In this episode of the Industrial IoT Spotlight podcast, we spoke with Robert Plotkin, Founding Partner of Blueshift IP and author of AI Armor: Securing the Future of Your AI Company With Strategic Intellectual Property. We delved into the evolving landscape of AI and intellectual property, discussing how AI is revolutionizing innovation and the patent process. Robert shared his expertise on developing systematic IP strategies for AI-driven companies, highlighting the importance of protecting AI-generated content and navigating the complexities of patent law in the AI era.

Key Topics:

• Systematic IP Strategy for AI: How to create a comprehensive patent strategy that maximizes the value of AI-driven innovations.

• AI’s Impact on Innovation: Exploring the ways AI is automating research and development, and its implications for inventors and companies.

• Patent Rights for AI-Generated Content: Understanding the legal landscape around patenting technologies, algorithms, and content created by AI.

• Defending Against Patent Workarounds: Strategies to broaden patent scope and protect innovations from clever loopholes.

• AI Litigation: How smaller players can defend themselves against larger competitors in court.

• Trade Secrets vs. Patents: When it might be better to protect intellectual property as a trade secret instead of pursuing a patent.

Connect with Robert Plotkin:

• Website: https://blueshiftip.com/

• Book: AI Armor: Securing the Future of Your AI Company With Strategic Intellectual Property

• LinkedIn: Robert Plotkin

IoT ONE database: https://www.iotone.com/case-studies

Industrial IoT Spotlight podcast is produced by Asia Growth Partners (AGP): https://asiagrowthpartners.com/

Erik: Welcome back to the Industrial IoT Spotlight podcast. I'm your host, Eric IoT One, the consultancy that helps companies track and understand B2B technology trends. Our guest today is Robert Plotkin, founding partner of BlueShift IP and author of The Genie in the Machine and AI Armor. BlueShift IP is a patent law firm specialized in an AI computer technology and software patents that helps companies protect their key innovations by developing comprehensive patent strategies.

In this talk, we discussed how to develop a systematic IP strategy for AI and how AI is changing the game by reducing barriers to innovation and barriers to patent creation. We explored a range of topics related to AI and intellectual property, including AI automation of the R& D process, patent rights and copyrights for AI generated content, strategies to enlarge patent scope to defend against patent workarounds, how smaller players can defend against larger players in court.

And when it's better to protect IP as a trade secret, as opposed to seeking a patent. If you find these conversations valuable, please leave us a comment and a five star review. And if you'd like to share your company's story or recommend a speaker, you can email us at team at iot1. com. Finally, if you have a technology research or strategy initiative that you'd like to discuss, please email me directly at erik.walenza@iotone.com. Thank you.

Welcome to the Industrial IoT Spotlight, your number one spot for insight from industrial IoT thought leaders who are transforming businesses today, with your host Erik Walenza.

(interview)

Erik: Robert, thanks for joining me on the podcast today.

Robert: Thanks so much for having me, Erik.

Erik: Yeah, this is going to be an exciting one for me. So we're getting into IP law around the topic of AI. I think you've chosen a good profession to be in for the times. Tell me, how long have you been focused on the topic of IP related to artificial intelligence?

Robert: I've been doing this for almost 30 years, focused primarily on obtaining patents for new types of AI and other software.

Erik: Yeah, that's incredible. Right? Let's say, to some extent, you had good timing. You just got there way before everybody else did. What did this look like 30 years ago when you first started working on the topic?

Robert: Yeah, I mean, I started being interested in how AI was being used in the inventive process. In the late 1990s, I was already starting to see people using AI to automate the process of inventing, to do things like help to discover new drugs and design circuits and even write software code, which is things that most people have only been really talking about widely in the last couple of years. But it was already happening back then, just not on this large scale. That really fascinated me as a patent lawyer because I was really curious about what was the impact of that on patents. For example, could you get a patent on an invention that was created using AI? Who might the inventor be in that situation?

