objection arises under s1(2)(a) of the Act, and relates to inventions which are “a mathematical method … as such”. The UKIPO’s point here is very short, but again misunderstands the law. At the level of generality of the claims, the claims are to systems or methods of using systems including ANNs. Those are practical applications of any underlying mathematical theory. The law is clear on this – whilst one cannot claim mathematical concepts in themselves, one can claim applications of those mathematical concepts, and indeed many (perhaps most) inventions in electronics, telecommunications and engineering make use of underlying mathematical theories and can be explained in mathematical terms. See for example Gale at 327 (Nicholls LJ), Hallibuton at [72] and 34 – 42 AT&T at [17]. The position in the EPO is the same: a ‘pure’ claim to a mathematical idea or concept is not allowable, but its application in a technical process is. See the Caselaw at 2.2.2, citing Vicom (which addresses the issue at [4] – [9]).
L. C
ONCLUSION
111.The UKIPO states in §80 that the Judgment creates a divide between the approach of the English Courts and the decisions of the EPO, creating a carve out for ANN implemented inventions that cannot be justified under the EPO or domestic approach to excluded subject matter. That is simply not the case. The approach of the Judge was to apply the established law in this jurisdiction (which is not the same as that in the EPO) to a proper understanding of the facts and the claims as properly construed, and reached a result which is the same as would be reached in the EPO applying the different approach which is taken there. It is the approach of the UKIPO, which ignores the nature of ANNs, and seeks to carve out of the claim as a whole some extracted statement of the inventive step underlying it, and to compare that to the exclusion, which is wrong and which would further diverge the law here from that in the EPO. The Judge was right, and this appeal should be dismissed.
84 UKIPO skeleton §67.
28.
Skeleton Argument for the Respondent, UKIPO
Page 1 of 26
Comptroller’sAppealSkeleton
CA2024000036
‐‐
AppealNo.CH2022000144
‐‐
INTHECOURTOFAPPEAL ONAPPEALFROMTHEBUSINESSANDPROPERTYCOURTSOFENGLANDANDWALES INTELLECTUALPROPERTYLIST(ChD) PATENTSCOURT(SIRANTHONYMANN)
BETWEEN:
EMOTIONALPERCEPTIONAILIMITED Respondent
and ‐‐
COMPTROLLERGENERALOFPATENTS Appellant
___________________________________________________
SKELETONARGUMENTONBEHALFOFTHEAPPELLANT
___________________________________________________
BRIANNICHOLSONKC ANNAEDWARDSSTUART
‐ 11SouthSquare 8thMarch2024
InstructedbytheGovernmentLegalDepartment
onbehalfofComptrollerGeneralofPatents
Page 2 of 26
Comptroller’sAppealSkeleton
Introduction
1.TheRespondent,EmotionalPerceptionAILtd(theApplicant),istheproprietorof patentapplicationNo.GB1904713.3(theApplication).Byhisdecisiondated22June 2022(BLO/542/22)(theDecision),theHearingOfficer(theHO)heldthatthe inventionclaimedintheApplicationfellsolelywithinmatterexcludedundersection 1(2)(c)ofthePatentsAct1977(theAct)asaprogramforacomputer(theExclusion). TheApplicantappealedthatDecision.TheHOrejectedthemathematicalmethod exclusion(s.1(2)(a)oftheAct),inrespectofwhichtherewasnoappeal.
2.TheApplicant’sappealwasheardbySirAnthonyMann.Byhisorderdated19 December2023(theOrder),andjudgmentdated21November2023(theJudgment; references:J/¶),heallowedtheApplicant’sappealandsetasidetherejectionof ApplicationNo.GB1904713.3(theApplication).Insummary,theJudgefoundthat theExclusionwasnotengagedbytheinventionclaimedbytheApplication.Hadthe Exclusionapplied,theJudgeneverthelessexpressedtheviewthattheinventionwas patentableasaresultofhavingarelevanttechnicaleffect.
3.TheAppellant,theComptrollerGeneralofPatents(theComptroller)appealsthe Order,withthepermissionoftheJudge.TheComptrolleradvancesfourgroundsof appeal:
(1)Ground1:theJudgeerredinholdingthattheexclusionfrompatentprotection for”a program for a computer… as such”wasnotengaged;
(2)Ground2:theJudgewaswrongtorelyontheAppellant’s’concession’;
(3)Ground3:theJudgewaswrongtoexcludetheconsiderationofthe mathematicalmethodexclusionmodel;and
(4)Ground4:theJudgewaswrongtoholdthattheclaimedinventioninvolvesa substantivetechnicalcontribution.
4.WeaddressGrounds1and4first,assuccessonthesegroundsissufficientto determinetheappealintheComptroller’sfavour.Grounds2and3providefurther supportforGround1,shouldtheCourtconsiderthattobenecessary.
1
Page 3 of 26
Comptroller’sAppealSkeleton
Technicalbackground
5.TheApplicationrelatestothefieldofartificialneuralnetworks(ANNs).ANNs representanimplementationofmachinelearningtechnologyinthefieldofartificial intelligence(AI).ThenameandstructureofANNsareinspiredby,andintendtomimic, theinnerworkingsofthehumanbrain,bycreatingaconnectionofnodes(or ‘neurons’),dividedintoseverallayers,throughwhichinformationisprocessedfroma giveninputtoadesiredoutput.
6.ThespecialcharacteristicofANNsrestsintheirrelianceonsetsoftrainingdata,which theANNusesinordertoadjustitsinternalparameters(theweightsandbiases)viaa predeterminedalgorithm,forinstanceabackpropagationalgorithm,inorderto generateanoutputthatisascloseaspossibletothedesiredoutput.Thisprocessof adjustmentisoftenanthropomorphicallyreferredtoas’learning’,but,inreality,itis asimplemathematicalprocesstofindthesetofparameters(i.e.,theweightsand biases)forthemathematicalmodelsoastocreatethedesiredrelationshipbetween theinputsandoutputsoftheANN.
7.TheJudgeprovidesashortintroductionto”The nature of an ANN” atJ/¶¶1418.This ‐ issufficientlycorrectasintroductorymaterialforthepurposesofthisappeal,butwe morespecificallyaddressthedifferencesbetweenhardware&software,and operational&trainingaspectsofanANNforthepurposeofGround2at[59]et seq., below.
TheApplication
8.TheApplicationconcernsthetrainingandsubsequentuseofanANNtoprovidedigital filerecommendationsofvariousmodalities(inparticular,torecommendationsof songtracks)basedonemotionalperceptionandothersubjectivecharacteristics.The inventionoftheApplicationwassummarisedbytheHearingOfficerat[8][10]ofthe ‐ DecisionandbytheJudgeatJ/¶4.
9.Therearetwoindependentclaims:claim1(aproductbyprocessclaim)andclaim4(a process(ormethod)claim).Itwascommongroundthatthereisnomaterialdifference betweentheclaimsforthepurposeofassessingtheextenttowhichtheExclusion
2
Page 4 of 26
Comptroller’sAppealSkeleton
applies,andboththeHearingOfficerandtheJudgeproceededonthebasisofclaim 4.AsdoestheComptrolleronthisappeal.
10.Claim4,brokenintoappropriateintegers,canbefoundintheAppendixtothe Judgment.TheHearingOfficerconstruedclaim4inthefollowingterms(at[42]ofthe Decision)which,astheJudgeobserved(atJ/¶23)wasnotdisputed:
It defines a method in which an ANN is trained on pairs of modally identical files(forexamplepairsofsongs)inordertomapthedistancebetweentheir propertyvectorsinpropertyspacetowardsthedistancebetweentheir semantic vectors in semantic space. When an input (for example representativeofauserselectedsong)ispresentedtotheANN,itgenerates afilevectorinpropertyspaceforsaidinput.Thisfilevectoristhen compared to the property vectors of reference files (representative of a libraryofsongs,say)toidentifythosefileshavingsimilarvectorsinproperty spacetothatoftheinput.Suchfileswillbesemanticallysimilartotheinput by virtue of the ANN having been trained to map/converge distances in propertyspacetowardsdistancesinsemanticspace(i.e.sothatitgenerates similarpropertyvectorsforsemanticallysimilarfiles).Afilecanthenbesent to, and output by, a user device. In this way, it provides a tool for recommendingsemanticallysimilarfiles.
11.Therecanbenodisputethattheinventionofclaim4isacomputerimplemented ‐ invention.TheApplicantdidnotdisputethatthecontributiondidnotresideinthe NLP(naturallanguageprocessing-usedtofindthevectorsinsemanticspace)orthe extractionofthemeasurablefeaturesfromthefilesperseusedtoformtheinputsto theANN(whichgeneratesthevectorsinpropertyspace).AstheApplicantsubmitted “thefundamentalinsightisinthetrainingoftheANNwhichanalysesthephysical propertiesofthefilebypairwisecomparisonsoftrainingfiles”.
12.TheinventiveconceptwascharacterisedbytheApplicantinthefollowingterms(as recordedbytheJudgeat[24]):
…theinventionoftheApplicationisanANNbasedsystemforproviding ‐ improvedfilerecommendations.Theinventionmaybehardwareorsoftware implemented.ThefundamentalinsightisinthetrainingoftheANNwhich analysesthephysicalpropertiesofthefilebypairwisecomparisonsoftraining files.Inthesepairwisecomparisonsthedistanceinpropertyspacebetween theoutput(property)vectorsoftheANNisconvergedtoreflectthe differencesinsemanticspacebetweenthesemanticvectorsofeachpairof files.TheresultisthatinthetrainedANN,filesclusteredclosetogetherin propertyspacewillinfacthavesimilarsemanticcharacteristics,andthosefar
3
Page 5 of 26
Comptroller’sAppealSkeleton
apartinpropertyspacewillhavedissimilarsemanticcharacteristics.Once trainedthetrainedANNcanthenbeusedtoidentify,swiftlyandaccurately, filesfromadatabasewhichcorrespondsemanticallytoatargetfile,andto provideagainst[sic]swiftlyandaccuratelyfilerecommendationstoauser ‐ ‐ device(overacommunicationnetwork).
Thelegalframework
13.Apatentmaybegrantedonlyforaninventionwhichsatisfiestheconditionssetout insection1(1)oftheAct.However,variousexclusionstopatentabilityareidentified at section 1(2) which provides as follows:
It is hereby declared that the following (among other things) are not inventions forthepurposesofthisAct,thatistosay,anythingwhichconsistsof-
(a)adiscovery,scientifictheoryormathematicalmethod;
(b) a literary, dramatic, musical or artistic work or any other aesthetic creation whatsoever;
(c)ascheme,ruleormethodforperformingamentalact,playingagameor doing business, or a program for a computer;
(d)thepresentationofinformation;
buttheforegoingprovisionshallpreventanythingfrombeingtreatedasan invention for the purposes of this Act only to the extent that a patent or applicationforapatentrelatestothatthingassuch.
14.Sections1(1)and(2)oftheActimplementArticle52oftheEuropeanPatent Convention(EPC).
