AI and the Future of Work: Artificial Intelligence in the Workplace, Business, Ethics, HR, and IT for AI Enthusiasts, Leaders and Academics

Robert Plotkin, popular author and expert in AI IP law, discusses regulation for LLMs and legal advice for entrepreneurs

• Robert Plotkin • Season 3 • Episode 26

Share episode

No field is being upended as much as the legal profession. We’re all confused about how content generated by AI will be protected under the law and many lawyers are also asking how relevant they’ll be in a world where large language models can pass the bar and do legal research.

Robert Plotkin is a luminary in the software patent space having been in the field for 25 years and having been involved in important IP cases related to everything from AI to quantum computing to autonomous vehicles and speech recognition.

Robert also published the book Genie in the Machine back in 2009 which amazingly foreshadowed the legal implications of AI on IP. Robert has lectured at the Boston University School of Law and received his undergrad in Computer Science and Engineering from MIT.

Listen and learn...

  1. How we should regulate LLMs... from an expert
  2. What entrepreneurs most often don't understand about IP law
  3. Who has the rights to the inputs to LLMs?
  4. Can work derived from LLMs be patented?
  5. Is AI-generated work subject to copyright laws?
  6. What surprised Bill Gates when he saw GPT-4
  7. Is there an AI winter up ahead?

References in this episode...

Speaker 1:

Yeah, you know I mean. One that might be relevant to large generative language models is speech recognition. I had a number of clients in that field for many years, going back a long way and including a company that you know had a much, much larger competitor in that space who you know they needed to defend themselves again. So it's a good example where, if you're a small company, you have to ask yourself what's to stop the big gorilla in our industry from just waiting until we innovate and then copying what we're doing for much pennies on the dollar, essentially.

Speaker 2:

Good morning, good afternoon or good evening, depending on where you're listening. Welcome back to AI and the Future of Work. Thanks again for making this one of the most downloaded podcasts about the Future of Work. If you enjoy what we do, please like, comment and share in your favorite podcast app, and we'll keep sharing amazing conversations like the one we have for today. As always, i'm your host, dan Turchin, ceo of PeopleRain, the AI platform for IT and HR employee service. I'm also an investor in, an advisor to more than 30 AI-first companies and a firm believer in the power of technology to make humans better. If you're passionate about changing the world with AI, or maybe just looking for your next adventure, let's talk. We learned from AI thought leaders weekly on the show And, as you know, the added bonus is you get one AI Fun Fact each week.

Speaker 2:

Today's Fun Fact Kyle Wiggers writes in TechCrunch that Harvey, which uses AI to answer legal questions, raised $5 million this past week and around led by open AI. According to founders, harvey's co-pilot for lawyers lets them describe the tasks they wish to accomplish in simple instructions and receive the generated result. The founders go out of their way to say Harvey will not replace lawyers, but it will help them be more productive. It's a crowded space that already includes competitors like case text and clarity. Ai, we know, will go quite far, and we also know that there are many new startups feverishly building and raising money for new vertical SaaS offerings that augment generic LLMs like chat, gpt. As always, we'll link to the full article in today's show notes and have a hearty discussion about not just Harvey, but other developments in the legal space. Now, shifting to this week's conversation, no field is being upended as much as the legal profession, and not just by Harvey. We're all confused about how content generated by AI will be protected under the law, and many lawyers are also asking how relevant they'll be in a world where large language models can pass the bar and do their own legal research.

Speaker 2:

Robert Plotkin is a luminary in the software patent space, having been in the field for 25 years and having been involved in important IP cases related to everything from AI to quantum computing, to autonomous vehicles and speech recognition. Robert's also published the book Genie and the Machine back in 2009, which amazingly, foreshadowed the legal implications of AI on IP. Robert is lectured at the Boston University School of Law and he received his undergrad in computer science and engineering at MIT. Without further ado, robert, it's my pleasure to welcome you to the podcast. It's started by having you share a little bit more about your background and how you got into this space.

Speaker 1:

Thanks so much for having me, dan, and I'm really glad to be on a podcast about the future of work, although a lot of what I work on and have written about is about patents and IP. So much of automation in AI has implications for the work that humans do, the skills that humans need in their work, and for how humans are going to need to learn to update their skills and collaborate with computers as part of their work, so really excited to be on this podcast. In particular, as you said, i'm a patent attorney. I've been specializing in obtaining, enforcing, licensing, selling software patents for over 25 years. What that means is my clients are tech companies developing innovative software and I obtain patent protection for them.

