back to article AI-friendly patent law needed 'as a matter of national security', ex-USPTO boss says

America urgently needs to rewrite its patent laws to recognize modern artificial intelligence technologies, business and IP leaders have said. This sentiment emerged from a series of hearings organized by the US Chamber of Commerce, during which experts from academia, industry, and government were invited to speak. The …

  1. cornetman Silver badge

    > Thus, changes are needed in the law to allow machine-learning systems to be accepted as novel inventions, it was argued. And being able to patent and protect these inventions encourages businesses to build commercial products, we're further told.

    Therein lies the rub, as was hinted at a little in the article.

    Patents are (or are supposed to be) a social pact. In exchange for *full* disclosure of the innovation, then a limited monopoly is granted.

    The ability for someone skilled in the arts to be able to reproduce the innovation from the patent is built in. Without that, there is no justification for granting the monopoly.

    1. Yet Another Anonymous coward Silver badge

      That shouldn't be a problem for patenting a machine learning model.

      You have been able to patent a recipe for steel for 200years without knowing why adding these particular elements in these ratios give these properties. So patenting a Tensorflow model that can differentiate X from Y if you give it these weightings is no different.

      The main issue is that in 20 years when the knowledge becomes public domain it will be useless because we won't have Tensorflow (or at least that version) anymore

      1. Schultz

        It'll be a big clusterfuck

        Your statements nicely illustrate the problem with 'AI patents': So your novel, patentable method would be to use Tensorflow to recognize cat pictures? (Feel free to substitute the cats with whatever you consider useful.) Or do you really, in your heart, want to patent the idea of using AI for recognizing said cat pictures? Or do you want to only patent the use of your particular training data in combination with Tensorflow? Or maybe the use of any equivalent AI system that could give somewhat similar input-to-output results?

        Whatever you try to do, either you get a monopoly for an idea as opposed to a method -- and that would break the very innovation the patent system is supposed to foster -- or you try to patent data or software.

        AI is so interesting to so many, because it's a black box that promises to solve all problems of this world. Because it's a black box, it's really hard to prove that it cannot, potentially, solve a particular problem. You could spend decades of your life trying to understand and solve a hard problem. Or you could just use this magic box.

        I predict that AI will go out of fashion, gradually, over the next 10 years. All those great AI projects will be extremely successful. (You wouldn't tell your boss otherwise, would you? After all, it's research and it's OK to move the goalposts a bit, innit.) But after a few rounds of funding, the weeding-out will begin.

        1. Falmari Silver badge

          Re: It'll be a big clusterfuck

          @Schultz "AI is so interesting to so many, because it's a black box that promises to solve all problems of this world."

          You are right they went to be able to patent a black box. But if the patent is a black box how can it be defended. How can you prove that someone else's black box infringes your black box patent.

      2. cornetman Silver badge

        > You have been able to patent a recipe for steel for 200years without knowing why adding these particular elements in these ratios give these properties. So patenting a Tensorflow model that can differentiate X from Y if you give it these weightings is no different.

        The patent doesn't need to explain why the innovation works, merely contain all the information required for someone else skilled to reproduce it.

        The "skilled in the arts" part merely indicates the knowledge and materials available at the time prior to the innovation that would be required as a prerequisite. No-one seriously would claim to be able to reproduce an innovation merely from the patent from first principles, but it must be possible for *someone* to be able to do so with the appropriate amount of domain knowledge that was available at the time. This works because no innovation springs out of nowhere. It is all incremental in some respect, building on what has come before.

        1. Yet Another Anonymous coward Silver badge

          Yes that was rather my point. There is no difference in "add one cup of chromium" and "put the following magic values into Tensorflow"

          The problem is that, in return for telling the world how to make Invar the maker gets a 20year monopoly and then the whole world benefits and everyone can have Invar for ever-after.

