Episode 6 - AI or Human: Which is the Inventor?

From self-driving cars to voice automation in homes, inventors are developing ambitious AI technologies that will continue to impact the ways in which we learn, work, communicate, and travel. According to the United States Patent and Trademark Office (USPTO), the number of AI-related patents increased at an astonishing number, from 4,598 in 2008 to 20,639 in 2018. In this episode of Exploring AI Matters, Rama Elluru, a former Administrative Patent Judge on the Patent Trial and Appeal Board at the USPTO, and Christian Hannon, a patent attorney in the USPTO’s Office of Policy and International Affairs, share their perspectives on how AI is impacting patent law and the inventor community. [2022-10-17]

Transcript

SPEAKER_04 0:03

Welcome to Exploring AI Matters. This podcast series, previously known as Mind the Gap Dialogues on Artificial Intelligence, will continue to appear in the ABA series to the extent that. In addition, all of the episodes, old and new, will now appear under our new podcast name, Exploring AI Matters. Thank you.

SPEAKER_05 0:48

Given the innovative and creative applications of AI, its development and use raise important legal consideration, most notably in the areas of intellectual property and national security. Technologists introduce all manner of innovations, most of which disappear without much of an impact. The legal community seldom focuses on this froth until a technology begins to gain wide adoption and its use causes policy controversy. As AI evolves to increasingly perform innovative tasks, the idea that an AI-augmented machine can invent something useful when previously only humans could be inventors raises a number of complex issues, such as whether AI may be enlisted as an inventor on a patent. The comments that the US PTO received expressed the consensus that existing laws and regulations did not require change to deal with AI. On the other hand, the National Security Commission on AI subsequently recommended that laws be changed to protect intellectual property rights in AIs and their outputs as a matter of U.S. national security. In this episode, we are fortunate to have one current and one former member of the US PTO staff involved in the preparation of those two critical reports. Our guests today are Rama Eluru and Christian Hannan. Rama Eluru, until quite recently, was an administrative patent judge on the Patent Trial and Appeal Board at the USPTO, and she is now at a nonprofit formed in October 2021 called the Special Competitive Studies Project, or SCSP, which develops recommendations to improve U.S. competitiveness in AI and other emerging technologies. Christian Hannan is a patent attorney in the US PTO's Office of Policy and International Affairs and a co-leader of the agency's AI Policy Working Group. Hello, I'm Mark Donner, a computer scientist.

SPEAKER_02 3:18

Hello, I'm Roland Trope, a national security lawyer. We are your hosts for this episode of Mind the Gap Dialogues on Artificial Intelligence.

SPEAKER_05 3:28

In addition, we have two more hosts.

SPEAKER_00 3:31

Hi, I'm Anna Adams, a national security lawyer.

SPEAKER_04 3:34

And I am Charles Palmer, a computer scientist.

SPEAKER_05 3:38

Each episode will be led by two of us, with the other two adding impromptu questions and comments as the spirit moves them. So you are both experts in intellectual property, but when and why did you start to delve into legal issues involving AI?

SPEAKER_01 3:56

Sure, I can start us off. So this is Chris Hannon. Just to make a small disclaimer here, I am preparing uh these remarks in my own personal behalf. These are not official agency remarks. I'm here in my current personal capacity today, but I do thank the organizers and the podcast for having me and inviting me to speak. And I think, you know, for me, uh getting involved in issues of artificial intelligence was really uh to some extent a product of the portfolio that I cover at my day job at the US PTO. And part of that is I think I had a target on my back, so to speak, because I had my educational background was in Comp Psy, electrical engineering, computer engineering. So when uh the public momentum around AI and IP started to heat up, they of course said, well, Chris can do it because he's got the technical background. And of course, Chris couldn't do it without having done a whole host of legwork uh to really ramp up and get started. But I'll just leave that there and uh turn it over to my colleague Rama.

SPEAKER_06 5:00

Thanks, Chris. I too want to thank the organizer and hosts uh for including me today. And I want to make the same disclaimer I'm here in my personal capacity, and nothing I say should be attributed to any former or current employers. Um, so I actually also have a computer science background, have a BS in computer science, and my concentration was in something that's akin to AI. I don't know what it was called back then, but it was many generations of AI ago. Um, and then I also my patent work uh concentrated on electrical engineering. The USPT hosted up the AI policy committee. And I want to say in 2018, um and I became a member of that committee and I was just really interested in AI. I it was clear to me that AI was going to be our future, is our future. AI is, you know, it's ubiquitous to our our everyday personal lives. You can find it in our phones, you can it's used to make our you know recommendations to us on our social media platforms. Uh, you can find it in our cars. And so the more I learned, the more interested I became in the issues raised by AI and other emerging technologies. And um, you know, these issues span tech policy as well as legal. So uh that's how I got started.

SPEAKER_05 6:17

So before we get deep into the discussion about AI, can either of you explain what qualifies an invention for patent protection under US law?