Speaker 1:

I've done a lot of work on AI and myself I'm a computer scientist by background and training, as you said, so I'm sure we'll talk more about what I've done in relation to AI and automating of inventing, but I've been immersed in this field for a long time. I started programming computers as a kid TRS-80 was my first computer back in the early 80s So this is just a passion for me. I love working on it, talking about it, working with clients and talking with people like you who are at the cutting edge of the field.

Speaker 2:

Robert, I knew there was a reason we'd get along so well. You and I both started programming on a TRS80.

Speaker 1:

We all know it was called the Trash 80. The Trash 80.

Speaker 2:

It's a lovely lake in a museum in Mountain View, an early Trash 80. It's not the only place you can see them anymore. So, gosh, you're booked to any of the machine. It just seems fresh in. I mentioned you published it back in 2009. But I'm more interested in if you were to publish a new version, an updated version, in 2023. What's the new chapter that you would add?

Speaker 1:

Probably not surprisingly, i would add a chapter about generative language models and chat, gpt. So the book was basically about ways in which computers were automating the process of inventing how computers are being used to generate new inventions. At the time it was things like circuits, airplane wings, even software, and I was really fascinated by this, both as a software developer myself and as a patent attorney whose job is to help get protections for inventions which traditionally have been required. Humans Could only be created by human ingenuity, and so when I saw this starting to be automated, it really fascinated me. And the reason I called the book the genie and the machine was that I used this metaphor at the time of computers as a type of genie where you could provide them with a wish, meaning a type of description of the result you wanted to obtain, let's say an airplane wing that was more aerodynamic, and then get out of it your granted wish, meaning a design, maybe a model of a wing that satisfied those requirements.

Speaker 1:

Now, at the time it really wasn't possible to write what I called wishes in plain natural language. You had to write those wishes in very technical mathematical forms. They might be an objective function for an optimization process or a fitness function for an evolutionary algorithm. These are still very highly technical forms, but what generative language models and what the general public has seen through chat GPD has done is to just significantly increase the ability of computers to act like genies in a way where you can provide natural language, plain English wishes to a computer and get plain, natural language output out. So this is a natural extension of that genie metaphor I used in the book. It's very exciting to see it come to fruition in that way, and so that's definitely the update I would provide now as to talk about generative language models and how they're promoting innovation.

Speaker 2:

Usually, robert, when I talk to an IP attorney, it costs me $800 an hour, so I'm going to use my bully pulpit to ask you a question. What is it that you would say most tech entrepreneurs don't understand about how IP law works?

Speaker 1:

Yeah, i have found there is a very widespread not universal, but widespread belief amongst a lot of tech entrepreneurs that patents stifle innovation and actually go so far as to be a belief that patents are just evil, that they stifle innovation, they shouldn't be allowed at all, and that companies that obtain patents are obtaining them to block competitors, And so I think that is a real misunderstanding. Yes, that can happen sometimes, but in my experience, in the vast majority of cases, people and companies that seek patents are not doing it to stifle innovation. They're doing it to protect their own investment in innovation, and in most cases, patents promote innovation. So an analogy is patents are a kind of property right And they give you a kind of property in the inventions that you create. So, by analogy, if you own a home, you have a property right in the home And that means you have the legal right to stop people from walking in your front door without your permission right. Most people don't see that as a bad thing, even though in a sense, you could say you are inhibiting the freedom of other people to walk into your house and to make use of it.

Speaker 1:

Most of us understand that that kind of property right by imposing limits on what other people can do with your property gives you the freedom to make use of it in the ways that you see fit.

Speaker 1:

And so you know if, in general, if innovators, inventors, innovative companies didn't have the any ability to use patents or other kinds of intellectual property to stop competitors from copying their inventions, then they wouldn't be able to obtain the return on the investment they put in initially. And that's why we have patents, to enable any of your listeners who have invented anything. No, it takes time, effort, experimentation, money all of that to develop anything new, and that often then, once you release that invention to the public, it can be much cheaper for someone to copy it than it was for you to create it in the first place. And that's why patents and copyrights and other types of IP exist in the first place not to stifle innovation, but to enable the initial innovator to get some reasonable expectation that they could get a return on their investment. So that's the main thing I would like people to understand. And clients who come to us, they come generally for defensive purposes, not for offensive purposes, to stop other people from doing what they're doing.