          With a patent on an AI model there is no giving back to the world because in 20 years that list of weights to recognise a cat won't be of any use to anyone except computer historians

  2. Il'Geller

    Unfortunately, the creator of the AI, that is I, was not asked. The only innovation that led to the creation of AI is the new AI-parsing. Apart from it nothing new has happened, it is what allows to discover textual information in its context. This is the only novelty.

  3. razorfishsl

    Patents should be for Humans only,

    This prevents greedy bastards entering all man kinds knowledge then using AI to mine it bare all for the sake of a $

    Same with DNA, no one should be allowed to "patent " it, it would be like patenting chess moves.

    1. Short Fat Bald Hairy Man
      Pint

      Yes.

      Have a million of these. Or more.

  4. Flocke Kroes Silver badge

    The plan

    1) AIs to generate 100 patents per second.

    2) AIs to 'read' and award 99 patents per second.

    3) AIs to examine descriptions of profitable human inventions and decide which patents they infringe.

  5. amanfromMars 1 Silver badge

    To Whom IT and AI May Be of Grave Concern ....

    ..... when Words Can So Easily Create Dreams, Command and Control Existences and Destroy Worlds.***

    Is one correct in assuming that WMD* are a type of innovation which, whilst they can be patented, are largely agreed by more than just an elite chosen few to best be left unpatented?

    After all, who would be wanting to enable any Tom, Dick or Harry/Sheila, Janet or Jane person skilled in the art or science, of which it is a branch, or with which it is most nearly connected, to make, compound, and use the same." That means someone suitably skilled should be able to take a patent text and diagrams, understand what's going on, and reproduce the technology themselves.

    Whenever one realises pretty much everything new discussed here in this report by Katyanna on US Chamber of Commerce concerns is remarkably, disconcertingly easily rendered and wielded as a WMD, humans may have very good reason to not meddle, other than strive to ensure a safe and secure human prohibition in/to that which is clearly better left beyond their command and control because of the catastrophic dangers involved and evolving.

    Current IP laws do not recognize non-human entities as inventors, meaning machine-learning systems cannot be recognized as such.

    :-) Fortunately such recognition is not a pre-requisite for Alien Inventions and ACT**ive Interventions in Advanced Astute Agile IntelAIgent Developments.

    * .... Weapons of Mass Destruction or Distraction .... Deceit and Degradation

    ** ....ACTive CyberIntelAIgent Threat

    *** ... or would that be a Bastard Nightmare to Program and be held responsible and accountable for?

    A little something else to mull over in the club/around the conference table/before the board meeting/over tea and biscuits/at the bar as it wends its merry way on paths well established and long ago pioneered and perfected?

    1. Dave 126 Silver badge

      Re: To Whom IT and AI May Be of Grave Concern ....

      > Is one correct in assuming that WMD* are a type of innovation which, whilst they can be patented, are largely agreed by more than just an elite chosen few to best be left unpatented?

      During WWII a 1917 law was dusted off off, giving the US Patent and Trademark Office the power to prevent an inventor publically disclosing their invention if it was thought it would weaken the US war effort.

      Of course I could be in possession of the full blueprints and construction procedures of a hydrogen bomb and be totally harmless - I just can't get the raw materials. It's moot whether I have the plans. Doesn't matter if it's patented.

      Conversely, I already have all the materials required to create a biological weapon - I just need to know what order to put the A, C, T and Gs in. It is in this area that serious ethical questions are being by researchers of themselves.

      https://www.nber.org/digest/jul19/wwii-policy-kept-patents-secret-slowed-innovation

      And discussion from 1939 arguing that payments should be respected in time of war even if the invention is that of an enemy inventor (it saves on the backlog of payment claims come any armistice, apparently). Possibly a tangent, but charming:

      https://www.nature.com/articles/144580a0

      1. amanfromMars 1 Silver badge

        Re: To Whom IT and AI May Be of Grave Concern ....