SPEAKER_06 6:28

Sure. Um, I would say patent eligibility is a threshold for getting a patent. Uh the subject matter of the invention has to be eligible for a patent. And there are four broad categories of subject matter that are eligible for a patent processes, machines, manufacturers, and compositions of matter. Um, as well. And also the Supreme created three broad judicial exceptions over many years. Uh and these are national phenomena, laws of nature, and abstract ideas. These uh these subject matters aren't eligible for a patent. Uh, the invention also has to be new or novel. In essence, an invention can't be patented if it's certain public disclosures of the inventions have uh have already been made. If the public already knows about the invention and the inventor is not adding to the public knowledge base, then a patent won't be granted. Uh an invention also needs to be non-obvious. Um if it's not exactly the same or something that's already in the public domain, something the public already knows about, then which is referred to as a priority, then it's considered novel. But for an invention to be patentable, the invention also must be a non-obvious improvement over the prior art. Um, in simple terms, this means the invention sought to be patented would not have been obvious to a person of ordinary skill in the art, uh, to which a claimed invention pertains. And that's sort of a, it's a, it's a, that's a difficult or uh a determination that's also made when granting a patent or not. Um, in other words, invention is compared to the prior art uh and a determinist determination is made, uh, whether the differences in the new invention would have been obvious. Um lastly, a claimed invention also has to be useful in order for a patent to be granted on it. Uh, this means that the invention should have a use, must have a useful purpose and some practical utility.

SPEAKER_01 8:33

And I think I would just add that, you know, for for the listeners of the podcast that may be uh neophytes when it comes to patents, it's really, as Rama said, and anything that's new, useful, and not obvious. And AI is bringing a host of questions to sort of where those lines will be drawn, especially around what is obvious, right? If if we have this uh supreme knowledge that is capable of being bestowed by an AI, uh, how does that affect the law of invention in that sense of what is inventive or what is not obvious? And also how do you disclose those things, right? What does it take to actually teach the public how to use your patented invention when it comes to some new inventive uh machine, let's say, that you want to seek patent for? But I think uh, as Rama said, that's a pretty thorough and comprehensive overview of the sort of bare bones requirements for inventions.

SPEAKER_02 9:23

Can I just ask a couple of clarifications on that that may be helpful? How far-fetched can the usefulness be in the future? Because some things that people propose as AI inventions, they may not have reduced to actual practice when they submit the patent application. And what if they're thinking, well, when and if the technology arrives, this will be useful? Does the patent office make that kind of a determination as to how far out it has to be before it can be useful?

SPEAKER_01 9:51

Usefulness gets a little tricky. So I think what you're really getting to is with that concept of like, like sometimes you'll see these uh patents that are applied for that are, you know, claiming teleportation or or some you know brain interface or something like this that is probably more in the realm of science fiction. And that has to do with this uh, you know, the classical distinction with if you were going to need to, before we were in a uh a first-to-file system, we were in a first to invent system in the US. So that meant that you had to prove to the office that you actually either had possession of your working invention or you had constructively reduced it to practice. And that meant you could write it down and explain it to someone and it would, you know, hold up under scrutiny. So in in those cases, really that's that's more of a question of are you actually in possession of your uh, you know, can you written, can you write out your invention and adequately describe it to the patent examiner that's looking at it and convince that person that you are in fact, at the time of filing now in possession of that invention. So uh as far as usefulness, the agency doesn't really police utility so much. There's a line of cases that goes back uh uh in in um decades ago that sort of scrutinizes pharmaceuticals and things of that nature. You know, think about the whole realm of uh, you know, any any sort of material that's out there, and you can start to think, okay, well, is the PTO morality police? Do we have to, where do we draw the line when it comes to what is useful, right? So we we try not to play that game.

SPEAKER_02 11:21

Yeah, what prompted me to ask that was that the US Department of Commerce just recently requested comments on the imposition of export controls for certain brain computer interfaces. So somebody coming up with and and they explicitly say that includes AI, which sounds pretty far-fetched to me, but obviously for somebody skilled in the field, it's not at all.

SPEAKER_05 11:45

So hearkening back for a moment to your comments about one skilled in the art, Rama, uh when when I was a young engineer early in my career uh and was filing one of my first patents or working on it, uh, the patent attorney educated me by explaining that uh that that the disclosure had to be um sufficiently detailed and complete so that um my friend Art. Okay, Art is this wonderful engineer who can build anything given the instructions on how it works, but has not a single inventive or creative bone in his body. So is so basically the stuff that goes into the patent disclosure has to be enough for art to be able to recreate the invention. Um, that was sort of uh that that the that that fellow art has really sort of uh loomed large in my mind when I when I think about patents.

SPEAKER_06 12:48

Yeah, the person of ordinary skills a hypothetical legal uh person. Um and you know, part of the school's requirements is that you have to teach someone how to be able to uh make this stuff or do this stuff.

SPEAKER_05 13:01

And one of the interesting things about AI, which really is, I think comes to the crux of why this question is hard for us, is that one and one of the things we know about AIs is they can produce, let's say, solutions, answers, whatever it is, predictions that are demonstrably superior to anything else we can do. And yet we can't really quite explain how they did it. And you know, so in terms of telling art how to do this, other than say, well, write the same software as I used and give it the same data I gave it, and now you've got it. That somehow seems very unsatisfying.

SPEAKER_06 13:43

I mean, that's that's a great point, right? I mean, you can't just say apply AI and my invention is applying AI and coming out with this outcome, right? You have to sort of again teach, teach how to how to make that AI or how to use that AI. Um, so yeah, I think those are issues we're gonna be grappling with. I think those it those are issues the patent office is gonna be grappling with is what is sufficient disclosure.

SPEAKER_02 14:07

And it's true also, isn't it, that many of the uh inventions offered in the area of AI don't even mention the word artificial intelligence in the disclosure. They refer to the electronics, they alert refer to the algorithms and software, but they don't use the words artificial intelligence. And maybe that's because the Supreme Court has been so negative over the last dozen years or so uh in narrowing what can be patented in the AI area, if I'm not mistaken.