Speaker 2:

I mentioned in the intro some of the key technologies where you've had a hand in developing some core IP or helping entrepreneurs protect their IP. I mentioned AI, quantum computing, other spaces. Maybe can you pick an example of maybe how you've partnered with an entrepreneur in a space that has become very important and how you thought about the challenge of protecting their IP.

Speaker 1:

Yeah, you know I mean one that might be relevant to large generative language models is speech recognition.

Speaker 1:

I had a number of clients in that field for many years, going back a long way and including a company that had a much, much larger competitor in that space who they needed to defend themselves again.

Speaker 1:

So it's a good example where, if you're a small company, you have to ask yourself what's the stop the big gorilla in our industry from just waiting until we innovate and then copying what we're doing for pennies on the dollar, essentially, and then, given their huge marketing budget and sales force and distribution channels, just knocking us off?

Speaker 1:

Well, patents are the answer. So, with a lot of companies that are small and growing, what I do is we look at the competitive landscape, we look at what are the core aspects of your technology that are really innovative, that you really need to be able to protect in order to stop someone like that big competitor from competing with you, and then we focus the patent strategy on those core innovations, and that addresses what's often a concern of small companies, which is you mentioned the $800 an hour, the potentially very high cost of obtaining patents. Well, you don't have to patent everything If you do a thorough analysis and focus on the most essential inventions, the ones that give you the biggest competitive advantage. You can focus your patent budget and time there and not necessarily patent everything, but still get enough protection to get the protection that you need.

Speaker 2:

These days I'm doing about eight executive briefings a week for leadership teams that are trying to figure out what their generative AI strategy is, and among the topics we discussed the social and cultural implications And we certainly discussed the legal implications. There's a very high profile example in Samsung where their engineers were inadvertently feeding training data that was proprietary back to open AI. That's a no-no And that frankly throws a lot of organizations realizing that there are implications that they don't fully understand. I would love to just start with an open-ended question What do and don't we know at this point about how the law is or potentially may treat generative AI?

Speaker 1:

A lot of questions in there. The one that you just mentioned is who has the rights to the input to generative AI? That is, for the most part, under the law, a contract question, so read the end user license agreement carefully of any AI tool you're using, such as chat GPT. I believe that, generally, open AI has made it clear that this is a publicly available, mostly free, tool. You can pay for a higher level of service from it, but for the most part, you should assume, when you're using a tool like that, that any input you provide into it is going to be accessible to. Open AI is not going to be proprietary to you And that, as you said, it seems like not everyone at Samsung was aware of that or attuned to it. And so already, though there are responses to this, there's going to be market responses to this companies providing tools where maybe the technology isn't much different from chat GPT, but the end user license agreement is different, because corporate customers and individual customers will be willing to pay for a product where, according to the end user license agreement, the vendor agrees to maintain the confidentiality of the customer's input or agrees not to make use of the input in training models, and I know that, as in the legal world, i've seen lots of AI tools that have been marketed to my law firm over the years And the first thing we always ask is, if we provide input into it, can you use that to train your model? And one reason for that is very often probably your listeners are sophisticated enough to know.

Speaker 1:

Often AI models do what's called memorizing the training data, which means they might just spit that input right back out to another customer. I barely touched on the question, but I'll just list a couple of the other topics without going into them. Who owns the output of something like a mid journey or GPT based images? What rights do the owners of the training data have to the outputs? I know there's a lot of controversy going on now with images because of image generators using copyrighted artwork and spitting out output. That is pretty similar in some cases to the copyrighted training data, and I suspect where that's going to go and how it's going to get resolved is probably by a bunch of licensing deals, just like in the music industry. That's how things have panned out, but who knows, lots to be determined there, and there might be some copyright infringement lawsuits by copyright owners against companies like OpenAI before things settle and the dust settles.