Thelaw:theExclusionandtheneedforatechnicalcontribution
15.In order to determine whether a claim falls within the scope of the Exclusion, it must be assessed on the basis of ‘substance not form’. This principle was first articulated by the European Patent Office (the EPO)’s Technical Board of Appeal (the TBA) in Vicom T0208/84at[16]:
InarrivingatthisconclusiontheBoardhasadditionallyconsideredthatmaking adistinctionbetweenembodimentsofthesameinventioncarriedoutin hardwareorinsoftwareisinappropriateasitcanfairlybesaidthatthechoice betweenthesetwopossibilitiesisnotofanessentialnaturebutisbasedon technicalandeconomicalconsiderationswhichbearnorelationshiptothe inventiveconceptassuch. …
4
Page 6 of 26
Comptroller’sAppealSkeleton
Decisiveiswhattechnicalcontributiontheinventionasdefinedintheclaim whenconsideredasawholemakestotheknownart. …
16.Theprinciplewasfollowed(andtherebyincorporatedintoEnglishlaw)bythisCourt inMerrillLynch[1989]RPC1atp569):
…itseemstometobeclearforthereasonsindicatedbyDillonLJthatitcannot bepermissibletopatentanitemexcludedbysection1(2)undertheguiseof anarticlewhichcontainsthatitem-thatistosay,inthecaseofacomputer program, the patenting of a conventional computer containing that program. Somethingfurtherisnecessary.Thenatureofthatadditionis,Ithink,tobe foundintheVicomcasewhereitisstated:”Decisiveiswhattechnical contribution the invention as defined in the claim when considered as a whole makestotheknownart”.Theremust,Ithink,besometechnicaladvanceon thepriorartintheformofanewresult(e.g.,asubstantialincreasein processing speed as in Vicom).
17.ThesameprinciplewasappliedbythisCourtinGale’sApplication[1991]RPC305 (citingDillonLJ’sjudgmentinGenentech[1989]RPC147(alsoreferredtoinMerrill Lynch)andthedecisionsoftheTBAinIBMCorpT22/85andIBMCorpT65/86).At325 ‐ 327,NichollsLJheldthat:
…IntheGenentechcase,atpage240,DillonL.J.said: “It would be nonsense for the Act to forbid the patenting of a computer program,andyetpermitthepatentingofafloppydisccontaininga computerprogram,oranordinarycomputerwhenprogrammedwith the program; it can well be said, as it seems to me, that a patent for a computerwhenprogrammedorforthedisccontainingtheprogramis nomorethanapatentfortheprogramassuch.” It would equally be nonsense, if a floppy disc containing a computer program isnotpatentable,thataROMcharacterisedonlybytheinstructionsinthat programshouldbepatentable.Or,asFoxL.J.saidinMerrillLynch’s Application,[1989]R.P.C.561,569: “…itcannotbepermissibletopatentanitemexcludedbysectionI(2) undertheguiseofanarticlewhichcontainsthatitemthatistosay, ‐ inthecaseofacomputerprogram,thepatentingofaconventional computercontainingthatprogram.Somethingfurtherisnecessary.” ThisaccordswiththeapproachadoptedbytheEuropeanPatentOffice.In IBMCorp./Documentabstractingandretrieving(DecisionT22/85),[1990] E.P.O.R.98,theTechnicalBoardofAppealrejectedanapplicationfora patentrelatedtoamethodforautomaticallyabstractingandstoring documentsinaninformationstorageandretrievalsystemandtoa correspondingmethodforretrievingadocumentfromthesystem.Theboard
5
Page 7 of 26
Comptroller’sAppealSkeleton
heldthatanynewconceptintheapplicationcouldlieonlyintheprescribed procedures,andthoseproceduresorruleshadnotechnicalcharacterbut wereofapurelyintellectualnature.Inparagraph14,theboardadded: “Theforegoingconsiderationshavebeenmademainlyonthebasis thattheclaimedsystemsandmethodswouldinvolveaconventional computercontrolledbyasoftwareprogram…Analogous considerations,however,applyinthecasewherethecontrolofthe computerwouldbeeffectedbyhardware(specificallydesigned logicalmeans),anoptionalsofallingwithinthescopeoftheclaims,as the choice between the two possibilities is not of an essential nature butisbasedontechnicalandeconomicalconsiderationswhichbear norelationshiptotheinventiveconceptassuch.” Likewise, and even more recently, in IBM Corp/Text processing (Decision T65/86),[1990]E.P.O.R.181,188,theTechnicalBoardofAppealsaid:25 “Sincetheonlyconceivableuseforacomputerprogramisthe running of it on a computer, the exclusion from patentability of programsforcomputerswouldbeeffectivelyunderminedifitcould becircumventedbyincludingintheclaimareferencetoconventional hardware features, such as a processor, memory, keyboard and display,which,inpractice,areindispensableiftheprogramistobe usedatall.” … Iapproachthesubstantialissueinthiscase,therefore,onthefootingthatitis convenientandrighttostripaway,asaconfusingirrelevance,thefactthatthe claim is for “hardware”. The claim in the specification is, in substance, a claim toaseriesofinstructionswhichincorporateMr.Gale’simprovedmethodof calculatingsquareroots.Itisaclaimtoelectroniccircuitryintheformofa ROM which is only distinguishable from other electronic circuitry in the form ofaROMbythesequenceofinstructionsitcontains.Assuchthoseinstructions arenotpatentable,becausetheyconstituteacomputerprogram.
18.ThatjudgmentwasendorsedbyParkerLJandtheViceChancellorwhoheld(at333):
Turningnowtotheissuesinthiscase,inmyjudgmentthefollowing propositionsareestablished: (1)Anythingexcludedfrombeinganinventionbysection1(2)ofthe Act(“excludedmatters”)cannotqualifyasbeingpatentablewithin section1(1)bybeingincorporatedinaproductorprocessunlesssuch productorprocessiseitheritselfnovelorproducesatechnicalresult whichisnovel; (2)Theincorporationofadiscoverywhichisanexcludedmatterina productorprocessdoesnotrenderitpatentableiftheonlyeffectof suchincorporationistoproducesomethingwhichisitselfanexcluded
6
Page 8 of 26
Comptroller’sAppealSkeleton
mattere.g.thetradingsysteminMerrillLynch’sApplication,[1989] R.P.C.561. (3)Acomputerprogramisanexcludedmatterwhethertheprogramis onadisc(software)orinthehardware(aROM):GenentechInc.’s Patent,[1989]R.P.C.147perDillonL.J.at240;IBMCorp./Document abstractingandretrieving(DecisionT22/85),[1990]E.P.O.R.98;IBM Corp./Textprocessing(DecisionT65/86),[1990]E.P.O.R.181. MrGale’sdiscoveryisacomputerprogram(anexcludedmatter)incorporated inaROMwhichisadeviceofnoinherentnovelty.Themereincorporationof the program in the ROM does not alter its nature: it remains a computer program.Acomputerprogramremainsacomputerprogramwhether containedinsoftwareorhardware:proposition(3)above.Moreovertheresult of the incorporation of Mr Gale’s “method of calculation” or “computer program”(bothexcludedmatters)onlyproducesanotherexcludedmatter,viz. acomputerprogram:proposition(2)above.Thatisenoughtodecidethiscase.
19.The’substancenotform’testispartoftheassessmentofstep2ofAerotelvTelco Holdings[2006]EWCACiv1371.Havingdeterminedthataclaimistoacomputer programasamatterofsubstancenotform,whethertheExclusionappliesdepends onwhetherornotthereisatechnicalcontribution.Thisisclearfrom[43][44]: ‐
43….Howdoyouassessthecontribution?MrBirsssubmitsthetestis workableitisanexerciseinjudgmentprobablyinvolvingtheproblemsaidto ‐ be solved, how the invention works, what its advantages are. What has the inventorreallyaddedtohumanknowledgeperhapsbestsumsuptheexercise. Theformulationinvolveslookingatsubstancenotformwhichissurelywhat ‐ the legislator intended.
44….Intheendthetestmustbewhatcontributionhasactuallybeenmade, notwhattheinventorsayshehasmade.
20.In HTC v Apple [2013] EWCA Civ 451, Kitchin LJ reiterated the ‘substance not form’ approach(at[37])andexplainedthatfollowingtheAerotelapproachaclaimed inventionwhoseonlycontributionisnottechnicalorliesinexcludedsubjectmatter willberejectedundertheExclusion(at[41]).Hewentontoconsiderhowto determinewhetheraninventionhasmadeatechnicalcontributionat[45][51]and ‐ explainedthatthefive’signposts’inAT&TKnowledgeVentures[2009]EWHC343 (Pat)provideusefulguidance,althoughsucharenotdeterminativeand,indeed,not allsignpostswillberelevanttoeverycase.
7
Page 9 of 26
Comptroller’sAppealSkeleton
21.Finallywenotethatthetechnicaleffectinquestioncanbeeitherinternalorexternal (asperSymbianvComptrollerGeneral[2008]EWCACiv1066at[58]),butitneedsto bearelevanttechnicaleffect.Thisqualificationneedstobemadeinordertoavoid thepitfalloftooeasilyascribingatechnicaleffecttoinventionswhichinvolve programsforcomputersforthesimplereasonthattheyseem”technical”intheusual sense:seeProtectingKidstheWorldOver[2011]EWHC2720(Pat)at[14].
22.The position on the authorities is clear, irrespective of the form in which the invention is claimed (or can be implemented) if in substance the claim is to a computer‐ implemented invention (i.e. an invention performed by a computer program) then the Exclusion is engaged. The allowability of that claim must then be decided by determining whether or not it reveals a technical contribution. There is nothing wrong with computer‐implemented inventions (whether or not they incorporate an ANN) that make a relevant technical contribution; the exclusion catches those computer‐ implemented inventions (with or without an ANN) which do not make a relevant technical contribution and which are accordingly a “computer program…as such”.
Ground1:theJudgeerredinholdingthattheexclusionfrompatentprotectionfor”a
programforacomputer…assuch”wasnotengaged
23.BythisGround,theAppellantchallengestheCourt’sconclusions(at[61])thatclaim4 (and,implicitly,claim1)isnotaclaimtoacomputerprogramatall,thattheExclusion isthereforenotengaged,and(at[62])thatasamatteroflaw,itwasthereforenot necessarytoconsiderthequestionoftechnicalcontribution.
24.AstheJudgecorrectlyrecognised:
(1)theclaimsdefineacomputerimplementedinventionandtheANNincludedin ‐ theclaimsmaybeembodiedinsoftwareorhardware(J/[22]);and
(2)theANNdefinedintheclaimsmaytaketheformofadedicatedhardwareunit (inthecaseofahardwareembodiment)oritmaybeimplemented (“emulated”)byrunningappropriatesoftwareonthegeneralpurpose ‐ computerwhichisthesubjectoftheclaims(inthecaseofasoftware embodiment)(J/[23]).
8
Page 10 of 26
Comptroller’sAppealSkeleton
25.Acomputerimplementedinvention,whenclaimedassuch,seeksprotectionfor ‐ embodimentsimplementedinhardware,insoftware,orinacombinationofhardware orsoftware.Theprincipleof’substancenotform’meansitisinappropriatetomakea distinctionbetweenembodimentsofthesame(claimed)inventioncarriedoutin hardwareorsoftware,aswhatisdecisiveisthesubstanceoftheclaimedinvention (i.e.whetheritrevealsatechnicalcontribution)notitsform(i.e.whetheritis hardware‐ or software‐implemented).