SPEAKER_01 14:37

Right. I think you're hitting on a broader point here that has sort of always been associated with software when software and business methods to that end, sort of to get the attention, and this goes back to what Raman was mentioning about the judicial exceptions to what is eligible for patent. Because if you have plain as day in section 101 of the Patent Act that it says, you know, machine, let's say, could actually be a qualifying invention, then okay, well, if a software runs on a computer and I claim that computer is part of the software, what's the problem, right? Well, the Supreme Court told us there's a problem because certain software is too abstract. And of course, extend that over to AI, and that's I think something we'll get to later. But that's sort of what we've heard from the public is that AI, if you can call it that, right? Whatever that even means, uh, because there's a whole uh pregnant definition question in there. But if you can figure out what you mean when you say AI, it's really to some extent, it's just an extension of the software uh issues that the patent system has been facing for decades now.

SPEAKER_02 15:38

Well, let's look at another sort of public assumption. It's a common assumption that a patent provides a patent owner with the right to practice and commercialize the invention covered by the patent. Rama, could you explain what rights a US patent actually grants the patent owner and for how long?

SPEAKER_06 15:56

Sure. That's not actually a precise assumption. Um a patent is a type of intellectual property that gives its owner the legal right to exclude others from making, using, or selling an invention for a limited uh period of years. Um that time in the US, a patent provides 20 years of protection from the earliest filing date of the patent application. So a patent owner has the right to decide who may or may not use the patented invention for the period. Uh, but in so in other words, patent protection means that the invention can't be commercially made, used, distributed, imported, or uh sold by others without the patent owner's consent. But it is important to know that you know, a patent does not grant the patent owner any right to make their own invention. Um, that's going to be determined by other things like other people's patents that might not be able or that allow that might not allow you to uh make or use your patent. It also is going to depend upon um uh regulations, antitrust laws, FDA laws. Um that may restrict the ways in which a patent older patent holder can uh use his or her invention.

SPEAKER_02 17:12

So, for example, if I patent a new kind of lens for a camera, that doesn't mean I can put it on any camera that I want because there may be patents on the cameras that I have to get license permission to use.

SPEAKER_06 17:25

That's right, but you could exclude others from practicing that invention.

SPEAKER_02 17:30

Right. Um, picking up on this theme of rights, Chris, what is the basic bargain between the government, the US government in this instance, and the inventor when the US PTO grants the inventor a patent?

SPEAKER_01 17:43

Sure. Yeah, no, I think that's the sort of the classic quid pro quo of the patent law is that in exchange for the inventor's disclosure of precisely how to make and use that invention, in return, what what you're granted by the government is a time-limited, government-backed ability to actually exclude others from making, using, et cetera, uh your inventions. And so I think if there's any any new tension today, at least beyond those those we just spoke of on your business methods, software, and all that sort of thing. I think one of the main issues now with AI is sort of, well, what does that even mean to be an inventor anymore, right? There's a lot of uh cases the world over right now that are grappling with this question of well, what does it mean to invent? So if a machine creates something, uh, is does that satisfy that classical notion of, you know, my inventive mind giving something to the public in exchange for time-limited rights? And I think it's a mistake to try to personify AI in that regard, because I think once you start to mistakenly personify AI, you start to think of it as maybe what we've always thought of as inventors, you know, you and I, human beings, right, as inventors. And I think if you actually dig deeper, at least at a snapshot of like right now, you can start to see that these AIs, whatever they are, quote unquote, are actually just tools that are facilitating very advanced, sophisticated, as I think Mark put it, uh, predictions really. And we're we're very astute to leverage those predictions and have all kinds of neat things come out of them. But the second that you take it a bridge too far, in my opinion, that that's when you start to call AI a an inventor, maybe, and and what what are the ramifications of doing that? Then maybe we're outside of this classical quid pro quo, and maybe that makes the metaphor uh inapplicable today.

SPEAKER_02 19:33

And if I'm right, the the patent act defines inventor as an individual, and we don't normally refer to AI as an individual.

SPEAKER_01 19:42

Well, we're we're fortunate in the US that we have uh some strong statutory hooks, I think, to hang those types of arguments on in other jurisdictions less so.

SPEAKER_02 19:51

Well, let me follow up on the point you were making. Back in August 2019, the USPTO issued a request for public comments to help it gather information on patent-related issues regarding AI inventions. Just a month earlier, two patent applications were filed with USPTO that identified an AI system called the device for the autonomous bootstrapping of unified sentience, or as the acronym calls it, Dalvis, as the inventor. Chris, did the Dalvis application play any role in USPTO seeking comments shortly thereafter?