Speaker 2:

So you hit on one of the most important topics. Exactly what you said. Can I patent work that was derived from an LLM, a large language model And look, we're six plus months into this experiment. Gpt and other AI models have been around a long time, but since the commercialization of chat, gpt, just the public's imagination has just gone wild in a way. That's exciting. What do we know so far? Have any cases been decided that maybe would be good coaching for listeners?

Speaker 1:

Yeah, so I mentioned the copyright side of things. On the patent side, i think it's unlikely that we're going to see patentable inventions coming out of chat GPT directly. By that I mean someone just asking a question to chat GPT and getting an answer out which on its own is a patentable invention. I don't think chat GPT is likely to do that. You notice I'm sort of couching my answer because I suppose with a sophisticated enough prompt maybe you could get a description of a patentable invention out. But more likely, where we're going to see patentable inventions and this is what I focused on in the book is that people are going to have to use probably non-language model AI. I described a lot of use of evolutionary algorithms to do things like take an objective function, meaning a high level description of sort of requirements that you want. Again, i'll use that airplane wing some sort of description of efficiency of airflow over the wing, let's say and then have a piece of software that can iterate over possible designs for that wing and then simulate them, evaluate them and then go back and iterate. OK, so people have been doing that for a long time now and then you get a design out at the end that may be patentable on its own or you get a few possibilities which at least then can be evaluated and refined and finalized by a human. So the AI might not sort of fully invent, but it might make the inventive process a lot faster, simpler and easier than it would be if you had to come up with all those possible designs on your own, build physical prototypes of them, test all of them.

Speaker 1:

I mean this has been being done in drug discovery already. The CEO Moderna pretty famously said that they may not have ever been able to develop their COVID vaccine without AI and if they did, it would have taken them much, much longer than it did. So there are lots of inventions like that have been patented already. Recently there's been cases in which someone has tried to get their AI software named as the sole inventor on a patent. interestingly, those cases have all failed so far. But the broader question of what are the rules that should apply to determine whether an invention that was generated using the assistance of AI should be patented, that's a very open, evolving question. I talk about a lot in the book and I'm glad to talk about that more in detail.

Speaker 2:

Robert, does your answer change if we expand it to include trademarks and copyrights on, let's say, brand assets that were enhanced or inspired by, maybe, a text image generator, like a mid journey or a Dolly too?

Speaker 1:

Yeah, so I'm going to be speaking a little bit outside my area of expertise because I'm a patent lawyer. I don't handle trademark or copyright too much, But there's two things about those areas of law that are different than patent which I think would make I think it's more likely we'll be able to see AI generated works be susceptible to trademark and copyright protection. One is that we don't have the same sort of inventorship requirements in trademark and copyright law. You can have a corporation that's just the owner and applicant for a trademark, At least under US.

Speaker 1:

Patent law still puts the human inventor on somewhat of a pedestal and requires the inventor, the human inventor, to have engaged in some sort of conception of the invention. So there's some hurdles there to patentability that may be hard to satisfy that don't apply necessarily in the trademark or copyright context. The other thing is that you don't have the requirements for what I'll call inventiveness in trademark or copyright law. You do need to have a work, needs to be what's called original and copyright law, And a trademark does need to be different enough from previous trademarks to be satisfied to qualify for trademark protection. But as long as it does, it doesn't really matter how it was created. So I think that you could probably have AI generated trademarks and not really have much of a problem with those gaining trademark protection.

Speaker 2:

So you mentioned that there have been attempts made to have the AI be the owner of a patent And I know I've read some interesting proceedings and I think that's a conversation we'll continue to have. But here's a thought experiment. As these AI assisted devices, maybe they get implanted in humans or the ways they extend humans may cause us to question At what point is it a human? At what point is it a machine? Is it a cyborg? Even though right now it's become somewhat easy to say there needs to be a human inventor named in the patent, are we ever back having a conversation about what's a human? Are we granting the patent to the half human, the human side of the cyborg? Would love to get your thoughts on that.

Speaker 1:

Yeah, it's a fascinating question. There's someone named Andy Clark who I know was already writing a lot about this before I wrote my book. He wrote this book called Natural Born Cyborgs. I'd recommend it to everybody.