26.The Judge appeared to recognise (at least implicitly) that a claim to a computer‐ implemented invention (protecting hardware and software embodiments) should be assessed as a matter of substance not form when applying the exclusion to a program for a computer (J/[56] and [78]). However the Judge did not apply this principle.
27.The actual approach taken by the Judge is difficult to discern from the Judgment. Rather than seek to identify and assess the substance of the invention, the Judge adopted an alternative approach assessing the hardware and software implementations separately in each case assessing whether the Exclusion was engaged by reference to the identification of a ‘computer’ and ‘computer program’ to which the Exclusion could apply.
28.We attempt to summarise the Judge’s approach below:
(1)Rather than determine the appeal on the basis of the Grounds of Appeal advanced by the Applicant, the Judge decided instead to ascertain whether or nottheExclusionwasengagedbyaskinghimselfwhatwasthe’computer’and whatwasthe’program’saidtoengagetheExclusion(J/[32]).Asthisquestion onlyaroseafterthehearing,theJudgerequestedsequentialwritten submissionsfromtheComptrollerandtheApplicantrespectively.
(2)Havingfoundthatacomputerisinvolvedinboththehardwareandsoftware ‐ implementations,theJudgeproceededtodisregardthehardware implementationonthebasisof(amischaracterisationandmisunderstanding of)theComptroller’s’concession'(addressedinmoredetailinrelationto Ground2)(J/[43][47]). ‐
9
Page 11 of 26
Comptroller’sAppealSkeleton
(3)Insofarasthesoftwareimplementationisconcerned,theComptroller submittedthattherearetwoprogramsinvolved:(i)atthetrainingstageand (ii)inthecomputerwhichoperatesthetrainedANN.TheJudgeagreedwith(i) butnot(ii)(J/[48][49]&[57]).Rather,theCourtfoundthatthesoftware ‐ ‐ implementedANNcanbe”decoupled”fromthecomputersoftwareemulating theANN(J/[56][59]),thattheremainingdecoupledANNoperatesinthesame ‐ way as the hardware ANN and that, extrapolating from his interpretation of the Comptroller’s concession, the decoupled ANN does not represent a program for a computer.
(4)The Judge therefore concluded that the only program involved in the invention is the training program (J/[59]), which “involves setting the training objectives in terms of the structure of the ANN (the software) and the training objectives” (J/[60]).
(5)The Judge concluded that the training program is only a subsidiary part of claim 4 and therefore that the Exclusion is not engaged because “as a matter of construction the claim is not to a computer program at all” (J/[61]).
29.Having erroneously found that the Exclusion is not engaged at all solely on an assessment of the technical constituent elements of the different forms which the claimed invention can take, the Judge stated (at [63]) that it was not necessary, as a matter of law, to consider the question of whether the actual contribution of the inventioninclaim4istechnicalinnature.
30.Inthecircumstancesinwhichaclaimisdirectedtoacomputerimplemented ‐ invention(irrespectiveofwhetherthatincludesanANN),theinventionbeing implementedinhardwareorsoftware(oracombinationthereof),theJudgeshould havedisregardedthepossiblealternativeformimplementationsandshouldhaveheld thattheclaimsweredirectedto”aprogramforacomputer…assuch”,thatthe Exclusionwasengaged,andthatthequestionoftechnicalcontributionshouldbe asked.
31.Wehavesetouttheconstructionofclaim4andtheallegedinventivecontribution above.ThesubstanceoftheinventionresidesinthetrainingoftheANN.Thisis
10
Page 12 of 26
Comptroller’sAppealSkeleton
recognisedbytheApplicant,whosubmittedthatthe”fundamental insight is in the training of the ANN”i.e.theuseofatrainingprogramprovidingatrainingmethod (pairwisecomparison)andobjective(convergingdistances).
32.Thatthetrainingprogramisthesubstanceofthecontributioncanalsobedetermined bylookingattheclaimedinventionitself.Claim4coversamethodofproviding semanticallyrelevantfilerecommendationsinasystem.Thesystemissimpleto deconstruct into its constituent elements. It consists, in the training phase, of (1) a training program, with accompanying training data on which the instructions are run, and (2) an initially untrained ANN. Once the ANN is trained, the system consists of (1) the trained ANN, (2) a database of files on which the trained ANN operates, (3) target data given by the user, (4) a communications network, and (5) a user device.
33.Simply put, everything else in this system apart from the training program is already well‐known in the field. The application is silent as to any specific details of its implementation. As the Hearing Officer noted at [50]‐[51], this is not a case where the contribution resides in the hardware per se. Nor does the contribution reside in the NLP (natural language processing) or the extraction of the measurable features from the files per se, for both of which off‐the‐shelf software is taught to be used.
34.Accordingly, once the claimed method is stripped (as it must be) of its form, those characteristics which are common to any (or at least most) ANNs, and the network, media file database and user device (all of which are known in the art), all that is left isthetrainingprogram.Thatis,insubstance,thecontributionofclaim4.
35.Havingconstruedclaim4asaclaimtoacomputerimplementedinventionthat ‐ includesanANN,andhavingidentifiedtheactualcontributionoftheinventionin claim4asbeingthetrainingoftheANNasper[34]above,theJudgeshouldhave concludedthatclaim4isaclaimtoaprogramforacomputerassuchandtheExclusion isthereforeengaged.TheCourtshouldthenhaveproceededtoassesswhetherthe actualcontributionsoidentifiedistechnicalinnature,pursuanttoAerotelstep(4).
36.Further,insofarastheJudgefound(contrarytotheHOandtheApplicant’sAppellant’s Notice)thattheANNcouldbe’decoupled’fromthecomputersoftwareemulatingthe
11
Page 13 of 26
Comptroller’sAppealSkeleton
ANN(ashedidatJ/[58]),theJudgeshouldhaveheldthatthe’decoupled’ANNwas excludedasamathematicalmethod.ThisisaddressedinGround3.
37.Yetfurther,theJudgewaswrongtorelyinhisassessmentoftheExclusion(ashedid atJ/¶56)ontheComptroller’s’concession’asregardshardwareimplementedANNs asabasisforholdingthatthesoftwareimplementedANNisnotoperatingaprogram. ThisisdealtwithinGround2.
Ground4:theJudgewaswrongtoholdthattheclaimedinventioninvolvesasubstantive
technicalcontribution
38.TheonlypartoftheinventionrelieduponbytheApplicantasbeingrelevanttoany questionoftechnicalcontributionwas”thesendingofamessageovera communicationsnetwork”(Applicant’sSkeletonbelowat[122]).TheJudgeheld(at J/¶76)thatthefiletransmissionwascapableofconstitutinga’technical’contribution becausethefilehadbeenidentifiedandmoved”becauseitfulfilledcertaincriteria”.
39.Theactualtransferofthemediafilesselectedfromadatabasetotheenduserisdone ‐ inanentirelyconventionalway.Thereisnoteachinginthespecificationastohowit istobedone,soitmustbeCGK.Soitisthecriteriathatneedtobeconsidered.
40.Havingrecognisedthatthecriteriabywhichmediafilesareselectedfor recommendationbytheANNare”nottechnicalcriteriainthesensethattheycanbe describedinpurelytechnicalterms”(J/¶76),theJudgeshouldhaveconcludedthatthe criteriaprovideananswertoasubjective(i.e.nottechnical)problemonbehalfofthe enduser,andthatthecontribution(ifany)oftheApplicationisthereforeincapable ‐ ofgivingrisetoatechnicaleffect.
41.Thatthefulfilmentofsubjectivecriteriacannotconstituteatechnicalcontributionis consistentwithMerrillLynch,inwhichthejustificationforatechnicalcontribution wastheneedfor”sometechnicaladvanceonthepriorart”(seethepassagecited above).Oncetheactualcontributionoftheclaim,properlyconstrued,isstrippedof anyexcludedsubjectmatter,theremustbe’somethingelse’leftwhichjustifies ‐ grantingtheapplicanttherequisitepatentmonopoly.Thismustbecorrect,orelse anyclaimwhichis,forinstance,aclaimtoacomputerprogramassuchcouldescape
12
Page 14 of 26
Comptroller’sAppealSkeleton
theExclusionsimplyby,forexample,outputtingafile.Thelawisseekingtoreward inventions,bywhichitmeansthat’somethingelse’whichisatechnicaladvanceover thepriorart.
42.TheEPOdoesnotconsidersubjectivecriteriatofulfilatechnicalpurpose.InYahoo!T 0306/10(inconsideringinventivestep)theEPOheld(at[5.2]):
In the Board’s view, the selection of an item, for example a song, for recommendationtoauserdoesnotqualifyasatechnicalpurpose.Froma technicalpointofviewitisirrelevantwhatsongsarerecommendedtoauser. While making “good” or “bad” recommendations may lead to different user reactionsandthereby,intheend,todifferenttechnicalresults(theusermight forexampleplaymoreorfewersongs,orissuemoreorfewersearchqueries in order to find other songs), such results do not qualify as a technical effect of therecommendations,astheydependonsubjectivechoicesmadebytheuser (cf.decisionT1741/08of2August2012,reasons2.1.6).
43.FurtherinDesigningofBuildingStructures/ENUT1567/05inthecontextofthe presentationofinformationexclusionofArt52(2)(d),itwasheld(at[3.5][3.6]): ‐
3.5Nordoesthechoiceofinformationtobepresentedhaveatechnicaleffect. Article52(2)(d)EPCdoesnotdistinguishbetweendifferentkindsof information. Therefore the presentation (as such) of any information must be regardedasa”noninvention”.Itisthusirrelevantthatthepresentinvention ‐ isarrangedtodisplaythelargeststressvalueforeachbuildingstructural component (rather than some other stress value). Although relating to technicalphenomena,thestressvaluesaremerepiecesofinformationaimed exclusivelyatthehumanmind.Itfollowsthatalsothefeaturesdetermining the kind of information displayed do not contribute to an inventive step. 3.6Theaboveconclusionissupportedbytheobservationmadeinthe description(paragraph[0004])totheeffectthatstressvaluesdisplayedinthe formofnumericaldataaredifficulttounderstand.Thiswillhoweverdepend onthepersonstudyingit.Anengineerwouldreadilyunderstandnumerical valuesandmightpreferthemtocolourssincetheyaremoreexactandallow anindepthanalysis,whereasalaymansatisfiedwithaquickoverviewmight ‐ prefercolours,especiallyifthesehavebeenchoseninaconventionalway(eg redforcriticalcomponents;cfparagraph1above).Theproblemindicatedin thedescriptionisthusnotofatechnicalcharacterbecauseitdependsonthe skillsandpreferencesoftheparticulargroupofusersforwhomtheinvention isintended. 44.Thereisnoreasonofpolicyorprinciplewhythesubjectivepreferencesofuserswould beoftechnicalcharacterinthecaseoftheExclusion,wheretheydonotsoqualifyfor
13
Page 15 of 26
Comptroller’sAppealSkeleton
thepresentationofinformationexclusion.Onthecontrary,considerationsof consistencywouldrequiretheCourttoadoptthesameapproach.
