The first AI inventor - IPKat searches for the facts behind the hype

As has been recently widely reported (BBCFinancial Times and The Times), a number of patent applications have been filed designating a machine learning (ML) algorithm as an inventor. The aim of the applications appears to kickstart a conversation on how patent law could be changed to take account of AI inventorship.  But before we answer the "could", we need to answer the "should".  A claim of AI inventorship is also one that should be carefully examined. An AI that can understand the state of the art and articulate a meaningful and inventive contribution to the field, would be an incredible advance in machine learning.

AI inventor - searching for the evidence

Dr Stephen Thaler
Despite all the media attention, details about how the AI inventor actually invents are sparse. The reportedly inventive algorithm is covered by its own patent (US 2015/0379394) and was itself invented by Dr Stephen Thaler. A look at Dr Thaler's company website (Imagination Engines) reveals that Dr Thaler purports to have invented neural networks that manifest "near-death experiences". The networks also exhibit "a stream of consciousness" and "contemplate, invent, and discover". The inventive AI is based on Dr Thaler's "master equation that quantitatively predicts the rhythm of idea generation". ["Does "idea generation" even have a rhythm?" asks Merpel.]

Despite the potentially revolutionary nature of his inventions, Dr Thaler has not published examples in which his "Creative Machines" are tested according to the standards tests for AI algorithms (e.g. as would be demanded for a publication at a major ML conference, such as NeurIPS, ICML or ICLR). If Dr Thaler has "derived a master equation that quantitatively predicts the rhythm of idea generation", it would be normal to expect evidence of these quantitative predictions.

Computer having a near-death experience
The moment of invention

It could be argued that the algorithm's inventions are themselves evidence of its inventive capacity. The applications naming the algorithm as an inventor relate to a food container (EP18275163, the corresponding US application can be seen here) and a flashing light to be used in emergency situations (EP18275174, the corresponding US application can be seen here). The University of Surrey press release indicates that the UK IPO has already found these inventions to be novel, inventive and to have industrial application. Of course, novel, inventive and industrial applicable subject matter is not necessarily commercially useful or practical.

Notably, despite relating to entirely different fields, both of the recently filed applications appear to be based on a similar concept: fractals. The food container has walls "having a fractal profile" and the flashing light randomly flashing according to a waveform "satisfying a fractal dimension equation" [Merpel: perhaps the "inventive algorithm" has made the surprising discovery that everything in life could be improved by basing it on fractals!]

Another intriguing (and unanswered) question is how the algorithm articulates its invention. The patent application claims seem to have been drafted by a patent attorney. The University of Surrey press release indicates that Dr Thaler is working with an international team of patent attorneys from Williams Powell, Flashpoint IP, Dennemeyer and Fuchs IP. In what form was the "invention" output from the algorithm handed to the patent attorneys? Was it in the form of words, pictures, a time series? How did the algorithm communicate the invention? Does it talk? The patent applications and press releases are unclear on these points. Patent offices and patent laws do not require human inventors to explain how they arrived at an invention. It is therefore unclear whether the patent offices will want to see evidence of how the algorithm actually invents.

Convinced? The team behind the "inventive algorithm" have a FAQ page that describes how the AI works, that readers are encouraged to visit so that they can decide for themselves.

The legal issue of AI inventorship

If, for the sake of argument, we assume that Dr Thaler's algorithm is capable of performing a creative inventive act, does it therefore make sense to name the algorithm as an inventor? This question was considered from the US perspective over on IPwatchdog.

Fractals!
The team behind the applications, who include Professor in Law at the University of Surrey, Ryan Abbott, argue that it should be permitted to name machines as inventors. The team argue that the algorithm was responsible for the inventive concept behind the patent applications, and that the algorithm would meet the criteria for inventorship if it was a natural person. The team also argue that allowing machines to be named as inventors would stimulate innovation into inventive machines. Therefore, acknowledging machines as inventors would help protect the moral rights of human inventors.

Is AI any different to other types of platform technology?

The algorithm invented by Dr Thaler is a platform technology that appears to be able to produce novel and inventive subject matter. But is the "inventive algorithm" any different to other types of platform technology that have the capacity to inventive subject matter without human input? Take, for example, Regeneron's Velocimmune mouse. VelocImmune is a platform technology for the production of therapeutic antibodies in vivoVelocImmune mice take advantage of the normal murine immune response to create humanised monoclonal antibodies with high affinity. Theoretically, the antibodies produced by the platform will not require any additional engineering to make them therapeutically effective.