Speaker 1:

From my main memory of his insights, whether we consider someone a cyborg or we tend to think of cyborg as a cyborg from the DC Comics, a person with physical robotic type implants in them His point is well, really, why does the physical part of it matter? If you think about how you as a human interact with Google, google is external to you, but doesn't it play the functional equivalent role of your memory? If the nature of the interaction between you and something like Google is very much like your interaction between you and your own internal memory, doesn't the you plus Google system really qualify as a cyborg, even if Google isn't inside of your body? Conversely, you could have an attachment on you physically. that's just an add-on that doesn't integrate with your being so much. Then you might say well, that doesn't really make you a cyborg.

Speaker 1:

This question of what does and doesn't make us a cyborg is really very interesting. I'd recommend you to him for going into that topic in a lot more detail. But chatGPT is a great example where the fact that it's a chat bot and you can engage in a back and forth that helps you with your own flow of thought, you could say, is a very cyborg, like it's more of a brain extender in a way. So I don't know the answers to any of these questions, but all of the latest technologies do really raise these questions in ways that we have to address and we can't avoid anymore.

Speaker 2:

I chose the term cyborg because I think it's a little bit more definitive and it conjures up, like you said, some notions of the physical portion that's the bot. But I agree with you when the reasoning is significantly influenced by a machine, it's casually say your colleague is a bot, but at some point you are the bot, or the fusion of you and the bot.

Speaker 1:

And.

Speaker 2:

I wonder how long is it before an AI runs for political office And how long before the one who's granted the degree? when you go to college or something, do you have to give credit to the AI coach that got you through the degree? Prognosticate for us. So these conversations that are science fiction or are we looking at? in the next three to five years, we're going to actually have some case law that answers these questions.

Speaker 1:

I think we're going to be seeing case law that answers it. And in terms of three to five years, i think lots of people have been wrong in both directions about the amount of time. When chatGPT came out, i thought I was pretty well versed in language models and what they could do and it really took me by surprise. I've spoken to lots of people in the field who said the same thing, who are more immersed in the technology than I am, and so you know. But then I felt a lot better about being surprised when I saw an interview with Bill Gates and another one with Steven Wolfram, who both said they were also surprised. You know Bill Gates talked about I'm sure you could all find this interview online. It was interviewed by someone from OpenAI who talked about going to Bill Gates' house last summer to show him GPT-4, you know, and he said he was surprised and had thought we were still five to ten years away from anything like that. So you know, if he can be wrong, any of us can be wrong.

Speaker 1:

So you say, two to three years from now, who knows, things could also slow down. You know progress doesn't go in a linear or even a predictable exponential fashion. Sometimes there's big leaps ahead in a short amount of time And I think you know, release a chat, gpt and all this big spur of innovation we're seeing now with large language miles might be one of those where there's a big spurt and then things plateau for a while. Maybe not, i have no idea. You know, and I just say that as someone who works with innovators all the time, i'm just trying to be humble enough to say that I could make some prediction. It's just so hard to extrapolate as to where things are going to be, especially, i mean, the pace of innovation is so fast right now that I'm seeing from clients of mine people who came to me a month ago with something that might have been patentable and today it might not be anymore. You know, it's just hard to say. I normally don't see things moving forward that quickly and yet they are.

Speaker 1:

But I think what we're going to have to see in terms you mentioned, case law or the patent office, i think the patent office is going to have to grapple not so much with the inventorship question but the question of what's sometimes called inventive level and, to use the US patent law term, it would be obviousness What's the degree of advance that something has to satisfy to be patentable, and how does the widespread availability of AI tools impact that?

Speaker 1:

What I mean by that is, if AI tools make it easier to develop significant new improvements to existing technologies, should that raise the bar for obtaining patents? And there is a requirement in patent law called a non-obviousness In Europe they call it inventive step meaning the degree of advance over the current state of the art that your invention has to embody in order to qualify for patent protection. And normally the way we ask whether an invention satisfies that requirements is to say what would a person this is the term of art in the law, the person having ordinary skill in the art have known to invent? And now and this is what I said in the book, you know, 15 years ago I think we should be asking what should the cyborg or the person with commonly available AI have been able to invent? You know, that person with their skills augmented have been able to. And if the law adopts that standard, it is going to raise the bar for patentability And I'd be very curious to see the patent office and courts try to apply that standard in practice.