        During WWII a 1917 law was dusted off off, giving the US Patent and Trademark Office the power to prevent an inventor publically disclosing their invention if it was thought it would weaken the US war effort. .... Dave 126

        That appears to suggest that first degree murder is the only surefire power one has to exercise if one is unable to persuade a human inventor against publication ...... Ye olde 00 Licensed to Kill franchise.

        A slippery slope that whenever it can so easily be copied and executed by aggrieved enemies or then seriously disadvantaged frenemies.

        And non-human entities with secrets to tell are something else altogether quite different to come to terms with ...... which might be just as simple as covering their agents and agencies lavish expenses with stashes of flash cash. It’s certainly worth a try whenever it works so well and so often for humans of all shapes and sizes.

        Thanks for all of the info, Dave 126. Much appreciated.

        1. Yet Another Anonymous coward Silver badge

          Re: To Whom IT and AI May Be of Grave Concern ....

          It may be an urban legend - but supposedly the Germans patented an anti-tamper fuse they invented in the 1930s.

          When they subsequently delivered some samples to London in the early 1940s the Brits were able to read all about the design and the description of its operation

  6. Neil Barnes Silver badge
    FAIL

    good reason to give machines at least authorship rights

    My left elbow!

    My pen's just claimed authorship rights over the deathless prose I wrote using it.

  7. heyrick Silver badge
    WTF?

    Greedy bastards blatantly milking a broken system

    The number of legal cases regarding patents suggests that the granting of such is more rubber stamping them anything resembling diligence.

    A patent is granted and then the holder gets to shake down everybody else in court (it's set up so they pretty much have to defend the patent) until such time as the patent is declared invalid.

    And now? They would like a machine, that is not a moral "person", a machine capable of creating hundreds of patents every second, to not only create patents but also to be able to be awarded them?

    This ought to count as some sort of abuse of the legal system, but lawyers are likely too busy rubbing their hands in glee to care about the abuse angle...

    1. Michael Wojcik Silver badge

      Re: Greedy bastards blatantly milking a broken system

      USPTO rejects about half of all applications. That's hardly "rubber-stamping".

    2. Neil Barnes Silver badge

      Re: Greedy bastards blatantly milking a broken system

      Sounds rather like Charlie Stross' "Accelerando"

  8. Anonymous Coward
    Anonymous Coward

    Less, not more!

    What the US patent system needs is to be shrunk, not expanded, starting with removing the ability to patent software!

    1. Doctor Syntax Silver badge

      Re: Less, not more!

      Make the USPTO liable for negligence for patents it grants which are later proved to be invalid.

      1. Yet Another Anonymous coward Silver badge

        Re: Less, not more!

        So making it much harder and more expensive to obtain a patent. That will discourage huge corporations with 1000s of lawyers and billions of $ - while protecting the ingenious inventor in his or her shed

  9. Doctor Syntax Silver badge

    Most experts people with something to gain by it agree AI algorithms should be patentable

  10. Doctor Syntax Silver badge

    The two paragraphs starting at "But take a system with a trained neural network." say pretty well what's the whole problem with this: trusting a black box to get the right answer. "But reproducibility is notoriously difficult in machine learning."

    However the next bit: "You need access to the training data and other settings to recreate it. That becomes problematic if the data is medical or personal info, or proprietary" really suggests the solution to the conundrum. It's data that matters. It can be protected by copyright, assuming, that is, that the trainers had the rights to that data in the first place.

    1. Ken Hagan Gold badge

      Plus the fact that, if all you have done is build a neural network and trained it, then I'm afraid it isn't novel. We already know how to do that. The fact that we haven't yet built and trained all possible NNs does nothing to make your one "non-obvious" to "one skilled in the art""anyone with a clue".

      No "invention" that consists of just a trained NN (with and input and output mechanism) should be patentable. It is old and obvious. There is no benefit to society in granting you a monopoly.

      1. Graham Cobb Silver badge

        I don't think that is quite right. Think back to when Edison got his patent for the light bulb...