45.At[76]theJudgedrewsupportforhisfindingthatthetransferofafile”because it fulfilled certain criteria”couldamounttoatechnicalcontributionfromthedecisions inGemstarvVirginMedia[2009]EWHC3068(Ch)andProtectingKids.However,both thesecasesprovidefurthersupport(albeitimplicitly)forthepropositionthat subjective criteria cannot constitute a technical function.
46.Gemstar concerned three patents relating to electronic programming guides (EPGs). The first patent concerned a method for navigating around TV programmes. It claimed a specific grid format in which the programmes were displayed on the TV screen in the EPG (see [38]‐[39]). The second patent similarly provided for an improved television schedule system and process based on favourite channels (see [141]‐[142]). The third patent, however, was more complex. It assumed the existence of a set‐top box capable of receiving and recording TV programmes and of displaying an EPG, the latter of which was used to control and programme the recording of TV shows in the known way. The alleged invention rested in the additional facility in the EPG which provided a means of controlling the transfer of an already recorded programme to a secondary recorder (e.g. a VCR).
47.In relation to the first two patents, Mann. J found that they covered inventions which were programs for a computer as such and were therefore excluded from patentability.Inrespectofthefirstpatent,MannJheld(at[50])that:
ThetechnicaleffectreliedonbyGemstarisabetterinterface,oradifferent interfaceif”better”isnotrelevant.Thatisanabstractconcept.Itdoesnotin termsdescribesomephysicalactivityoreffect.Thereisadifferentdisplayon thescreen,butthatisnotenough,inmyview.Thatisstillpartofthecomputer programandisnotanexternaleffect(MrBirssdidnotrelyonanyinternal effect).Manycomputersrunningaprogramarelikelytohaveadisplayoutput, andifthatwereenoughtobeatechnicaleffecttheneveryprograminsucha computerwouldbelikelytofalloutsidetheexclusion,whichisunlikelytohave beentheintentionofthedraftsmanoftheact.Adifferentdisplaytothat shownbeforedoesnotseemtometogofarenoughtoamounttoatechnical effectwhichmakesadifference.MrBirssdescribesthetechnicalcontentas beingabetteruserinterface(usually)orauserinterface(sometimes).That wayofdescribingitdoesnotovercomethedifficultyhefaces.Ultimatelythey
14
Page 16 of 26
Comptroller’sAppealSkeleton
arebothwaysofdescribing,indifferenttermsfromthepatent,whatthe inventionissaidtoachieve.Buttheyarebothjudgmental,thefirstmoreso thanthesecond.Thefactthatwhattheuserperceivesandinteractswithis “better”doesnotmaketheadvancetechnicalatall(norisitpartoftheclaims). Nordoescharacterisingitasaninterfacegiveitatechnicaleffectthatitwould nototherwisehavehad.Onehastolooktoseewhattheeffectactuallyis,and inmyviewitisnottechnical.Infact,inthesenseinwhichMrBirssusesthe expression,”interface”confirmsthis-itisanabstract,notaphysical,concept.
48.Similarly,fortheSecondPatentheheld(at[147]):
Thisseemstomeplainlytobeacomputerprogram.Itisaprogramwhichtakes data, makes a display, permits visual and internal tagging, and then displays a subsetofthatdata.Sothereislittledoubtthatitisaprogram.Whatisthe contributionsaidtobe?ItwasdescribedbyMrBirssasbeingtheimproved use of the screen ‘real estate’ as per the user’s interactively applied specification.Soithasworkedonthedataandfilteredit.Isthatatechnical contribution?Inmyviewitisnot.Itdoesnotmakethecomputerabetter computer. It does nothing to make the computer do anything other than to sortsomedata,andthatisnotsufficient.
49.However,thethirdpatentwasfoundnottobeexcludedfrompatentability,asaresult ofhavinganactualtechnicaleffect.Whatdistinguishedthethirdpatentfromthe otherswasthephysicaleffectoftransferringdata1.MannJheld(at[234]):
Againthequestioniswhetherwhattheinventionachieveshasarelevant technical effect. This time I think that it does. This is not merely a computer runningaprogramwithoutanyeffectinwhatmightberegardedastheoutside world.Whileitdoesnotproducea”bettercomputer”itdoesactuallyachieve something which can be regarded as a physical effect, namely the initiation of movementofdatafromonedisktoanother(bothmetadataandTV programmecontent).Thatseemstometobeenoughtopreventitbeingjust acomputerprogramassuchandtorenderitpatentablematerial.Itistruethat itdoesnotproduceaneffectoutsidethesystemitself,butitisstillaneffect.
50.ProtectingKidsconcernedapatentapplicationrelatingtoasystemformonitoringthe contentofelectroniccommunicationstoensurethatusers,especiallychildren,are notexposedtoinappropriatecontentorlanguage.Theappealfocusedonclaim33, whichclaimedadatacommunicationsanalysisenginewhichsamplesdatapackets fromacommunicationschannelforfurtheranalysisofthecontent.Oncethecontent isanalysedandanalertlevelassigned,anadministratorsuchasaparentisnotified
1 On this point, see also Lantana v Comptroller [2014] EWCA Civ 1463 at [40]‐[41].
15
Page 17 of 26
Comptroller’sAppealSkeleton
andisrequiredtosendacommandinreturn(e.g.terminatingtheelectronic communicationsorshuttingdowntheequipment).
51.Infindingthattheinventionofclaim33doesnotlackarelevanttechnicaleffect,Floyd J(ashethenwas)engagedinthefollowinganalysis(distinguishingthefirsttwo Gemstarpatents)(at[34][35]),reiteratingthatthestartingpointforageneratedand ‐ ‐ transmittedfileisthatitisnotarelevanttechnicalcontribution:
34….Istartwiththepropositionthatthegenerationandtransmissionofan alertnotificationtotheuser/administratorisnotarelevanttechnicalprocess. Iacceptthatinmanycasesthismaybecorrect.Plainlyitwascorrectinthe case of two out of the three patents considered by Mann J. in Gemstar, where informationwassimplydisplayedonascreen.Butwhatisinplayinthepresent case,namelyanalarmalertingtheuser,ataremoteterminalsuchasamobile device, to the fact that inappropriate content is being processed within the computer,isinmyjudgmentqualitativelydifferent.Firstofall,theconcept, althoughrelatingtothecontentofelectroniccommunications,isundoubtedly a physical one rather than an abstract one. In that respect it was more akin to thethirdofthethreepatentsconsideredbyMannJ.inGemstar.Secondly,the contributionofclaim33doesnotsimplyproduceadifferentdisplay,ormerely rely on the output of the computer and its effect on the user. The effect here, viewedasawhole,isanimprovedmonitoringofthecontentofelectronic communications.Themonitoringissaidtobetechnicallysuperiortothat produced by the prior art. That seems to me to have the necessary characteristicsofatechnicalcontributionoutsidethecomputeritself. 35.AccordinglyIwouldholdthatthecontributionofclaim33doesnotreside wholly within the computer program as such exclusion. I think that conclusion isinaccordancewiththeAT&Tsignposts.InparticularIwouldsaythatthe inventionsolvesatechnicalproblemlyingoutsidethecomputer,namelyhow to improve on the inappropriate communication alarm generation provided by thepriorart.Thatissufficientinmyjudgmenttoovercometheobjectionon theveryspecificfactsofthecasebeforeme.
52.Wementionthatanalternativeapproachsupportedintheauthoritiesariseswherea technicalsolutionwhichisnotitselfintrinsicallyobjectiveortechnicalsatisfiesthe technicalcontributionthresholdifthecontribution,asawhole,addressesatechnical problem.AsstatedbyKitchinLJ(ashethenwas)inHTCvAppleat[49]:
…itisalsohelpfultoconsiderwhethertheinventionmayberegardedas solvingaproblemwhichisessentiallytechnical,andthatissowhetherthat problemliesinsideoroutsidethecomputer.[…]Aninventionwhichsolvesa technicalproblemoutsidethecomputerwillalsohavearelevanttechnical
16
Page 18 of 26
Comptroller’sAppealSkeleton
effect,forexamplebycontrollinganimprovedtechnicalprocess.Ineithercase itwillnotbeexcludedbyArt52asrelatingtoacomputerprogramassuch.
53.Wementionthisforcompletenessandalsobecauseitappearsatleastpossiblethat boththethirdpatentinGemstarandclaim33ofProtectingKidscouldequallybe justifiedonthe’solvingatechnicalproblem’approach.
54.WhicheverwaythethirdpatentofGemstarandclaim33ofProtectingKidsare assessed,theallegedtechnicaleffectofthepresentinventiondiffersfromthemboth.
55.First,theeffectofthepresentinventiondoesnotincludeanyphysicalconsequences (thetransferofpreexistingdatatotheuserisentirelyconventionaland,assuch,is ‐ notpartoftheinventivecontribution).Noneofthedata,themethodof communication,northecommunicationsnetworkitselfisthesubjectoftheinvention. Itonlyinvolvesdeliveringtotheuserdevicearecommendedmediafile,whichhas beenselectedfromadatabaseby,essentially,aprocessofdataanalysis.
56.Second,theeffectofproducingafilewithsubjectiveattributeswhichmayhavean effectontheuser(inthattheuserenjoystherecommendation,or,moregenerally, certainemotionsaretriggeredinresponsetothatrecommendation)isnotsimilarto thewidereffectofimprovingthemonitoringofcontent.Theeffectoftheapplication inProtectingKidswasnotsimplythetransmissionofanalerttotheparent-itwas thefactthat,bysearchingtheelectroniccommunicationsforinappropriatecontent, informingtheparentofsuchcontentandthenactingaccordingtoacommandgiven bytheparentinresponse,anentiresystemiscreated,whichismeanttoensurean overallimprovedmonitoringprocess(therebysolvingatechnicalproblem).Thereis noequivalentinthepresentcase.Indeed,theJudgerightlyheldthattheuserdoes notevenneedtolistentothesong(J/¶76).Thereisnowidersystem,theonlyeffect isthedeliveryofarecommendedfilewithcertain(subjective)characteristics.
57.Indeed,itisnotablethattheJudgeneverexplainedwhathemeantby”becauseit fulfilledcertaincriteria”,beyondaccepting(J/¶76)thatthecriteria”arenottechnical criteriainthesensethattheycanbedescribedinpurelytechnicalterms,buttheyare criterianonetheless”.Nordidheseektosuggestthattherewasanytechnicalproblem
17
Page 19 of 26
Comptroller’sAppealSkeleton
whichtheinventionwassolving.Inshort,hefoundthatanontechnicalproblemwas ‐ beingsolvedbythetechnicalapparatusofanANNbasedcomputer. ‐
58.HadtheJudgenoterredashedidinrespectofGrounds1and4above,heshouldhave foundthattheApplicationisacomputerimplementedinventionthatdoesnot ‐ discloseasubstantivetechnicalcontributionandisthereforeexcludedfrom patentabilityasaprogramforacomputerundersection1(2)(c)oftheAct.