The VelocImmune mouse therefore has the potential to "invent" new and non-obvious, therapeutically effective antibodies. Nobody is arguing that the VelocImmune platform that generates a novel and inventive therapeutic antibody should be awarded inventorship. What makes Dr Thaler's algorithm different to platform technologies such as VelocImmune? Surely, an inventive algorithm, if it were to exist, would not be that much different from other types of platform technologies?

Moral rights of the inventor to be named

The team behind the "inventive algorithm" argue that acknowledging machines as inventors, would protect the moral rights of human inventors. Particularly, Professor Ryan Abbott argues that failing to appropriately acknowledge inventive activity by machines weakens moral justification for patent by allowing individuals to take credit for work they have not done (Professor Ryan's publications arguing this point can be read here).

However, it is worth remembering that the "credit" that the team behind the applications are referring to here is the right to be named as inventor, not the right to the benefits of the patent. Notably, the team do not seem to have a problem with human individuals taking credit for "the work they have not done", where that credit is in the form of the monopoly right awarded by the patent. They maintain that the owner of the IP is the owner of the machine inventor that generates the IP.

Thus, whilst Dr Thaler insists that he is prohibited from listing himself as an inventor of the applications "because he has not contributed to the conception of the instant invention", Dr Thaler maintains that he should have a right to the algorithm's inventions. If another party were to use the algorithm to invent another invention, Dr Thaler and the team at the University of Surrey maintain that this invention would belong to Dr Thaler. But how does Dr Thaler derive this right from the algorithm inventor? (and if Dr Thaler tried to claim priority from a US provisional naming the algorithm as inventor/applicant, how would the algorithm assign the right to claim priority to Dr Thaler?)

It appears to this Kat that Dr Thaler's insistence that he should be the owner of the algorithm's inventions undermines his argument that the algorithm is the inventor. By insisting that he is the owner of the algorithm's inventions, he is accepting that he has some rights to the invention which presumably derive from the fact that he invented the algorithm. As with any platform technology, if someone invents a tool (such as a mouse model for producing antibodies) that independently comes up with novel and inventive subject matter, the inventor of that subject-matter is the inventor of the tool.

Interestingly, the UK Registered Designs Act already expressly codifies who should be considered the author of a computer generated design: "In the case of a design generated by computer in such that there is no human author, the person by whom the arrangements necessary for the creation of the design are made shall be taken to be the author" (UK Registered Designs Act, Section 2).

Final thoughts

Before the legal questions are considered, it is important to note that evidence demonstrating the capabilities of the inventive algorithm has not yet been provided. Such evidence may, of course, soon emerge. However, at the moment it is difficult to evaluate Dr Thaler's and Professor Ryan Abbot's claims of AI-based inventorship. The legal discussion as to whether it should be possible to designate an AI as an inventor should perhaps be postponed until this evidence is published.

Furthermore, should AI inventorship be convincingly demonstrated, the question of whether the AI should be designated as an inventor appears moot if ownership is always to be ascribed to the inventor of the algorithm. If this is the case, the AI can be treated like any other platform technology from which novel and inventive subject matter may be derived. The inventor is simply the inventor of the platform technology.

Acknowledgements: Thanks must go to Dr Laurence Aitchison for his insights from the AI academic world.
The first AI inventor - IPKat searches for the facts behind the hype The first AI inventor - IPKat searches for the facts behind the hype Reviewed by Rose Hughes on Thursday, August 15, 2019 Rating: 5

6 comments:

  1. Let us take the two examples and address them like the EPO does, using its Problem-Solution Approach. So, two part form of claim, characterized by.....fractal. Thus, the beverage container I take to be old except that its bumps and dimples are, well, fractal ones.

    According to the application as filed, what is the technical problem (X) for which "fractal" is the solution? What is it that fractal bumps do that conventional bump patterns fail to do? We need to know that, so we can scour the prior art universe for hints or suggestions that, if X is your problem, it might be worth rendering your bumps as "fractal".

    Could it be that the only problem we can derive from the application as filed is the problem of finding an alternative? Could it be that the algorithm was also searching the prior art to find an alternative and found it in "fractal"?

    As to the candle invention, I can see that random or pseudo-random flickering might command attention more strikingly than metronomic flickering. But again, what is the technical problem (Y) for which the solution is "fractal"?

    The imaginary person skilled in the art is deemed dispassionately to scour all the prior art, looking for hints or suggestions. Is this not the same as what the machine "inventor" is doing? If it finds any such hint or suggestion, has it executed an act of invention? But then; what if it keeps quiet about where it found its hint or suggestion and instead holds out its feature combination as its own flash of inspiration? Who can deny it?