        He had an idea for an electric lamp. He probably wasn't the first person to have the idea. But ideas are not patentable - society requires the idea to be implemented, and that implementation to be described so it can be reused.

        So, Edison invested a lot of money in creating a working light bulb. He built labs, he employed many people, they tried many different designs (the bulb, the type of wire in the filament, the way it was wound, the atmosphere inside the bulb, ...). That investment of time and money seems, to me, to be worthy of being granted a patent.

        If we apply that to today, I agree there should be an incentive for people to train NNs to do certain tasks. That training seems analogous, to me, to the labs of technicians building and testing light bulb designs.

        So, I can see that some patents for NN machines may be reasonable. We want to reward the people who turn the idea of "use NN for task X" into a working solution. On the other hand, society needs the other side of the coin as well: the result must be that once the patent has expired, others can build the same device. So the how can't just say "train an NN" - it has to include either the results of that training or the training data, or something so that others can re-use it.

        And as for the ludicrous idea that some machine or software could be treated as "the inventor"...

      2. Michael Wojcik Silver badge

        if all you have done is build a neural network and trained it, then I'm afraid it isn't novel. We already know how to do that

        This is flat-out wrong.

        There are infinitely many ANN architectures, and even if we confine ourselves to those that are both practical to implement and useful, it's a vast space. There are a similarly huge number of training processes, objective functions, rectification functions, and so on.

        Moving from GANs to Diffusion networks was most definitely a novel innovation, for example.

        And none of the interesting models are just a single ANN. They're complex stacks of ANN layers with complex interconnects. Even with a single layer, it's not just a question of "a neural network" – you haven't even specified recurrent or convolutional or whatever.

        Your argument is absurd, frankly. It's like claiming we know how to write software, it's "old and obvious", and so nothing new can ever be done with software.

        1. Anonymous Coward
          Anonymous Coward

          You can't patent software, try again.

          1. Yet Another Anonymous coward Silver badge

            You can patent software algorithms and even business models

        2. amanfromMars 1 Silver badge

          A Quantum Leap for Understanding Enlightening Master Piloted Projects

          There are infinitely many ANN architectures, and even if we confine ourselves to those that are both practical to implement and useful, it's a vast space. .... Michael Wojcik

          With command and control of the prime supply of novel undeniably accurate product/information/intelligence presentation, are television channels/audio visual broad banded streaming services servering images/pixel clusters and messages, one of the most useful practical implementations of an ANN architecture’s sublime and surreal abilities/facilities/possibilities for the creation of future secured events and their subsequent provisional supply programming with progressive performance virtually guaranteed a metadata base physicalised augmented reality.

          And a simple documentary series of the myriad complex processes involved and enabled by primed premium state and any number of rogue freelancer pirate non-state actors and their agents presenting the product of the realities created, is well within the reach and capacity of current universal systems administrations with Quantum Leaping Enlightening Master Piloted Projects Plan Blueprints to Follow/Narrate and FutureBuild Upon.

          A little something different for the likes of an Apple TV Empire or Middle Kingdom Media HeadQuarters to Run With and Take and Make Over Worlds by Remote Virtualised Storm? Yes,.... that’s easily done nowadays with safe and secure command and failsafe control of monumental 0Days.

  11. yetanotheraoc Silver badge

    Reproducibility?

    If a black box can't be reproduced exactly, and yet a patent is granted, then in litigation how will the defendant prove that they did _not_ infringe the patent? -- No doubt the lawyers already thought of that, but they charge by the hour.

  12. Long John Silver
    Pirate

    1793 - too far behind the times for IP applied to digital artefacts to continue?

    In 1793 and thereafter until onset of the digital era matters IP were relatively simple to understand.

    Slight correction: arrival of non-messy and cheap to use photocopiers together with home taping set the seal upon nightmares yet to come for IP rentiers. In retrospect, that worry was trivial.