Ground2:theJudgewaswrongtorelyontheAppellant’s’concession’
59.Followingthehearing,furtherwrittensubmissionswererequestedbytheJudgeinhis noteof15September2023,andweresubmittedsequentiallyon22and27September bytheComptrollerandtheApplicantrespectively(seeJ/¶3233). ‐
60.Themainquestionswhichthefurthersubmissionsaddressed,assetoutbytheJudge, were(1)whatistherelevant’computer’forwhichthereisarelevantprogram?and (2)whatistheprogramwithinthatcomputertowhichtheexclusionapplies?The Comptrollerthereforehadtoengageinanartificialdistinctionbetweenthehardware andsoftwareimplementationsinordertoaddressthesequestions.Itspositionwas summarisedbytheJudgeatJ/¶33.
61.BytheirSubmissionstheComptrollerstated(at[3]):”Insofarastheinventionis implementedinhardwareform(suchaswheretheANNisembodiedonspecial ‐ purposeelectroniccircuitrythatdoesnotrelyonexecutableinstructionsand/or hardwareonanintegratedcircuit),therelevant”computer”isthehardware(i.e.the specialpurposeelectroniccircuitry)onwhichtheANNiscreated,trainedand ‐ ultimatelyimplemented.”And(at[5]):”Insofarastheinventionisimplementedin hardwareform,thereisnorelevantcomputerprogramtowhichtheexclusionapplies. Inthepremises,hadtheinventionbeenlimitedtoa’hardwareonly’implementation, ‐ itisunlikelythatthecomputerprogramexclusionwouldhavebeenengaged.”Here, theComptrollerwasconceivingthatitispossibleforaninventiontoliein’better hardware’forimplementingANNsgenerally,i.e.,an”architecture”,”operateinanew way”and/or”bettercomputer”(AT&Tsignposts2,3and4).Thisappearstohaveled toamaterialmisunderstandingbytheJudge.
18
Page 20 of 26
Comptroller’sAppealSkeleton
62.TheJudge,encouragedbytheApplicant([2(b)]oftheApplicant’sSubmissions)seems tohavethoughtthattheComptrollerwassaying(inthewordsoftheApplicant)that “a hardware ANN is itself a computer, albeit they say a computer with no program”. ThatisnotwhattheComptrollerintendedand,sointerpreted,isacomplete nonsense.AstheApplicantsaid”This is an unsupported notion that is obviously wrong: computers are devices for executing computer programs”.Weentirelyagree.
63.Irrespective of the possible interpretations of the Comptroller’s Further Submissions, giventheimportanceofthiscase,theComptrollermakesclearhispositionasfollows:
(1)AhardwareANN: the hardware itself is a computer which has a specific hardware architecture. Rather than the architecture familiar to conventional computers (as described by this Court in Gale at 320 lines 29 to 321 line 30), as the Judge explained [J/15] a hardware ANN computer has “layers [of neurons] connected to each other, or at least some others, and to layers below…one feeds data in at the ‘top’ and it is processed down through the layers in accordance with the states of the neurons, each applying its weights and biases and passing the result on, until the result of the processing is reflected in an output at the bottom.”
(2)TheprogramforanANN: the ANN hardware is a known mathematical model which can be ‘trained’ to perform a specific application. In order to customise thegenerichardwareforaparticularapplication,itisthesetof’weightsand biases’ (which the Judge referred to as the “network parameters”, J/¶17) whichhavetobeconfigured.Itisthesetofweightsandbiaseswhichtherefore formthe’program’ofthisunconventionalcomputer.Thisaccordswiththe definitionsofacomputerprogramintheCambridge,CollinsEnglishDictionary andMacmillandictionaries,quotedbytheJudgeatJ/¶36:”asetofinstructions thatmakesacomputerdoaparticularthing”;”asetofinstructionsfora computertoperformsometask;and”asetofinstructionsstoredinsidea computerthatallowstheusertodoaparticularthing…”.Thedefinitioninthe FreeDictionaryisnotsatisfiedasthisincludes”asequenceofinstructions…” andisclearlylimitedtotheconceptoftraditionalsequentialcomputing.ANNs
19
Page 21 of 26
Comptroller’sAppealSkeleton
areadifferentarchitecturetoasequentialcomputer,buttheExclusionisnot limitedtoanyparticulararchitecture.
(3)SoftwareANNs:AstheJudgeheldatJ/¶18:”…an ANN can also exist in a computer emulation. In this scenario a conventional computer runs a piece of software which enables the computer to emulate the hardware ANN as if it were a hardware ANN. Professor Pardoe explains that this is slower than a dedicated hardware ANN. However, it has the same effect. It should be appreciated that there is…a distinction between the underlying software on the computer which “creates” the emulated ANN, and the emulated ANN itself.” We agree, in the case of a ‘software ANN’ there are two programs: (i) the generic software which emulates the hardware; (ii) the set of weights and biases which constitute the ANN programming, as in a hardware ANN.
(4)ANNTrainingSoftware: The ANNs discussed above are in the ‘operating’ or ‘inference’ stage (see J/¶27 – which distinguishes between the “inference stage” where the trained ANN is operating, and the “training stage” where the weights and biases are determined). In the training stage, additional software is used, together with a set of training data for which the desired output is known, to evaluate a cost function to incrementally change the sets of weights and biases (see J/¶16). The basic training program is a known algorithm, but the necessary configurations (inputs, training set, cost function etc.) are determinedbythepersonseekingtotrain,orprogram,theANN(seeJ/¶25). The ANN being trained may be a hardware or emulated (software) ANN.
(5)HardcodedANNs:Justasitispossiblewithaconventionalcomputerto ‐ permanentlyfixtheprograminthecomputer(e.g.liketheprogrammedROM inGale),itispossibletohardprogramaparticularsetofweightsandbiases ‐ (i.e.,theprogram)inahardware(orsoftware)ANN,e.g.,forpermanent applicationssuchasinadishwasherortoaster.Nosuch’permanent programming’hasbeenconsideredtoavoidtheExclusionforconventional computingarchitectures,andthereisnobasisforadifferentapproachinANN ‐ architecturecomputers.
20
Page 22 of 26
Comptroller’sAppealSkeleton
64.Aswillbeapparentfromtheabove,theComptroller’spositionisthat:(1)allANNsare computers(hardwareoremulated);(2)allANNsintheinferencestagehavea program:thesetofweightsandbiases;(3)foremulatedANNs,thereisadditional genericsoftwareforemulatingtheANNonaconventionalcomputer;and(4)inthe trainingstagethereisadditionalsoftwaretoperformthetrainingfunction.
65.TheComptrollerdidnotconcede,andnordiditfollowfromtheirsubmissions,thata computer‐implemented invention (whether or not it involves an ANN), whose claim necessarily encompasses software executed on a general‐purpose computer, is outside the scope of the exclusion. Indeed the Comptroller submitted (at paragraph
5) that “…where (as here) the claim covers both hardware and software implementations then the software implementation should be assessed according to the guidelines set out above in respect of computer‐implemented AI inventions”. The Application is not to ‘hardware’ or ‘software’ but to a computer‐implemented ANN.
66.Further, the Judge’s conclusion is based on the false premise that the software form of the ANN (in the inference stage) “is not implementing a series of instructions pre‐ ordained by a human” (see J/¶54). As we have made clear, an ANN (whether hardware or software) in the inference stage has such a computer program. The Judge gives no justification for his implicit contention that such program itself must be “pre‐ ordained by a human”. Insofar as he is however right, the program that results from thetrainingactivity(putinplacebyahuman)isnolessaworkderivedbyacomputer (in this case, the training program) under the direction of a human, than an ordinary machinecodeprogramusedtoprogramaconventionalcomputerhavingitselfbeen ‐ derivedbyacomputer(inthenormalcase,acompilerprogram)underthedirection ofahuman.Bothstartwithhumaninputandconfigurationanduseanintermediate computersystemtogeneratethefinalprogram.Noproperdistinctionexists.
67.EveniftheJudgehadbeenrighttotrytodissecttheclaimintocomputersand instancesofprogram,theJudgeshouldhavefoundthatatleasttworelevant programsexistinthecontextoftheANNalone.Namely,thefinalprogramcausing theANNtoevaluatethepropertyvector;andthetrainingprogram.Additionally, thereisaprograminvolvedinusingthepropertyvectortomakeafileselection,and
21
Page 23 of 26
Comptroller’sAppealSkeleton
(inasoftwareimplementation)afurtherprogramfortheemulationoftheANN.Had theJudgetakenhisowncourseproperly,hewouldhavefoundtheExclusionengaged.
Ground3:theJudgewaswrongtoexcludetheconsiderationofthemathematicalmethod
exclusionmodel
68.TheHearingOfficerheld(atDecision,[63])thatinsofarastheANNcantrulybe decoupledfromthesoftwareplatformthatsupportsitinthewaythattheApplicant suggested,thenitrelateswhollytoamathematicalmethodandisthereforeexcluded frompatentabilityundersection1(2)(a)oftheAct:
IamnotpersuadedthattheANNcantrulybedecoupledfromthesoftware platform that supports it in the way Mr Chacksfield suggests. However, even ifitcan,itisimportanttoconsiderwhatanANNisatthislevelofgenerality.It isanabstractmodelwhichtakesanumericalinput,appliesaseriesof mathematical operations (applying weights, biases and an activation function), andoutputsanumericalresultatsuccessivelayers.AclaimtoanANNorthe algorithmbywhichitistrained,inageneralandabstractsense,relatedwholly to a mathematical method and it fails at step 3. …
69.TheApplicantsoughttorely(at[107]ofitsSkeletonArgumentbelow)ontheHO’s findingthatthemathematicalmethodexclusiondidnotapplybecausetheANNcould notbedecoupledfromitsspecificapplication.TheComptrolleracceptedtheHO’s decisiononmathematicalmethod(hence,noRespondent’sNotice).
70.WhentheJudgeimplicitlysuggestedbyhisposthearing’NotetoCounsel’that ‐ (irrespectiveoftheApplicant’sAppellant’sNotice),hewaspotentiallyconsidering decouplingtheANNfromtherestoftheclaim,theComptrollersoughttocautionthe Judgethatheriskedwalkingontothemathematicalmethodexclusion:
AstotheANNitself,inageneralandabstractsense(i.e.strippedofthe softwarebywhichitistrainedand/orimplemented),astheHearingOfficer held(at[63]),thatissimplyanalgorithmand,assuch,relateswhollytoa mathematicalmethodandisexcludedfrompatentabilityonthatbasis.
71.Inresponse,theApplicantsaidthispointcouldnotbetakenonaproceduralbasis becausetheComptrollerhadnoRespondent’sNotice.TheJudgeacceptedthe Applicant’scontentionstating(atJ/¶82)thatiftheComptroller”wishedtorunitas analternative[theComptroller]shouldhaveservedsuchanotice”.
22
Page 24 of 26
Comptroller’sAppealSkeleton
72.Thismissesthepoint.TheComptrollerwasnotrunningthepointasanalternative. HewascautioningtheJudgeastotheconsequenceofanapproachwhichtheJudge appearedtobecontemplatingnotwithstandingthat(forverygoodreason)the decouplingpointwasnotbeingadvancedbytheApplicantontheAppeal.
73.BeforethisCourt,theComptrolleradvancesthesameposition:theCourt,proceeding inaccordancewiththesubstantivelawonexclusions,shouldnotdecoupletheANN fromtheinventionasawhole.However,iftheCourtismindedtodoso(suchthat theprogrammedANNaloneisthetechnicalcontribution)thentheCourtmust separatelyreconsidertheHO’sdecisiononmathematicalmethodexclusion;theHO rejectedtheapplicabilityofthemathematicalexclusiononlybecausethemethodhad tobeviewedinthecontextoftheinventionasawhole.