So I started really looking into this in depth. And for quite a few years, I worked on a book on this topic, which was published in 2009, called The Genie in the Machine. I interviewed tons of people, including people who had gotten patent on AI-assisted inventions back then. The book didn't get a whole lot of attention, I think, because again, AI wasn't as powerful then and wasn't being used on as wide as scale. There weren't that many patents being impacted. But of course, ever since ChatGPT was launched in 2022, this topic has gotten the attention of the world, including the patent world. So I decided to write a second book which was just published a couple of months ago called AI Armor, which is a much more practical book targeted at growing AI companies on how to use intellectual property to protect their innovations, which is what I do as a patent lawyer. But of course, it's a lot more of practical topic now that a lot more people are interested in. In fact, the patent offices around the world have been addressing all kinds of questions relating to AI. Like, can an AI system be named as an inventor on a patent? What's the impact of AI on the standard for how inventive essentially something needs to be in order to be patented? All kinds of questions have been raised by AI for patent law. So it's been very exciting for me to be interested in and attached to this topic for so long and to see it finally getting a lot of attention, be involved in the public discussion and also to be working directly with lots of innovative companies on obtaining IP protection for them to help them in their growth.

Erik: Okay. Yeah, well, fascinating. I mean, it seems, from my somewhat of a layman's perspective, the IP topic, there's kind of two tracks of challenges at least that I see my clients wrestling with. One is this topic of AI in the product development process. And as you mentioned, the question, what can you patent if an AI is involved, and then how do you differentiate between the level of involvement? There's a separate topic which a lot of the more traditional companies that I'm also working with struggle with. And so a lot of these companies, they're coming from a background where they use patents to protect their market space, right? They're in chemistry. They're in pharma. They're in electronics, machinery. They're dealing with atoms. They can be patented. Now AI is becoming a more important part of their product portfolio. So then there's this question of, within our solutions that we're developing with AI, what aspect of that solution can we patent? And if we can't patent it, then of course, how do we still protect that IP? I suppose you cover both of those. Where tends to be your focus? Then also, is there any third or fourth area of patent law that is important for us to cover in addition to those two?

Robert: Yeah, I mean, I'll say one interesting development that's related to what you mentioned is that in the chemical and biological fields, I'll say again drug design, traditionally, companies would develop a drug and then they'd patent the drug in terms of its chemical structure, right? The patents would describe that drug, maybe how to manufacture it, what its structure is. Then the patent would be on that. Now that people are using AI and other types of software more and more regularly to design drugs, people are seeking patents on the AI algorithms or software that's being used to design the drugs. In some ways, those patents can be broader. In that if you obtain a patent on such a process, the patent would enable you to block anyone else from using the same method to design any drug.

So it's important for companies that are developing their own innovative algorithms, for things like drug design to decide, should I patent just the drugs that come out of that process individually, which is one option? Should I be patenting the algorithms for designing the drugs or both, or should I keep one or more of those as trade secrets instead? Usually, for a drug, you wouldn't be able to keep the drug as a trade secret because you want to sell it. But I've written about this pretty extensively. AI raises a lot more challenging questions about evaluating and weighing trade secret versus patent protection because of AI's ability to essentially churn out patents. They always have that question. Now, do you want to keep that AI method as a trade secret instead of exposing how it works to people, or do you want a patent that itself to give you that ability to block people from using it? Well, the reason I mentioned the response to your question is, this is coming up in all kinds of other traditional fields that normally wouldn't need to be thinking about software as part of their IP portfolio development.

Erik: Yeah, so I was chatting last week with somebody from a company that basically manages a patent database. They have a set of tools that will enable you to review through the structure of patents and understand where there might be infringement. They mentioned a bit offhand that about 40% of the users are now based in China, and they're basically using the database to do very structured kind of breakdowns of patents to navigate them, to understand where can we go make minor changes to accomplish the goal that this patented technology accomplishes well, avoiding being in infringement of the patent, right? So, obviously, this is a very kind of relatively matured technique right now. AI, I suppose, allows you to take that to the next level where you can automate largely that process.

But Robert, let's, for the sake of kind of structuring the conversation, break this down and maybe first focus on this topic of patenting AI. So what can be and cannot be patented? Then what are the alternative strategies, as you mentioned, to protect IP? Then as a separate, a second conversation, we can discuss this topic of to what extent you can patent the outcomes of AI, the innovations that an AI creates either independently or with a person or human. On this first topic, patenting AI, when can AI solutions be patented? When can they not be patented?