SPEAKER_01 20:28

Yeah, it's it's a great question, Roland. Thanks. And you know, obviously, I'm not able to talk about the internal deliberations of the USPTO here, but you know, I'll just say that for decades, the USPTO has been examining and granting patents on on what many would consider fields of artificial intelligence. And so I think that that point is is quickly lost that we have this, you know, really strong examining uh patent examiner workforce that has is very adept and astute at examining these sophisticated technologies. They've been doing it for a long time. And I think to your question, though, really on the policy side of the house is you know where I sit. And we we always, of course, try to keep abreast of these uh emerging technology issues. Obviously, the technology is not going to wait for us to figure out how to catch up to it. So we had been studying these issues for quite quite a time. And you know, our first, I think the really modern era data point would go back to January 2019 when we held a uh a roundtable uh number of panels covering all of the realm of IP. We we dipped into some international perspectives, we did an economics panel, and so I think that's really the culmination of those latter um requests for comments that we saw put out. It sort of can go back to those stakeholder comments that we heard at those particular roundtables back in January of 2019. And so the the other work sort of flowed from that. And just to put that in perspective of Davis, I think that team, at least in the US, they first filed their applications around summer of 2019. So we'd already Been sort of looking at these issues long before, and I think maybe if anything, uh perhaps uh we might have stoked that inventorship issue at that time, and and then you saw this flux of uh the Davis cases around the world thereafter.

SPEAKER_04 22:14

Yeah, this is Charles. Um uh the the other nerd. So I guess what what really piqued my interest is when you said prediction, because uh one of our other uh interviewees talks about prediction machines. Uh AI is they're just prediction machines. The better the prediction, the the better the utility. And I wonder can you think either of you think of any patents in the past that it solely patented a process for coming up with a prediction? And and if so, how would it be different if an AI is predicting what compound is going to be the best uh you know to maintain the virility of a vaccine or or whatever?

SPEAKER_06 23:04

So I think that gets back to what Chris was mentioning earlier, that AI is really just an extension of some of the software algorithms that we've had in the past. I mean, I can't think of a specific uh software algorithm that used statistics for example to come up with predictions. But I think this is just a sophisticated way of doing that. So, but it does, you know, it does raise unique issues that we're possible, you know, probably didn't have with our more simplistic algorithms.

SPEAKER_01 23:32

And and I'll just add, um, you know, that that nod to prediction machines, I think, is as apt as because that's definitely one of the books that I read in my pursuit of trying to get up to speed on when AI. That's an excellent book. I'd recommended prediction machines for sort of the economic side of analysis. And I'll look forward to that future uh episode of this podcast. But yeah, you know, there are many, um, you know, we've got over 10 million issued patents today, right? And that's not even including some of the very beginning historical patents, if you will, the ex patents and all that. So a lot of patents that I'm sure there's a few that are actually processes or methods for um generating predictions, but all of those will necessarily be a matter of law, they will be different and and novel and bring something to the field with which they were um there to define over the prior art as that term, that's everything that came before that particular um process or method for making that prediction. And so if you can think about how how many different ways there could be to paint a house, right? There could be that many different ways to predict how someone would pick a color to paint a house, right? So it's uh there's a vast uh spectrum, I think, for coming up with this inventiveness. And uh, you know, naming a precise patent number, you know, I can't do that, but I think that there's, yeah, there's definitely some that are out there. And that that gets into a very interesting question of liability for for infringing these particular patents that uh we may or may not touch on.

SPEAKER_05 24:58

So many of the public comments that USPTO received indicated that no major changes were necessary in IP law to address issues raised by AI technologies. At a high level, could you summarize the key issues or themes that USPTO laid out in its October 2020 report following the comment period? Ronald?

SPEAKER_01 25:25

Sure, maybe I can I can get us started and then Robert can uh can correct me. Um so I think one of the the really overarching themes that we we took away was the one that I made a nod to earlier, is that is you know, you didn't define what AI is. So if we're gonna talk about it with any sort of concrete, you know, uniformity, we should probably at least define what it is that we're talking about. And so I, you know, I'm very sympathetic to that argument because obviously there's there's so many different uh you know tentacles flowing from from the AI body there that you you really have to pin down what it is that you're trying to talk about to get there with any, especially when you're making policy to make sure that good policy is being made. Um, another major theme was the this question of where are we in the state of the art of artificial intelligence, right? You know, everyone can can break it out into these two camps of you know, sort of the artificial general intelligence versus the more specific intelligence, narrow AI. So um, for listeners that may not be familiar, AGI is really just this the singularity moment when computers know how to think and conceive of anything and they'll create the problem and then solve it. It's not us giving the machine a problem and figuring out what it predicts the solution could be. Um so that was uh a general agreement, I think we saw in the comments that we're not at that singularity moment quite yet. Uh, we're still sort of in this uh more narrow AI domain. And that's really important to think about, right? Because uh some some people will advance things to the contrary at times. Uh a third major takeaway that's that's in our report is that when it when it came down to you know what is the level of uh human engagement, there was this general agreement that we as human beings are are necessarily uh required to operate many of the sophisticated algorithms that are that are out there. And that I think flows from that understanding that we're not at the AGI stage just quite yet. And in most importantly, I think for the PTO, one of our major concluding points was that the on balance, the majority of those that that weighed in and sent us comments were really that the the current architecture of our IP system was appropriately calibrated to handle the you know near term of of sort of IP innovations that relate to AI. So the one question where things got a little bit more split was when you think about sort of the fuel of AI, if we can term it that, right? The data. So there's some out there that want to seek more protections for data, whatever, again, define that, whatever you're talking about there, because that's a loaded term that has many different meanings in many different policy contexts. But then if you could do that, uh, you know, maybe you need rights for data that are more robust than than they currently are available under copyright law, let's say, or you know, even through trade secret protection. But Rama, I'm sure I'm forgetting something.