74.The Comptroller’s position is that an ANN, whether programmed (i.e., with weights and biases set) or unprogrammed is nothing more than a mathematical method.
TheapproachintheEPO
75.For completeness we summarise the approach of the EPO to assessing computer‐ implemented inventions under Art 52 EPC, and ANN inventions more generally.
76.InassessingcomputerimplementedinventionstheEPOadoptsa’twohurdle’ ‐ ‐ approachwhichincorporatesitssocalled’COMVIK’,amodifiedproblemsolution ‐ ‐ approachforassessinginventivestep.Theapproachissummarisedintheleadingcase ofG1/19(PedestrianSimulation)at[37][39]and[78][79],reproducedbelow: ‐ ‐
37.Tobepatentable,anyinventionhastopasstheeligibilitytestunderArticle 52EPC(i.e.itmustnotfallunderthe”noninventions”mentionedthere)and ‐ alsofulfiltheothercriterialistedinthatarticle(novelty,inventivestep,etc.). Forcomputerimplementedinventions,thetwofoldtestforpatenteligibility ‐ andforinventivestep(usingtheCOMVIKcriteria)isoftenreferredtoasthe “twohurdleapproach”… ‐ 38.Itmaybethatashifthastakenplaceintherelativelevelofeachofthese twohurdlesinthesensethatithasbecomeeasiertocleartheeligibilityhurdle ofArticle52EPC(seepointB.II.aabove)andmoredifficulttopassthe inventivestephurdle(seepointB.II.babove)ofArticle56EPC.Asresultofthis shift,itcouldbesaidthatthereisnowineffectanadditionalintermediatestep toassessthe”eligibilityofthefeaturetocontributetoinventivestep”.
23
Page 25 of 26
Comptroller’sAppealSkeleton
39.Thetwohurdleapproachforcomputerimplementedinventionsactually ‐ ‐ entailsthreesteps.Establishingwhetherafeaturecontributestothetechnical characteroftheinventionconstitutesanintermediatestepbetweenassessing (i)theinvention’seligibilityunderArticle52EPC,and(ii)whethertheinvention isbasedonaninventivestepvisàvistheclosestpriorart.Thisadditional ‐‐ intermediatestepservesasafilterforfeaturescontributingtoatechnical solutionofatechnicalprobleminviewoftheclosestpriorart.Onlythose distinguishingfeaturescancontributetoinventivestep. …
78. Patent eligibility, the first hurdle, is to be assessed under Article 52 EPC withoutconsideringthepriorart,i.e.withoutregardtowhethercomputers existedattheprioritydateoftheinvention.Theuseofacomputerinthe claimed subject‐matter therefore makes it eligible under Article 52 EPC (point B.II.aabove). 79.Forthesecondhurdle,thepriorartistobeconsidered.Inventivestepis based on the difference between the prior art and the claimed subject‐matter. Therequirementthatthefeaturessupportinginventivestepcontributetoa technicalsolutionforatechnicalproblemmeansthattheinvention, understood as a teaching based on existing prior art, has to be a “technical invention”.Theuseofageneralpurposecomputeralwaysconstitutesprior ‐ artinthiscontext.Theinventiontobeassessedunderthisprovisionneedsto be “technical” beyond the use of a general‐purpose computer. For this assessment,thedefinitionofatechnicalinventioninArticle52EPC,in particularthelistof”noninventions”inArticle52(2)EPC,canbeusefulfor ‐ determining whether specific features contribute to inventive step (see G 3/08,Reasons,point10.13.1). 77.AlthoughtheEPOapproachcombinestheassessmentunderArt52withthatunder Art56(inventivestep),ultimatelyboththeUKandEPOapproachseektodetermine whetherornotthereisanonexcludedtechnicalcontribution. ‐
78.Further, the EPO’s Guidelines for Examination (G‐II 3.3.1) take the view that ‘neural networks’ are based on computational models and are inherently mathematical in nature. Thus, at the EPO, ANNs necessarily engage the program and the mathematical method exclusions. Thus, computer implemented claims incorporating ANNs will only be patentable if they have a relevant technical effect such that a technical problem is solved in a non‐obvious way. See also Mitsubushi T 0702/20 at [10]‐[11].
79.TheComptrollerdoesnotsubmitthattheUKshouldadopttheEPOapproach,forto dosowouldrequiretheamalgamationoftheassessmentofexcludedsubjectmatter withinventivestep(whichhaslongbeeneschewedbytheUKCourts).But,the
24
Page 26 of 26
Comptroller’sAppealSkeleton
ComptrollerobservesthattheJudgmentcreatesasubstantialinconsistencywiththe EPOapproachandtheexistingdomesticapproach,whichhighlightstheerroneous approachadoptedbytheJudge.
80.TheJudgmenteffectivelyprovidesacarveoutforANNimplementedinventionsthat ‐ isnotjustifiedundertheEPOordomesticapproachtoexcludedsubjectmatter.This isbecauseanyANN(orinventioninvolvinganANN)maybeimplementedinhardware. FollowingtheratiooftheJudgment,itcannotbeexcludedasacomputerprogram undersection1(2)-evenwhere(ashere)theclaimalsocoverssoftwareimplemented ‐ embodiments.ThiscrosscheckillustrateswhytheJudgmentiswrong. ‐
Conclusion
81.Forthereasonssetoutabove,theCourtshouldallowthisappealandrejectthe Applicationasbeingexcludedfrompatentabilitypursuanttosection1(2)(c)oftheAct.
82.The Comptroller’s view is that, irrespective of the result in the present Application, the Judgment is at very material odds with the legal approach of the EPO and as previously understood in the United Kingdom. So much so that in the month between the Judgment hand‐down and the final Order hearing, the Applicant informed the Court that it had counted no fewer than 77 reports as to the change in the UK’s law on the patentability of ANNs. The message received by the industry appears to be that claims covering AI, and in particular ANNs, are not caught by the exclusions on patentability. We conclude with a quote from one such report (from Mewburn Ellis):
TheanswersprovidedbyMannJwereatoddswiththecurrentapproachof theUKIPOtoexaminationofcomputerimplementedinventions,andtherefore likely to be surprising for [those] with experience prosecuting patent applicationsforcomputerimplementedinventionsintheUK. ‐ MannJconsideredthattheprocessbywhichtheartificialneuralnetworkis trained is a program for a computer as such. However, he considered that, evenwhenimplementedinsoftware,thetrainedartificialneuralnetwork itselfisnotaprogramforacomputer,andthereforethattheclaimasawhole could not fall under the exclusion. … …ifthejudgmentwereconfirmedbytheCourtofAppeal,thiscouldindeed markthebeginningofamoreAIfriendlyeraattheUKIPO,albeitonewhich ‐ might lead to different results from their European counterparts. ______________
25
Appellant Submissions, Draft 4.1
Page 1 of 13
Emotional Perception Lady Justice Nicola Davies UKIPO Professor Pardoe Arnold LJ Brian Nicholson Edmund Eustace Birss LJ Ms-Edwards-Stewart
The Application
I would like to start with the Application, as to which there was some confusion yesterday, but I think we got there in the end.
The Judge summarises in [4] – [13] of the Judgment, and which we summarised in Annex 2 to our High Court skeleton which is in [SUPP/1/p/41] –
If I may start in the Judgment at [8] and [9]. There the Judge sets out the two ANNs which are involved in the training process.
The first, which he discusses in [8], is the so-called semantic analysis. This is a process used only in training, and what it involves is taking a written, hence the word semantic, description of a track and running it through an ANN which can analyse text, for embedding the track in the semantic embedding space.
Text analysis of this type was established at the priority date, and this analysis could be done with an off the shelf NLP text analyser, which is a kind of ANN.
This produces a set of coordinates, that is a vector, which points to a particular location in the semantic embedding space.
Answering ML LJ Birss’ question of yesterday, our understanding is that it is a vector in the sense that it describes the movement necessary from the 0,0,0 origin.
Note that NLP simply looks at the words use and perhaps their relationships in the text in a purely mechanical way. Whilst it manages to group text with similar meaning, it does not do so by any understanding of that meaning, and it does not work in that way at all. It is a mechanical analysis with no intelligence or understanding behind it.
For that reason a human cannot ‘read’ that NLP vector and understand what it is indicating – that for example a particular vector means that the description is of a happy, light track with a summery feel or whatever. “”
The embedding space does not work in that way, and that the vector cannot be read off by a human like that.
But what one does know is that whatever the semantic meaning of the text associated with a first track, another track whose semantic vector is closely associated with it will have a similar semantic description; whatever that description might be.
And conversely one located distally to the first track will have a different semantic description; again without any particular knowledge from their vectors of what that meaning might actually be.
Act S1(1), s1(2), s125, s130(7); EPC 2000 Art 52 (54 Novelty., 56 Obv); TRIPS Art 27 -1-.
Page 2 of 13
Emotional Perception Lady Justice Nicola Davies UKIPO Professor Pardoe Arnold LJ Brian Nicholson Edmund Eustace Birss LJ Ms-Edwards-Stewart
Thus the scalar distance between two tracks is used in the invention as a measure of semantic similarity or dissimilarity; but without any understanding of their actual semantic nature.
Secondly, as the Judge sets out in [9], tracks are analysed by a different, separate ANN for their physical properties, for embedding in the physical properties embedding space. Appn 23 This is a purely physical analysis, for music based on for example beats per minute, also 31 tonality, rhythm, timbre, based for example on a thing called a Mel spectrogram. Note that the properties themselves are extracted using standard techniques known at the time, and it is those which are fed into the physical ANN. The physical ANN then processes those inputs to produce a final output vector, which is the one referred to as V. FILE
But again, that vector has no specific or readily human identifiable meaning.
Although one knows that it is based on an analysis of the input physical features, one cannot ‘read’ the vector and understand anything about it.
That because there is no obvious relationship between the inputs and the output, and as you may have picked up when MLF went to Fig 7 of the Application yesterday, there Appn p.131 is no reason to think that even there will be the same number of dimensions in the input and output of the ANN.
And you can also see that from the structure in Fig 6 [SUPP/7/p.131] because you can see there that all the say tonality inputs feed into every node on the layer below, and ” so on, down to the output. So there is no ‘siloing’ of the inputs to particular outputs. So if you say had 10 inputs about tonality, 10 about rhythm, 10 about timbre, etc, you might have, say, 20 or 50 outputs, and none of them would be a ‘tonality’ output as such. The outputs are each individually a kind of blended consequence of the total inputs (albeit each individual differently blended).
As the Judge notes at the end of [9], this second ANN is the only one which is used once trained.
In training the two tracks in the first pair are run through this second physical ANN, and two vectors are spat out.
The scalar distance between those two tracks in the physical embedding space is calculated.
Act S1(1), s1(2), s125, s130(7); EPC 2000 Art 52 (54 Novelty., 56 Obv); TRIPS Art 27 -2-.