    Digitally represented information, regardless of whether conveying entertainment or used as 'recipe' in 3D printing, has turned the tables upon conventional thought about what may be deemed 'property' and how property rights may be protected. At present threat is greatest for digital artefacts covered by copyright. Eventually patents will be damaged and later go into abeyance too. Patents are prone to two kinds of attack.

    First, a poor nation might sensibly decide no longer to abide by international conventions; pharmaceutical products being likely precipitating factor. Its leaders if neither 'bought' by IP interests nor within reach of US Marines may do their sums and conclude the national economy (including individual disposable incomes) would be bettered despite invalidation of IP owned by its citizens. Should a major nation, e.g. China or Russia, decide to scrap all IP law (replacing it with entitlement to attribution for creative people), it would be the 'nuclear' economic option; however, despite loud bleating by people whose livelihood depends upon distributing IP 'products', closure of law firms, and loss of sundry middlemen, the entire globe would end up the better for it.

    Second, despite patents (including those involving digital artefacts) at present being far more easily enforced than for copyright digital materials, their vulnerability is immense. Automated manufacture is becoming both easier to organise and cheaper to implement; advances in 3D printing presage return to small scale local manufacture to meet local demand: cottage industries. Depending upon complexity of that to be printed and demand level for output of particular physical artefacts, we may expect 3D printing plant to spring up in towns, neighbourhoods, and family residences.

    This ubiquity will extend to pharmaceuticals because technology is in the offing to brew drugs etc. in small amounts and bespoke for particular patients. Starting in hospitals this will spread to primary care services, perhaps beyond too.

    The point is that 'recipes' can be shared just as at present is the digitised howling of a 'pop star'. Access to required ingredients is unlikely to be difficult; perhaps, spawning further cottage industries.

    It's no good jumping up and down complaining about broken law and supposed 'property rights' denied. That which is not enforceable - prospectively or retrospectively - makes for bad law and becomes a matter of personal rather than societal morality.

    Inception of the digital era exposed the fundamental flaw in the concept of IP. Hitherto, IP-related products, e.g. books, recorded music, and, latterly, A/V 'content', depended upon being 'inscribed' on a physical medium, e.g. paper and vinyl disk. The medium containing the 'content' could be sold according to traditional market-economics supplemented by legal monopoly protection.

    Digitally expressed 'content' no matter upon what inscribed is separable from its physical substrate. Digital sequences cannot be corralled once arriving in public spaces. Sequences, are easily replicated with accuracy and readily distributed electronically with a variety of storage media available. Thus, they lack the "scarcity" necessary for market price-discovery; no amount of legal proscription and technological defences can force digital 'content' back into being ersatz physical property.

    Truly creative people will still be able to make a living so long as they convince others to offer patronage for the construction of creative 'content'. Patronage differs from 'investment' because finished products, being digital, possess no intrinsic (scarcity derived) monetary value. The Internet offers global reach in seeking patronage e.g. by crowd funding. Reputation, derived from attribution, underlies a creative person's imagination and skills being marketable (not sale of end products).

    Leonardo da Vinci made physical constructs and disseminated ideas. He attracted greater patronage as his reputation spread. Even should it have been feasible, da Vinci, and his ilk, would never have impertinence to demand every viewing of his works, and every 'derivation' from his ideas', attract royalty payment.

  13. Mike 16

    Reasonably Skilled

    It has ben clear for decades that the descriptions and figures in patents are _NOT_ designed to be understandable to one reasonably skilled in the art. They are designed to be foggy enough to allow wide ranging calls of infringement, and to be so obfuscated as to drive the would-be implementer mad. I'd be hard pressed to implement one of my own patents from _only_ the description. This obfuscation is what wer pay lawyers for.

    As for what the "originals" knew, well Boole's writings were originally about a multi-value logic (hand-waving, "A bit like Fuzzy Logic"). Only later did the bit-heads decide that "well, only zero and one really count". Much like those who think a "baud" is a "bit per second". Your gig-ethernet begs to disagree.

    (BTW: Boole also wrote on theology and related subjects. Much like others of his era.)

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like