SPEAKER_06 28:23

I know I think I think that's just right. I think that, you know, overall the public comments were the house isn't falling down, thankfully. But that we should look out on the horizon and see where we're going and what changes we might need to make in IP policy. Um, the technology is evolving at a very rapid pace, uh, uh faster than prior technologies. So it is something that we need to track and make sure that our IP regimes are on point to address the issues that are raised by these technologies. Um, and as uh Chris pointed out, one of these is data. Data has always been important to us, but with AI and emerging technology, it's becoming even more critical. So uh that is one area that I think we need to track and uh make sure that uh we're in good shape with our IP laws and policies to address those issues.

SPEAKER_05 29:14

So one of the core questions uh under discussion is whether AI may be identified as the inventor in a patent application or the author in a copyright application. Where did the USPTO come out in the DABAS issue and why?

SPEAKER_01 29:30

Sure, I can I can uh lead us off on that one. So I think uh the way that the USPTO looked at this, and you know, this is all publicly available on our our website. So I'll just make a plug for that real quick. If you go to the main USPTO.gov site, you'll see links to the AI corner of our website, and then you click there and you'll find all these reports we're talking about. You'll find um some relevant uh legal decisions the offices have made. And that was really um how we handled this Davis case that was at the USPTO. Um in uh, I believe it was April 2020, we actually issued a decision in the application to say, look, you have not completed perfectly your application. You have left off this uh critical component of who was the inventor of the human being that is the inventor. Uh you have here listed Davis, and that's not a person. And so uh rather than uh you know modify the the naming of a human being, uh they sort of pressed the issue with the agency, and the agency issued a final petition decision. Uh, and that petition decision really relies on the you know the statutory constructs that we operate under at the USPTO. Uh, you know, section 101 says whoever invents, right? So we said look at whoever. That's that's probably a person, right? Whoever um referring to a person, we see individual, right? As was mentioned in section uh, I think it was 115 that lists that as well. So uh replete throughout the statutory schema for for invention, you're gonna find these references that you can make strong arguments, I think, that this is a requirement that Congress has put there. So the flip side, of course, is well, Congress didn't know about AI when they they drafted these laws. So, okay, they wanted to promote innovation, and that's what you need to interpret those words to consider AI, again, whatever that means, falling under underneath of those sort of uh classical human being type uh pronoun inferring words. Um, so that's ultimately how we handled it. We we looked at the statutory schema, we made our decision that this was really uh a question of humanity, and we said uh this DABAS is not an inventor, and therefore, you know, please uh either correct that or are you over on this. And so uh ultimately they appealed that decision to the Eastern District of Virginia, which is uh uh maybe more infrequent path chosen for appeals from the US PTO. Usually they will go to the Court of Appeals for the Federal Circuit, but there uh we already have our decision uh from from that court, and that court basically sided with the PTO and said, yeah, I agree, they did a pretty thorough statutory analysis here, and it's it's reasoned and based in the law. So yeah, we agree with the PTO. Davis, you can't, you know, this is this is too great of a policy change. This would need to come from Congress type of thing if we're gonna start to reinterpret this, and and that's where we are today, in the US at least. Ama?

SPEAKER_00 32:26

So so Chris, we've we've been hearing from some of the users of AI, you know, that to think of AI as a tool, right? It's a sophisticated innovative tool. So putting that kind of construct in the framework of the Davis decision, you know, is one way for our listeners to think about this decision. Um, that AI is a sophisticated tool that's used to create and innovated. But you know, as we sit here today, noting that things can change and emerge, you know, the use of that tool is at the direction of natural persons, right? Um, the inventors. And that's sort of the framework or the lens with which a lay person could kind of understand what this decision means.

SPEAKER_01 33:08

Yeah, absolutely, absolutely. Um I think uh one way to think about it right by way of analogy, at least, is right, there's CERN has their uh large hadron collider and they do all sorts of sophisticated experiments, and and one day they exploded some things together and they found that there's a particle, the Higgs boson existed. Now no one sat down and was like, the machine made that particle. It was like, no, we use the machine as a tool to find that. And maybe if that's if that's too uh, you know, quantum physics, then maybe like the electron microscope, right? Like if I can see things and discover things through a tool that I otherwise could never have done, nobody thinks that the electron microscope should get the pattern rights or whatever, right? It's the it's the human that's behind that tool that ultimately appreciates it. And you see this a lot, I think, in a couple of use cases that have come up, uh, right? Obviously, the pharmaceutical industry using these um sort of like AI search algorithms to sort of figure out which combinations of materials may yield a viable drug component or even a material science. Same thing, looking at elemental structures of compounds and how will they interact to form you know new and novel materials. And rather than having to you know brute force all of those combinations, you just sort of get a top list of candidates from the machine, and then the you know, human beings go in and they try to verify which is best, or if maybe I got a recommendation from that tool, but it wasn't really a viable candidate because of you know some reason that came up in that practical application of sort of dividing that, whatever it was, a compound for a therapeutic uh drug or or for you know a new type of you know cement or something, right? So um I do think that's a really useful way to think about this.

SPEAKER_02 34:53

Without going into USPTO's internal deliberations too much, did it also consider the fact that the Constitution grants Congress the power to create, you know, to create laws concerning patents and copyrights because it wants to create incentives for the creation of inventions, the creation of works of authorship. And it's not clear that a machine would be receive any incentive whatsoever if granted a patent.