Page 3 of 13
Emotional Perception Lady Justice Nicola Davies UKIPO Professor Pardoe Arnold LJ Brian Nicholson Edmund Eustace Birss LJ Ms-Edwards-Stewart
There then follows what the Judge described as a significant ‘trick’ of the invention, which he sets out in [10] & [11], and if I could invite you to read that. Appn 7 – As the Judge sets out in [12], once trained, the result is that when faced with new – tracks for which there is no semantic description – the ANN can analyse their physical properties, find where they lie in the physical embedding space, and know that other tracks close to them in that physical embedding space will in fact have similar musical properties.
Thus by its analysis of the physical properties of a track, the ANN can recommend other tracks which are at musically similar, even if they are completely different in style or genre or date.
And conversely, just through the analysis of physical features the ANN will avoid recommending tracks which are dissimilar in their semantic content, even if they are similar in their style, genre or date.
But there are four further points to note here.
First the process which has been described does not converge the physical and – semantic spaces themselves. The physical space does not take on the same form as the semantic space – and you can see this most easily from the fact that they may well, and likely will, have a different number of dimensions, i.e., the respective vectors may have different lengths of 512 bits and 1024 bits.
What has been used is the scalar (that is directionless) distances in the semantic space to train the physical ANN, but the two spaces do not become the same.
Second I hope that this will be completely apparent, but the ANN is not making any – musical or taste-based judgment. The analysis is based only on the physical properties, and indeed as ML LJ Birss noted yesterday, even the semantic analysis is just a comparison of text qua words there is no understanding of their meaning in the NLP – process – nor anywhere else in the process.
The ANN neither understands nor expresses an opinion on the music itself.
Third – the resulting ANN of this training process, indeed the ANN of any training process, will have a structure which is different from any other ANN. A trained ANN is a new machine with a new structure.
Finally, fourth as is obvious, in an ANN the human (I avoid any contentious word) not only does not prescribe how it is that the ANN analyses the physical characteristics so that musically similar tracks are closely located in property space, but that is something which the ANN has to determine for itself from the data.
Act S1(1), s1(2), s125, s130(7); EPC 2000 Art 52 (54 Novelty., 56 Obv); TRIPS Art 27 -3-.
Page 4 of 13
Emotional Perception Lady Justice Nicola Davies UKIPO Professor Pardoe Arnold LJ Brian Nicholson Edmund Eustace Birss LJ Ms-Edwards-Stewart
And MLady&MLs will understand the reason for that it is not known to a human how – the physical sounds of a song actually relate to its semantic nature. No one, at least as far as we are aware, has ever worked out how to do that, but by using the invention a machine can be produced which is able to do that task.
And that is an important point no programmer is able to instruct the machine what to – do, since no programmer knows the answer. Conceptually, the “programmer” doesn’t even know how to start.
This is what we refer to as the intractable problem in our skeleton, and it highlights one difference between an ANN and a common parlance computer program.
The contribution and the benefits of the invention
MLF has shown you the claims, and I am not going to spend time on them now.
The Judge set out the contribution in [24], citing the Hearing Officer’s [53], but if I may go to the Hearing Officer’s decision to pick up also on where he set out the benefits of the claimed invention.
HO Decision §§53 & 54 and then back to end §49 at [SUPP/6/p.117]
Drawing these together, six points: First, the claimed invention is to an ANN-based system which encompasses the ANN but is more extensive than that. Second, the ANN, as is conventional, may be hardware or software implemented. But that is a matter of its implementation. Third, the system is one which in use receives a track (or other media file) and performs an analysis of the physical properties of the track; Fourth, the ANN is, from that analysis of those physical properties, able to physical identify other tracks which will have similar properties, but which will therefore have the benefit of being similar musically to the initial track. Fifth, a message containing the track recommendation and the recommended file is then sent to a remote device. And sixth, the system has real benefits: namely increased simplicity, speed and accuracy. As a result the recommendation message is quicker, better, and therefore different from the message of the prior art and therefore new.
One could summarise that even more simply by saying that the invention is a better system for analysing the physical properties of a music track … faster, more accurately and more simply … and sending a better recommendation message the message itself – being new, faster and better in the sense of it contains a better recommendation.
Act S1(1), s1(2), s125, s130(7); EPC 2000 Art 52 (54 Novelty., 56 Obv); TRIPS Art 27 -4-.
Page 5 of 13
Emotional Perception Lady Justice Nicola Davies UKIPO Professor Pardoe Arnold LJ Brian Nicholson Edmund Eustace Birss LJ Ms-Edwards-Stewart
Technical introduction
I am shortly going to look at the MLF’s submissions on the technical background, where he sought to give a particular, and we say entirely wrong, view of what an ANN is, and how he says it can be carved up between a general purpose structure and the weights and biases.
Before doing so, I think it is helpful to take a quick detour via the Act and the question of construction as to what a computer program is, so that we know what the targets are.
THE STATUTE
I’d like to start with by turning to the statute itself [AUTH/28/p567 (bundle)].
s1(1) says that a patent may be granted for an invention in respect of which certain conditions are satisfied.
The fourth of these is that the invention is not excluded by sub-sections 2 and 3 (also 4A).
And subsection lists out the prohibitions, with the ‘as such’ provision at the end.
s.125 says that the invention is, at least in general, that which is stipulated in the claim.
Now that is all obvious, but it makes a really important but very simple point what is – critical is to focus on the claimed invention.
And the importance of that is that one has to have regards to the invention at the level of generality at which it is claimed.
In this case, the invention is to a system or method of using a system containing inter alia an ANN.
It is common ground that an ANN is a structure of nodes, link, weights and biases.
In a hardware ANN, that is a box of electronics – Judgment [14].
And as the Judge held at [18] and [53], in a software ANN the same structure is emulated as if it were a hardware ANN on a software platform. And MLF confirmed that he agreed with this yesterday.
The point about discussing a hardware ANN, is that this is a convenient way of understanding what an ANN itself the thing at the level at which it is claimed is, — free from any confusion which might be caused by the presence of the software platform underlying it.
Act S1(1), s1(2), s125, s130(7); EPC 2000 Art 52 (54 Novelty., 56 Obv); TRIPS Art 27 -5-.
Page 6 of 13
Emotional Perception Lady Justice Nicola Davies UKIPO Professor Pardoe Arnold LJ Brian Nicholson Edmund Eustace Birss LJ Ms-Edwards-Stewart
As it is, MLF has rightly focussed his submissions on the ANN as a structure at that level, and quite rightly does not – at least as far as I can tell – seek try to make any point about the fact that a software platform can be used in some implementations.
So in answer to ML LJ Arnold’s question yesterday, the point about a physical ANN is that it helps clarify about the nature of ANNs themselves, free of other distractions.
The other way this can be viewed is the important passage from Vicom at [16] – which are important for understanding here, and if I could just go to those now.
[Auth/13/p.328 (bundle)] [15] & [16] which are the passages most widely cited in – the authorities.
This is all important stuff! But there are three points I would particularly emphasise: First [read out end of para 15] reiterating a computer running a program — which is directed to a technical process is not within the exclusion.
That is because the invention is then to the technical process. Not that process may be implemented via a computer, but it is not a claim to a computer program ‘as such’.
Second – [Read out first para of 16] – It is no objection that the invention can be implemented as specialist hardware or on a conventional computer, and it is wrong to distinguish between embodiments carried out in hardware or in software since those do not change the essential nature of the invention.
What the Board is getting at here is that if the invention is one which is agnostic of the exact manner of its implementation, then generally it will not fall within the exclusion since it will properly be characterised as a claim to that invention, and not to the manner of its implementation – whether that be by a computer program or not.
It is similarly or accordingly no objection that the invention can be implemented as specialist hardware or on a conventional computer. That simply does not matter.
Mlady & ML will obviously see the relevance of that to the present case.
So those are the two point about hardware ANNs, and they are really facets of the same issue which is that one has to look at the invention at the level of generality at which – it is claimed, and the fact that it could be implemented using a computer is no objection if it would otherwise be claimable.
Sorry – that was a bit of a detour from the point we were on – which was the Act and the fact that that requires a close focus on that which is claimed.
The second preliminary point is then the construction of the Act.
Act S1(1), s1(2), s125, s130(7); EPC 2000 Art 52 (54 Novelty., 56 Obv); TRIPS Art 27 -6-.
Page 7 of 13
Emotional Perception Lady Justice Nicola Davies UKIPO Professor Pardoe Arnold LJ Brian Nicholson Edmund Eustace Birss LJ Ms-Edwards-Stewart
Yesterday there was a discussion over what a computer was, and ML LY Birss raised things like analogue computers – which I have been business Googling overnight, but am still not much the wiser!
I can see that there could be an interesting debate about the wider edges of what a computer is, but what is critical to this case is what a program for a computer is. ”
That is, or at least I think was, common ground.
The Judge addresses this in [36].
He there sets out four definitions, each of which is broadly a sequence or set of instructions for a computer to cause it to do a particular thing.
And you will note that underneath, the UKIPO accepted these definitions at the hearing, and moreover have not sought to appeal them.
Indeed they put in some of the same in their skeleton at 63(ii). ϕ
And as we point out in our skeleton at 34, the description in Gale is possibly more ϕ accurate in that it refers to a series of instructions capable of being followed by a CPU; which is just a further clarification of what the word ‘computer’ in those dictionary definitions should itself be understood to mean.
And as we also note there in our skeleton, the language of instruction is also used in Aerotel.
That is why, in respond to a question of ML LJ Birss to MLF yesterday, there is a focus since it is agreed that that is what the statute is all in this case on the ‘instruction’ – about.
Further, as we will come to, MLF wider construction which he posited on his feet which ran ‘input, mathematical processing, output’ [T1/70] leads to what we submit are 9 absurd results.
The point then is really what an instruction is, as to which we say that this is a term freighted with two elements: First, that there is an element of command or order about it. It is something which tells someone (or here something) else what to do. Second, in particular in this context, that the instruction is a human direction as to what to do.
Tech background proper
With those introductory comments, I can now turn to the technical background proper.
Act S1(1), s1(2), s125, s130(7); EPC 2000 Art 52 (54 Novelty., 56 Obv); TRIPS Art 27 -7-.
Page 8 of 13
Emotional Perception Lady Justice Nicola Davies UKIPO Professor Pardoe Arnold LJ Brian Nicholson Edmund Eustace Birss LJ Ms-Edwards-Stewart
As we set out in our skeleton at §11, the Judge addresses the technical background in [14] – [18], … and Prof Pardoe sets out the basic structure of an ANN in his paragraphs 19 – 43 [SUPP/tab 9/ p.212].
Yesterday MLF sought to give YL’s a very carefully curated view of an ANN, designed to support his submission that an ANN can be divided into a general purpose structure ” laid on top of that. of nodes and links, with a ‘program’ of weights and biases
MLF seeks in his submissions to split the links and nodes, from the weights and biases, to say that the former is a computer and the latter the program.
He was at great pains to give the impression that there was a generic or general purpose set of links and nodes, which was programmed with weights and biases.
But we submit that that is artificial and wrong, for six reasons.
CHANGING STRUCTURE IN TRAINING
First, I would like to start with an important point, which MLF did not bring to your attention, which is that during training the structure of the links and nodes themselves may be changed.
If I could show you that – go to Pardoe ϕ39 [SUPP/9/p.216]
It is a short piece of evidence, but given the new way that the UKIPO now puts its case, it takes on some real importance.