Speaker 2:

Robert, there's a whole other episode just in unpacking some of the questions that you raised in that last answer. Unfortunately, we're bad at a time, but I'm going to ask you to come back so that we can have the second part of this conversation. Before I can let you off the hot seat, there is one other important question that I have to ask you. You are on a podcast called AI in the Future of Work and I opened up with a little bit of a teaser about Harvey and the impact of AI in the legal profession. When we're back here, whether it's version two or three or X of this conversation, is there a digital or an AI version of Robert that has replaced the version that I'm talking to now?

Speaker 1:

I hope not, but it's not just a hope.

Speaker 1:

I actively work on staying on top of what the latest tools are that are available that I can leverage, so that Robert or Robert plus the tools that Robert uses, are always more skilled than just the tools on their own or the tools when used by the average person.

Speaker 1:

I think any professional, anyone now has to look at themselves and their work that way now, which is to ask what is my value? What can people potential customers or clients of mine or my business obtain just using AI tools in their own skill? How can I justify my own value in light of what people are able to obtain just using AI and commoditized skill? When I say I hope that I will still be needed, it's not, it isn't just a hope. I think it'll be the result of me continuing to actively ask myself that question about myself, my own law firm, and making sure that we are constantly augmenting ourselves so that we can be ahead of the curve and provide and demonstrate that added value that we can provide that you couldn't get just using off the shelf AI without the extra skill that we have. Does that make sense?

Speaker 2:

Yeah, in an existential way no, but in a practical way, absolutely. it's a good answer. Is there a time when the roles of, let's say, the paralegal or the junior associate who's primarily not to oversimplify, but doing research that supports, maybe, a partner's work, is there a way that those lower tiers in a traditional big firm get subsumed and the partner can do a lot of the research? I think that's more likely than the paralegal or the junior associate being able to do the work with a partner. Do these law firms get flattened over time?

Speaker 1:

I think so. I think that's the general trend. After using chat GPT for a little while, one thought I had is that, in general law and otherwise, anyone whose job involves primarily finding and obtaining information from somewhere is at real risk. That's what you do. That is something. Increase in search the role of search And then of compiling, aggregating, even putting search results into a more easily usable form. That's one of the things that chat GPT is able to do.

Speaker 1:

Despite all of its limitations and hallucinations and so forth, it's demonstrated that we are heading in that direction of search or just finding, obtaining information, even describing it and summarizing it in relevant ways, is at risk of being really automated.

Speaker 1:

The other skill comes closer to what the lawyer does. It's just that of writing generally, Maybe writing based on existing information, to describe, summarize, characterize in existing known ways. That might take a little bit longer, and when I say a little bit, i don't know if that's one year or 10 years right now, but that's another skill that's at risk, which means, just as you said, what is left, those higher level skills. I'll just put it out here because we can leave it for the next edition, which is you and your listeners are probably familiar with that hierarchy of data, information and knowledge. What I've been thinking lately is we're starting to see knowledge, the ability to apply information to specific situations and come up with solutions. We're starting to see knowledge being capable of being automated. So humans, lawyers and others are going to have to then develop wisdom, which is the next level above knowledge. It's really a challenge, but that's the challenge we all face now as the technology keeps moving up those layers of abstraction and skill.

Speaker 2:

I think that's the topic for your next book, thanks, let alone the next conversation on this podcast. Well, hey, robert, this has been great. I really wish we had longer, but I am going to take you up on the you're approving my request to have you back, so I'm looking forward to that next one. Already, same here, same here. Good stuff. And before I can let you go, where can the audience learn more about you and your work? Yeah, go to blueshiftipcom.

Speaker 1:

Check us out. You can check me out on LinkedIn, which is LinkedIn slash in slash Robert Plotkin, just my name, and we've got tons of free content on the website. Most law firms don't do that. We've got blogs, podcasts, webinars, all on a lot of the topics that we've just talked about on this podcast, most of them more geared specifically towards patent implications, but also just generally about developments in AI and impact on innovation.

Speaker 2:

Well, this one's been a lot of fun. I learned a lot. I hope the audience did as well. Thanks so much, robert, for hanging out. Thanks so much for having me, dan, you bet. Well, that's all the time we have for this week on AI and the future of work. As always, i'm your host, dan Turchin from PeopleRain, and of course, we are back next week with another fascinating guest.

People on this episode

Head Shot

Dan Turchin

Host