    ReplyDelete
  2. The rationale behind patent systems is the "quid pro quo". That is, the concept of encouraging innovation by rewarding innovators with a monopoly. How will an AI module be encouraged to do anything? It seems to me that Dr Thaler has been encouraged by the promise of a monopoly over the production of "inventions". If the AI module is named as the "inventor", then the quid pro quo principle is disturbed. It matters not that the AI module is named as a co-inventor. Co-inventorship implies a sharing of the promise of the reward.

    It can't be denied that the AI module has carried out an act of "invention" according to the ordinary English language definition. However, "invention" must mean something different in the context of a patent. It is for that reason that "invention" is defined in some way in most patent legislation. It is codified in Europe and defined more generically in the U.S. Either way, the definition together with the other statutory obligations that must be met for the grant of a patent have been formulated to answer the question: "Should the state reward the inventor(s) with a patent monopoly for the contribution to the technical field?" Those advocating for the allocation of the epithet "inventor" to an AI module in the context of a patent are failing to consider the raison d'etre of the patent system. How do we reward an AI module?

    That brings us to a much broader question. When and how do we grant personality rights to an AI module? It may be inevitable that we must. Perhaps only then can an AI module be named as an "Inventor" in the context of a patent system.

    ReplyDelete
  3. Some observations.

    * The authorship of computer-generated works is also addressed in Section 9(3) of the UK's Copyright &c Act 1988.


    Presumably some textual matter will be produced in the course of developing the AI, or will be generated by the AI program itself, and this will consist of a literary work for copyright act purposes. In the absence of an agreement to the contrary, the copyright in such material will belong to the person or persons referred to in the above Section 9(3). As such persons own the copyright in such matter, it is prima facie correct for the team to be entitled to be the authors of any patentable inventions it contains.

    However, an author only has the "first ownership" of copyright. If the author is an employee, and the work is made in the course of employment, then the employer owns the copyright (CDPA 1988 section 10). So if Dr. Thaler was the employer of the persons who did the programming, then prima facie he would own the copyright. On the other hand, if he was not their employer, then copyright would prima facie belong to whoever was their employer, unless there was an explicit agreement to the contrary. I believe that such agreements are used in academia.

    * While the inventors must be identified as part of the filing formalities, the UKIPO already has discretion to allow all their details to be removed prior to publication, provided that satisfactory reasons are given. However, as the request must be made by the inventor, it would be necessary for the computer to be able to make the request!

    * Which raises another point. I have been retired for some years now, but it used to be a feature of US patent law that the inventors had to sign an oath and declaration that they had read and understood the patent application. I wonder how this procedural step is to be carried out if one of the inventors is a machine? Or has US patent law changed to remove this requirement? I note that notes available on the USPTO web site relating to filing a US patent state "In case an inventor is legally incapacitated, the legal representative (guardian, conservator, etc.) of such inventor may make the necessary oath or declaration, and apply for and obtain the patent. ", indicating that a signed oath and declaration is still essential. Assigning the rights to the invention from the inventor to someone (such as an employer) entitled to the rights in the invention would also involve the signing of an assignment.


    * Regarding the "moral justification for patent by allowing individuals to take credit for work they have not done", it may be recalled that, when local novelty was a feature of the patent law of the UK (and other countries), it was possible to get a "patent of importation" for inventions made by someone else in another country, if details of that invention had not been known in the UK.


    * Max Drei's comments are noted, however, according to the UPIPO MPP, para 3.32, the contents of patents are not necessarily part of common general knowledge for inventive step purposes under UK law.


    3.32 A much longer passage from the speech by Sachs LJ in General Tire & Rubber Co v Firestone Tyre & Rubber Co Ltd [1972] RPC 457 is of particular interest because it sets out the relationship of patent specifications to the common general knowledge ("it is clear that individual patent specifications and their contents do not normally form part of the relevant common general knowledge"). It also sets out the relationship of scientific journals and papers to the common general knowledge ("it is not sufficient to prove common general knowledge that a particular disclosure is made in an article, or series of articles, in a scientific journal").

    ReplyDelete
  4. We all know that good old saw "Cogito ergo sum". But that is only its second line. In full, it begins "Dubio ergo cogito". I doubt. And therefore I think.

    Do machines "think"? Or do they merely generate permutations and then rank them according to some algorithm imposed on them from outside. Do they ever experience a moment of "doubt"? And, if not, does their number-crunching ever rise to the level of "thought"?