What this underlines is that it is not so simple as MLF seeks to portray it in training – the nodes themselves (and therefore links too) may be pruned.
So, for example, in training some ANNs may prune nodes and therefore alter the remaining weights and biases in one way, and another ANN may prune different nodes, or none at all, and therefore alter the remaining weights and biases in another way.
So the question of whether and what nodes to prune (if at all, sure) and what weights and biases to ascribed are all interlinked.
MLF is just wrong to suggest that in ANNs the links and nodes are a fixed given, and the weights and biases the only variables.
Whilst Prof Pardoe says that this may only occur in some cases, the importance of the point is what it tells us about the nature of ANNs themselves namely that it is not right – that the training process is limited to only weights and biases, but rather the structure of the ANN is a composite whole. The point is that that the elements of that structure are interlinked and cannot be divided up as MLF would like to have it.
Weights and nodes are inherent features and cannot be divided out.
Act S1(1), s1(2), s125, s130(7); EPC 2000 Art 52 (54 Novelty., 56 Obv); TRIPS Art 27 -8-.
Page 9 of 13
Emotional Perception Lady Justice Nicola Davies UKIPO Professor Pardoe Arnold LJ Brian Nicholson Edmund Eustace Birss LJ Ms-Edwards-Stewart
Second, but relatedly, it is simply not possible to have an ANN which does not have all of the features – links, nodes, weights and biases.
You cannot remove the weights and biases from an ANN at all. Every link must have a value and likewise every node, since it is a feature inherent in being a link or a node.
It is quite unlike a computer, where one can download or remove programs at will, and the computer remains.
Again the point is that these features are intertwined.
Function is defined by its structure as a whole.
Third, relatedly, the function of an ANN is determined by its structure as a whole.
The links and nodes only work with the specific weights and biases, and vice versa, and they are all entirely intertwined.
You cannot take the weights and biases from one ANN and run them on another ANN as a discussion. -it just doesn’t work, or even really make any sense
But that is because an ANN can only function as a complete machine.
And where MLF yesterday went on about his examples of training one ANN and rolling it out across other ANNs, what is important to understand is that in each case you are rolling out the same complete package of links, nodes, weights and biases. You are not rolling our weights and biases on arbitrary generic ANNs, but simply replacing the entire complete thing in all its elements.
There is no evidence that anyone else anyway decomposes an ANN in the manner
the UKIPO now tries to do.
Fourth, there is no evidence that anyone, other than the UKIPO in this court, has ever sought to characterise or decompose an ANN in the way that the UKIPO does.
In his report Prof Pardoe describes it as a machine as a whole, indeed saying that after each iteration during training the new machine is a better machine. That is his report
24. He does not describe it in terms of an improved program for a machine. ϕ
And nor does he anywhere suggest that an ANN can be described as a general purpose computing part with a program overlayed.
– Indeed you may be aware that the OECD definition of an AI describes it as a ‘machine based system’, not a computer or computer program, and that same definition has been taken into the new EU AI Act too.
Act S1(1), s1(2), s125, s130(7); EPC 2000 Art 52 (54 Novelty., 56 Obv); TRIPS Art 27 -9-.
Page 10 of 13
Emotional Perception Lady Justice Nicola Davies UKIPO Professor Pardoe Arnold LJ Brian Nicholson Edmund Eustace Birss LJ Ms-Edwards-Stewart
Further the UKIPO is unable to cite anyone one else who describes an ANN in the manner they do either.
Given that they are advancing an argument which is completely new, and indeed a reversal of their original concession, one would have expected them to come up with articles or the like explaining the operation of an ANN in the terms they now urge on the Court, but they have nothing (and it is not like they are scared of a hand-up or two, or indeed referring to trade articles as they have included in the bundles already). But on this critical point, they have nothing.
And we do pray in aid, not only that lack of any support for this new point (one we note not supported by Prof. Pardoe), but that this was not a view which was espoused by the hearing officer, or by the UKIPO in front of Mann J, where the UKIPO explicitly took in fact the exact opposite position – a position which was no doubt entered into with careful consideration and upon advice.
All of that indicates the artificiality of the UKIPO’s position, which has no grounding in any evidence, and is not supported by the only evidence in this case from Prof Pardoe.
FUNCTION vs COMPUTER PROGRAM
Fifth, the argument of the UKIPO risks confusing the function of a device and whether or not there is a computer program in that device.
The point is this the fact that a device is able to do something which could be described – mathematically and outputting a result, does not mean that it is a computer or more relevantly using a computer program to do so.
MLF put it as ‘it inputs, mathematical processing, outputs’ – that was at [T1/70]. 9 But we submit that that is a flawed and unreliable guide.
The basic point is this. You could create an electronic device the function of which was to perform a particular process, which process could be described mathematically and to output an answer; and you could write a computer program to control a computer to perform the same process, even to do so in the same way.
But that does not make the electronic device a program for a computer – it is just that different approaches are being used to achieve the same function.
Thus one has to be very careful with equating the function of a device – even a function which can be described in mathematical terms with whether there is a computer – program in it.
So to take a simple example, a basic ohm-meter (that is a device which measures the electrical resistance of things) uses a collection of simple electronic components to take in an input, do an analysis of it which if you were so minded could be described in
Act S1(1), s1(2), s125, s130(7); EPC 2000 Art 52 (54 Novelty., 56 Obv); TRIPS Art 27 -10-.
Page 11 of 13
Emotional Perception Lady Justice Nicola Davies UKIPO Professor Pardoe Arnold LJ Brian Nicholson Edmund Eustace Birss LJ Ms-Edwards-Stewart
purely mathematical terms, and gives an output being the position of the needle on the dial.
The function it performs is a result of the particular values ascribed to the network of electronic components within it.
[T1/70].: it inputs, mathematical To use MLF’s submissions from yesterday 9 processing, outputs.
On approach, a basic electronic ohm-meter would be a computer with a MLF’s computer program within the Act, but in our submission it is without doubt that that is not the kind of thing which the legislature had in mind.
Sure the structure of an ANN is much more complex, and so therefore also its analytical capabilities, but there is no qualitative difference between that and the ohm-meter.
Furthermore an electronic device of a little more complexity, say some kind of hardware filter arrangement for an electronic circuit, may well be designed by an iterative process by which the parameters of the resistors, capacitors and amplifiers involved are modelled on a computer and tweaked to optimise the filter spectrum, and then put into production afterwards.
Just to unpick that, in an electronic device such as a mobile phone the transmission is received at a certain frequency, but the received message may be interfered with by a plethora of other signals at other frequencies. What you can do is use arrays of capacitors, resisters, things called inductors etc which have the effect of draining off or blocking the unwanted frequencies, but allowing the one you are after through.
These are all electronic components arranged normally in one of a number of well known structures to achieve this function, but the fine-tuning in any case involves a decision of the exact values to ascribe to each individual component.
On MLF’s case such hardware filters would be considered computers with computer programs (being the values assigned to the components), and yet in my submission that would be considered an extremely surprising conclusion, these are just standard electronic devices and plainly not what the Act was ever contemplating as having or being a computer or a computer program.
One could even go further – a sextant for example?
That is a device, there purely mechanical, which takes in an input, performs a translation which can be described if one wishes in purely mathematical terms, and spits out an output.
A sextant has a basic structure, but it is the markings on the various bars and levers which enable it to perform its function.
Act S1(1), s1(2), s125, s130(7); EPC 2000 Art 52 (54 Novelty., 56 Obv); TRIPS Art 27 -11-.
Page 12 of 13
Emotional Perception Lady Justice Nicola Davies UKIPO Professor Pardoe Arnold LJ Brian Nicholson Edmund Eustace Birss LJ Ms-Edwards-Stewart
Again, as he puts it, input, mathematical processing, output.
Sure, that is not even electronic, but is it not, on his case, a computer and a computer program? There I think the computer program would be the markings on the bars and arms of the sextant.
The point we are making here is not really that these are computers and computer function the device has is an unsafe programs, but that MLF’s approach of asking what way of identifying a computer and, more particularly, a computer program.
And further it shows that MLF’s suggestion that what mattered under the act was ‘input, [T1/70] leads to what we submit are absurd and mathematical processing, output’ 9 unprincipled results.
The components are not sending instructions, and to what?
Sixth, a
nother point about MLF’s case is to dig a bit deeper as to what he really means when he says that the weights and biases of an ANN provide the computer program.
Let us take a hardware ANN, which can be implemented in pretty simple forms.
Links can be just tracks of wire or the like.
The weights can just be an amplifier or resistor on that track.
The node itself can be implemented as a simple amplifier, with the bias an addition similar component.
So on MLF’s case, some of the amplifiers and components are the computer and other the program.
But they are all interlinked and intermingled in one complex structure.
Furthermore one can ask what are the ‘instructions’ on his case, and what are they ‘instructing’?
What he refers to are the individual current flows through the resistors and amplifiers.
But they are not passing instructions to anything at all. All they do is their normal electronic job of reducing or amplifying the current passing through them; they just function to restrict or increase the current passing through themselves. Weights and which affect structure and function, and not a program for a biases tangible “things” computer as MLF would have you believe.
Act S1(1), s1(2), s125, s130(7); EPC 2000 Art 52 (54 Novelty., 56 Obv); TRIPS Art 27 -12-.
Page 13 of 13
Emotional Perception Lady Justice Nicola Davies UKIPO Professor Pardoe Arnold LJ Brian Nicholson Edmund Eustace Birss LJ Ms-Edwards-Stewart
Sure this may have a downstream effect on another component, but a downstream effect is not to be confused with an instruction to that component to do or not do something. It is not telling it do something, but rather it is just performing its function with a consequential downstream effect, as happens in electronics. Instructions and consequences are functionally different concepts.
Intractable problem
The sixth point to pick up here again is that the operation of the ANN is fundamentally different from a computer program, not only in the lack of any series of instructions, but in that such function as is present is not in the form of a series of logical instructions defined by a programmer. That is the intractable problem point which I have already addressed you upon the human in most cases (of which this is one) has no idea how – to address the problem, let alone to write a series of instructions to solve it. To put this in the context of the invention and musical texture or timbre, there is no definition of — so the what amounts to “timbre” -it is “illdefined” in the language of the ANN arts programmer cannot program instructions that define this characteristic. However, within the ANN of the invention and in practical terms, the invention can approximate musical texture similarity (“timbre”) without needing a strict definition of a timbe which, in any event, doesn’t exist.
As I have submitted the term ‘instruction’ is one freighted in this context as human instructions, and here in an ANN there is nothing like this.
At most the human is involved in setting the parameters for the training program and – we do accept that there is a program involved at that stage. But once you turn to consider the final ANN itself, the human is not, and in any case other than the entirely trivial never could instruct the trained ANN what to do or how even to attempt to solve the problem at hand.
The Judge made a similar or related point in [77] and [78] of his Judgment, where he said that even if the UKIPO was right that there was a computer program at the training stage, then in any event the result of that training process is itself a separate technical thing – the trained ANN.
© 2024 Hepworth Browne
Act S1(1), s1(2), s125, s130(7); EPC 2000 Art 52 (54 Novelty., 56 Obv); TRIPS Art 27 -13-.