SPEAKER_01 35:22

Yeah, I think I don't recall uh the text of the decision from Judge Brinkema that it was the Eastern District of Virginia case, but I do I do remember maybe from the oral argument, I think uh I'm remembering that this this came up, right? So obviously Article 1, clause 8, section 8 of US Constitution, you know, we're we're unique in in the world that we have this innovative system shrined in our constitution. So yeah, I think that's that's one way that uh the judiciary at least has thought about it is that yes, uh Congress is in charge of making these policy determinations. If for some reason, you know, a narrow, strict understanding that an individual can only be a human being, is that going to you know confound the framers' intention here? If if so, then maybe Congress needs to go and look at what we've done in our interpreting that law that's on the books today, because maybe it needs to change.

SPEAKER_02 36:14

Well, comparing Congress to other countries, uh I note that a number of countries, South Africa and Australia, have uh addressed the issue of whether Dallas can be the inventor of a patent application. I'm gonna leave South Africa out of the conversation because they don't really examine patents. An application is submitted and they just approve it. So that wasn't really a deliberative review, but a recent Australian court decision uh said you can identify under the Australian patent statute uh an inventor as an AI machine. Chris, uh Rama, what do you think are some of the key differences between these countries' laws and US PTO's consideration of the issue of who and what can be an inventor?

SPEAKER_01 37:00

Yeah, Rama, you want to start us off?

SPEAKER_06 37:02

Sure. I'm just gonna reiterate what you said under our US statutes and regulations. You know, I think the question before uh the PTO and the uh Eastern District of Virginia was whether an AI can be named as an inventor. And under the current doctrines, I think that you know that question was answered that it has to be an individual human being. Um, but you know, whether that's what you know we should have going forward for policy reasons like incentives and uh you know, though I think those are questions for Congress to address.

SPEAKER_01 37:37

And and Roland, the uh, you know, the Australian and South African case, I think, you know, they're they're interesting cases because they're sort of, I think the the legal team on Team Davis, you know, these are uh I would, if I were them, I would view them as wins, I guess, um, uh from their legal team's perspective. And another thing, yeah, it's it's easier to dismiss the South Africa uh case because it was sort of like a formality almost that they got the rights granted there in South Africa. But apart, like if you take a step back, uh there's a thing called the Patent Cooperation Treaty, which is uh an international treaty that allows for uh single filing. And from that filing, uh all the uh the WIPO member states, right? You have the possibility of perfecting your application in the national jurisdiction. And so they actually got WIPO to publish the PCT application, naming the machine Davis as as the inventor on that um PCT application. So that's an interesting thing because that's sort of you know the implementer of the the UN is behind WIPO, right? And so you have them making that determination that this is okay, this and this can fly out the door and land in all these different jurisdictions. But, you know, like uh in agreeing with Rama here, I think really what happened in Australia is again, they're looking at their law, they're interpreting what their statutory patent law says. And the judge in that case seemed to find some some room in saying that really, you know, we haven't defined inventor. And so we this is a chance that we can really get out ahead of our international um, you know, colleagues and sort of take a take a stance on this that's contrary in a way to what we've seen already. The European Patent Office took a determination, the UK's intellectual property office took a particular termination, all citing in the same manner as the US PTO that you can't name a machine as an inventor under the those patent acts. And so obviously it's it's different patent laws in every country. There's some some sort of uh, I don't know, I won't call it harmony, but there's some uniformity in patent laws thanks to the uh WTO's TRIPS agreement, uh, which governs sort of a floor for intellectual property rights. But that, of course, that was back, I believe, in 1994, and AI was not on the topics of conversation at that time for sort of how are we going to grant patent rights? And so maybe we're at a stage where the international uh community needs to get back together and start to think about this if this is really, you know, a ripe issue to discuss. Uh or maybe we're not even there yet. Maybe we're so squarely in this uh narrow AI world and we don't have to fear the AGI singularity upon us quite yet. And maybe that's uh not right yet.

SPEAKER_02 40:12

Well, I I read the Australian judge, Judge Beech's uh judgment much the same way you did, that he he read their patent law as not precluding a machine rather than requiring an individual. And he was very policy-oriented. I should also note that when I checked recently, the Commissioner of Patents in Australia has taken the case up on appeal. So this is an open issue, and we shouldn't be treating it as you know, the Australian position. It's one judge in the federal court in Australia. Rama, you were detailed for some time from USPTO to assist the National Security Commission on AI in its work in developing a report on AI, which it published in March 2021. Could you tell us what the commission's objectives were and how the commission went about its work to the extent you're allowed to describe that?

SPEAKER_06 41:01

Sure, sure. Thanks for the question. Uh, Congress established the National Security Commission on AI in the 2019 National Defense Authorization Act. Um, it was an independent federal commission uh with a mandate through October of this year to consider the methods and means necessary to advance the development of AI, uh, machine learning, and associated technologies to comprehensively address U.S. national security. Uh, the commission was composed of 15 commissioners appointed by Congress and the executive branch with uh experience in industry, academia, and government. In uh November of 2019, we submitted our first interim report to Congress on the current state of AI for national security. And this report um affirmed that the United States is in a strategic competition with AI at the center and that the future of our national security and economy are at stake. So it called for new imagination, common purpose, and decisive action. And after that interim report, the commission decided to release recommendations on a rolling basis through throughout 2020. We delivered three memos, uh, which culminated in a final report that we issued on March 1st of this year. Uh, the commission worked in close collaboration with Congress, the White House, um, and executive departments and agencies to develop its uh assessments as well as its recommendations. We consulted a range of experts inside and outside of government on AI and its responsible adoption. And so the goal was to produce a comprehensive and um enduring national approach national approach to uh maintain its leadership, America's leadership and advantages in uh AI and other technologies. Um the report considers where the US stands on AI compared to its primary competitors and presents first steps the United States uh should take to defend, compete, and win in this new tech era.