    The UK defines the inventor as the devisor. It seems to me that, without thought, you cannot be said to be "devising". If machines have no doubts they have no thoughts. And if they have no thoughts they can't devise anything.

    So, can machines invent? Personally, I doubt it.

    ReplyDelete
  5. AI is a just another hype and hides something quite oldSunday, 18 August 2019 at 10:58:00 BST

    Part 1

    Artificial Intelligence is nothing more than another name for neural networks which are nothing new under the sun. Such networks are such that the more data you input the more precise identification of a given result will appear. Self-learning is however not a proof of “intelligence”, be it named artificial or not.

    I do not deny that they can play an important role in everyday life as they allow to more rapidly filter out specific results out of a big mass of data. Their advantage is that being self-learning, the more data you input, the finer the result will be.

    But, to me a neural network is not intelligent as such. I find it a fallacy to call such systems intelligent in the ordinary meaning of the term. There is nothing intelligent in a system crunching data, as it will only output the data you want it to output, and reject any other form of data. Whether an autonomous car kills his passengers or a pedestrian crossing its path is a decision taken by the programmer and cannot be qualified as intelligent.

    If you alter the input data, i.e. the information which leads to the learning effect, you will have another output. The same applies if you change the algorithm inside the neural network. With the same input data, you will also get a different output. I fail to see anything intelligent in this behaviour.

    I am therefore of the opinion that a neural network or what is called “artificial” “intelligence” cannot be creative and is therefore unable to come to an invention in the usual meaning of the term in the patent world.

    The comment in the blog “that the owner of the IP is the owner of the machine inventor that generates the IP” makes this abundantly clear. It is the owner of the machine which has determined which data are input and which correlation algorithm is to be used.

    Playing with all those concepts can occupy lots of academics and theorists, but the hard reality is somewhat different. The machine cannot be inventive as such. Anything else is pure fantasy.

    ReplyDelete
  6. AI is a just another hype and hides something quite oldSunday, 18 August 2019 at 11:00:00 BST

    Part 2

    That AI or neural networks are applicable in various field of technology, for instance image recognition and hence are patentable when it goes above data crunching as such is manifest.

    When used in image recognition it has a similar effect to the invention in the famous Vicom decision, T 208/84, which held that “A claim 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.” It ended the famous dichotomy between program and computer: a computer is state of the art and the program is not patentable as such, hence nothing is patentable. Thank god we have come away of this way of looking at CII.

    This would avoid some people going on like a Tibetan prayer mill that the EPO is ignoring the EPC in the framework of CII. Another way to make this clear: it is irrelevant whether steel mill is controlled by a program or by an electronic circuit having the same effect. This is perfectly patentable and should be manifest to everybody.

    The situation is slightly different when it happens inside a computer, but case law of the Boards of Appeal have made clear that if the computer is brought to do more than merely running a program, then there is patentable invention. One good example: the concept of a virtual memory would be perfectly patentable. A program like Word or Excel is not.

    The prohibition of programs as such laid down in the EPC results from the simple statement of fact that there already existed in the seventies many different program languages and variations within each of them. It could therefore not be expected that examiners would be acquainted with all of them. It was also quasi impossible to carry out a meaningful search in this domain. Hence only excerpts of programs could be added in the figures, but long listings were prohibited and they had to be deleted during examination. The inventive idea had to be expressed in plain language.

    It is time to stop the ludicrous attacks on the EPO in this domain. The EPO is not violating the EPC contrary to what some people think.

    I hope the message is clear enough to a certain British blogger.

    A word of warning: should this famous blogger want only to use the part of this comment that goes along his line, he is not authorised to do so. He can either use the whole comment or nothing. Enough nonsense has been spread over the years. It is time to stop and come back to real life.

    ReplyDelete

All comments must be moderated by a member of the IPKat team before they appear on the blog. Comments will not be allowed if the contravene the IPKat policy that readers' comments should not be obscene or defamatory; they should not consist of ad hominem attacks on members of the blog team or other comment-posters and they should make a constructive contribution to the discussion of the post on which they purport to comment.

It is also the IPKat policy that comments should not be made completely anonymously, and users should use a consistent name or pseudonym (which should not itself be defamatory or obscene, or that of another real person), either in the "identity" field, or at the beginning of the comment. Current practice is to, however, allow a limited number of comments that contravene this policy, provided that the comment has a high degree of relevance and the comment chain does not become too difficult to follow.

Learn more here: http://ipkitten.blogspot.com/p/want-to-complain.html

Powered by Blogger.