SPEAKER_02 43:13

Can we put a little finer point on it? Can you talk about the impact on the commission's recommendations by its finding that I think it put in its executive summary that an adversary's potential use of AI, namely that defending against AI capable adversaries operating at machine speeds without employing AI, is an invitation to disaster?

SPEAKER_06 43:35

Yeah, the committee commission assessed that AI-enabled capabilities will be used as tools of first resort in this new era of conflict as our strategic competitors develop AI concepts and technologies for both military and nefarious use. There's also uh because uh there are cheap and commercially available AI applications ranging from uh, let's say, deep. Fakes to drones, it's also becoming available to rogue states, terrorists, and criminals. So I would say the biggest danger is automatic response systems. The nature of threats and the time to respond to the threat is getting shorter and shorter. And these capabilities, you know, they compress the time amount of time for human decision making. So in the military context, for example, let's say a machine tells you that an attack is imminent. If you don't press this button in five seconds, you're going to die. What do you do? How do you get a human decision making? How do you get human decision making involved into these systems? So, you know, we need to have agreements on, you know, maybe delayed responses. Um, you know, the automatic nature of these decisions needs to be prescribed. They need to be agreed upon that they're being, you know, that they're not a good thing and that humans should be involved in these decisions and that the response time should be long enough to incorporate principal decisions at least as a starting point. Um, and you know, I'll point out that in our final report, we said, you know, that uh, you know, China, Russia, and us should renounce automatic launching of nuclear weapons. And we shouldn't assume that there's one rule uh about human control or machine speed that should apply to all scenarios because um they raise you know different questions and different issues in different contexts. Um, it's important to determine where the types of errors machines make would be too consequential or might lead to escalation, and where humans are uh should be uh you know uh part of those decision-making systems.

SPEAKER_02 45:46

Well, I noticed also in the report that it doesn't just describe China as developing AI at a rapid rate, but it focuses on how China has recently made uh changes in its patent strategies domestically and globally, to position to try to become the global leader in AI. Could you describe some of the changes that China has made that the commission was responding to so this is more concrete for our listeners?

SPEAKER_06 46:13

Yeah, unlike the US, China has recognized the importance of IP policies as a critical tool within its broader national strategies for emerging technologies. Um until recently, um, it was in China's interest to have weak IP uh intellectual property rights to facilitate tech transfers, but China has changed directions. It's now focused on its tech capabilities and wants to strengthen its innovation systems as a key driver in its competitiveness. And one of the areas that it's focused on is intellectual property. Uh, China just issued guidelines to build a powerful IP rights country just last week. And this is the uh 14th uh five-year plan in IP, and it identifies goals that it wants to accomplish by 2025 with uh numerical benchmarks. Um, and so some of these include doubling the number of high-value patents per 10,000 people to 12, and uh it wants to more than double the number of foreign patents granted annually. Uh, it wants to have have a country in the val where the value of the patent-intensive industries account for 13% of its GDP and have annual trade in IP royalties of, I believe it's 350 billion yen. So uh while China is making it easier to acquire patent rights to inventions, uh, you know, I think the US courts have severely restricted patent protection for computer implemented and biotech-related inventions. And I think this could have domestic harms uh as far as incentives for RD, as well as international uh harms. Um, you know, the US is could potentially lose its IP leadership position and um allow China to attract innovation to its borders, um, as well as influence other countries to adopt its technologies.

SPEAKER_02 48:11

Didn't they also increase the penalties for patent infringement so that they're actually more severe than they are in the US?

SPEAKER_06 48:18

Um that's right. Um it it it's I think it's even working on increasing those uh patent infringement damages even more.

SPEAKER_05 48:25

So given given what you've described of China's patent strategy, could you uh explain the Commission's recommendations about what changes the US should make in its patent laws in order to remain competitive uh with China in the field of AI?

SPEAKER_06 48:42

Yeah, the the the commission recommended that the United States recognize intellectual policy, intellectual property policy as a national security priority, critical for preserving America's leadership in AI and emerging technologies. It didn't make specific recommendations on what that policy should be, but it did recommend that the US needs to visit its IP laws, regulations, and regimes to make sure that they are up to date to address technologies of today so that it can remain competitive. Um and the commission reckoned that the um uh that the Secretary of Commerce, in conjunction with the director of the USPTO, should be charged with developing these policies with in, of course, in coordination with the other departments and agencies in the executive branch, and that that should be elevated to the White House to decide which of these policy recommendations should be adopted into broader national security, um uh national security um policies. And it also recommended that there should, you know, it provided a non-exhaustive list of IP considerations that should be assessed as part of this policy formulation. We identified 10, um, although again it was a non-exhaustive list. And so some uh one example was patent eligibility. You know, the recommendation was to assess and articulate the impact of the current patent eligibility doctrine on innovation in AI and uh emerging technologies from an economic trade and national security policy perspective. Another one is IP protections for data, uh, an issue that Chris identified. Uh, the recommendation was to assess the need for additional protections for data. Um, and then also recommendations for combating IP theft to assess additional executive branch efforts to counter IP theft threats, including actions in collaboration with our allies and partners. Um, and another issue that we touched upon today, inventorship by AI to assess the need for policies relating to AI-generated inventions and creations. And lastly, one of the IP considerations that we identified was um the need to democratize innovation in our IP ecosystems, to expand the innovation base and uh you know make sure that everyone is uh has access to innovation in the IP ecosystems through streamlining guidance, for example, for startups or um and uh seeking uh for uh startups seeking IP protection.

SPEAKER_00 51:28

Rama, picking up on something you just were walking through on those recommendations, sort of circling back to a converse, the conversation we were having about um data with uh Christian there, um, you know, the commentators to the USPTO notice seemed to generally agree that data was generally adequately protected under existing laws, whether that be trade secrets, copyright, licensing. And as I reading through the commission's report, you know, there were specific statements or concerns expressed about whether we had the right legal protections for data. Um, and we've been hearing about how valuable data is in the context of AI innovation and technology, but the report seemed to put a national security lens on the protection of data. And it'd be great to sort of hear some more thoughts of about how you guys were considering the value and importance of data and protecting it for American strategic value here.

SPEAKER_06 52:25

Thanks, thanks, Emma. So, yeah, you know, I think as AI and emerging technologies evolve even further, data is gonna take on even a more critical role. Um, China, for example, has uh an advantage in that it has a bigger population and has access to more data than we do, as well as other countries. So I think how we can leverage data is gonna be very important to our tech competitiveness and innovation. So I, you know, I think data is just going to take on even a more important role. And that's why the recommendation to make sure that we are poised to be able to leverage, you know, leverage data and have proper incentives in place to make those data sets, whether they're organic or whether they're synthetic. Um, so uh, and by synthetic I mean made-up data sets. So just to make sure that looking on the horizon, that we do have the protections in place to make sure that we have the incentives to create the data sets that are that will be necessary.

SPEAKER_01 53:26

And I was just gonna I was gonna pick up on that point, is that you know, thinking about you know, your listening audience and sort of if they want to get into AI and like where should they focus their time and attention. So there's a there's a book out there by Kai Fu Lee that's called AI Superpowers, and it makes an interesting point relating to China and their data acquisition. And so, you know, through the WeChat app or you know, whatever other applications that their citizens use on a daily basis, if you want to make a dentist appointment or you want to eat dinner or you want to go see a show, all of that's through an application. And so I think uh the author there, Kai Fuyi, is making the point that China has this rich wealth of data already available to them. And so, you know, you can see it from sort of the the European lens with their GDPR and sort of putting limitations on data. And then you've got China that's sort of the wild west of data where they've got every sort of aspect of your daily life because they've collected a data point on on what you ate for dinner last Tuesday. And so, where do we as the US, where do we want to fall, right? And and how do we calibrate that so that we get it right? I think there's a Herculean task in defining what the data is going to be, and sort of uh only from there can you then get to that next question about okay, well, how do we protect what we mean when we say data?

SPEAKER_06 54:43

Yeah, and that's one of the, you know, those are recommendations that the commission made also that, yeah, sure, we need to, you know, go forward with AI, but we also need to do it responsibly. We're not gonna give up on our democratic values, our privacy, civil liberties, and civil rights. So while we do advance our tech capabilities, we do need to do it responsibly so that we protect our democratic values. So there's you know, there's two chapters dedicated to responsible AI in the final report. Um, so um uh yeah, those those are issues that we grappled with.

SPEAKER_02 55:17

Before we wrap up, I'd be interested to know since both of you have been discussing policy recommendations, what are each of you anticipating in the next year or two working on in the AI space with regards to pursuing those objectives?

SPEAKER_06 55:32

So, yeah, so um I'm at a new entity called the Special Competitive Studies Project, a nonprofit. Um, and I will be looking at issues related to impacts on society. Um, you know, how AI is going to shape our futures and our societies and our democracy and how we grapple with those issues, as well as intellectual policy issues.

SPEAKER_01 55:55

Yeah, I think from the USPTO perspective, I think what, you know, I would think that we're gonna sort of uh stay charted on really listening to what our stakeholders have to say, right? Are there really concrete issues out there that you know we need to sort of focus on and then how do we treat them? And we had some very exciting news the other day. We have our uh USPTO director nominee, Kathy Vidal, who's uh I understand has some some familiarity with AI, may have even uh worked in AI. So I'm sure we, if anything, you could expect more from the PTO. I would imagine if I'm reading the tea leaves correctly on AI, then less.

SPEAKER_05 56:30

Well, on that note, uh I think it's clear that that this conversation has opened up a broader horizon than I think we even realized when we started uh working on this session. And there's going to be a lot, uh, a lot for us and our listeners to learn uh over the next uh year or two as we interview people and uh explore this space. This is a huge space and a very exciting one. Uh, I'd like to thank you both for your commitment of time and effort and your thoughtful comments and and your sort of incredible uh insight into what's going on.

SPEAKER_02 57:13

I want to echo that. Thank you both for your contributions and taking time to talk with us today.

SPEAKER_05 57:19

We thank the business law section of the American Bar Association for their generous sponsorship of the production of this podcast. We welcome questions and comments from listeners. Send email to comments at mindthegapdialogues.com. We read all comments and questions and will try to respond in the letters section of a future episode. If you are writing about a particular episode, please do mention the specific episode number. Please also do include pronunciation tips to help us properly say your name when we reply in a subsequent episode.

SPEAKER_03 58:00

See you next time on Mind the Gap, Dialogues on AI.org slash Bizlaw. That's B-I-Z-L-A-W.