Robert: So first, at a high level, just in terms of generally what can be patented in the context of software, let's say. You need to have a software that performs some new method that does something useful, solves a problem. Let's say in a compression algorithm, right? So it has to be different than any previous compression algorithm, and it has to be useful. It has to compress data in a way that can be used, that is capable of being decompressed. They have to show some practical benefit of it. And it has to be, what's called in patent law, non-obvious, which is a very difficult term to define. But it basically means it needs to be an advanced, over the state of the art. It can't be just a trivial kind of an improvement.

In the context of AI — I walked through this in great detail in the latest book AI Armor — there's a few types of things that are most frequently being patented right now. One are algorithms for training models. Okay. Everyone is familiar with training of ChatGPT based on a large chunk off the Internet and other database like the patent database and Reddit and so forth. So if you have a new training algorithm, that's something that can be patented. Once you have a model and it's just sitting there, you need software that can — you can call it execute. Sometimes it's called performing inferencing. I'll use in the context of ChatGPT again. You type your prompt in. The ChatGPT software takes that prompt, and runs it through the model to produce the output. There's always some software that needs to do that. And if you have a new type of inferencing or what I'll call model execution software that's more efficient, for example, faster, maybe cuts down on hallucinations or things like that, that can be patented. Those two, training and inferencing, I'd say, are the two most common.

Another thing though I'm seeing is methods for post-processing the outputs. We know that AI systems produce outputs that aren't always completely accurate, or they may include a lot of low-quality outputs. If you have an algorithm that can find the needle in the haystack, in the output or, again, find and correct errors, rank, or filter, or prioritize the outputs of an AI system, those kinds of algorithms are things that can be patented. Those are the most common.

I wrote an article about the possibility of patenting prompts, which is a bit of agressive idea. But if you imagine that you could write a prompt to ChatGPT to say, "Hey, here's a new kind of method. Can you write me some Python code to do it?" Then it does that. That method is new and useful and non-obvious and satisfies the requirements for patent law. Now you've used ChatGPT to demonstrate that it actually could be implemented in a software. Why not? Why wouldn't the prompt be able to be the basis for a patent? So that's a little bit more speculative to patent prompts. Let me just say quickly. The kinds of things that are often kept as trade secrets in the context of AI are things like training data, training parameters. Those are often kept as trade secrets. One, because they're typically not patentable. And two, those are often the secret sauce that really give a company its competitive advantage. That's the high-level overview. Again, I go into it in a lot more detail in the book.

Erik: So the second question of when to patent and when to protect know how and specifically not to patent. How do you look at two patents or, let's say, two AI solutions and determine, is this a solution where we should patent because we believe it can be protected? And maybe a solution two. We believe we should protect. We should not patent because the IP will be better protected if we keep it as a trade secret as opposed to putting it in a patent database. How do you evaluate two different scenarios and determine the correct approach?

Robert: Let me just mention because it's a complex topic. I wrote an article specifically about this for AI that's on this blog, IP Watchdog, about patenting versus trade secret protection for AI use, this analogy of the old Aesop's Fable of "The Goose that Laid the Golden Eggs." Anyway, the first thing is, if you have a new AI algorithm or technology, investigate whether it qualifies to satisfy the legal requirements for patent protection. Okay? If it doesn't, then the answer is somewhat easy. You're not going to try to patent it, and then you have to decide whether to keep it as a trade secret. But if it does satisfy the requirements for patentability, you want to ask yourself: if we got a patent on this and our competitors started using the same method, would it be easy for us to detect that, to find it out? And for something like a training method, you might not ever know that your competitors are using your training method. Because it's not something they reveal to the public. So when a patent is difficult to detect, what we say detect infringement of — because you won't see it in use from your competitors because it's something they're doing within their company — then you might not want to obtain a patent on it. Whereas if it's a method, let's say, that involves a user interface and with a competitor of yours released the product, and it would be easy for you to tell through by using the product yourself that they've probably implemented your method, then that makes a better case for patenting. So the detection of infringement is really big.

The second one is how widely applicable is this method. If you think that this is a pretty foundational technology that's likely to be widely used, particularly by large companies who'd be infringing on a large scale and generating a lot of significant revenue, that would weigh in favor of obtaining patent protection for this. I'll just mention one more of many considerations, which is, if you're a small and growing company and this innovation is a key part of what differentiates you and provides you with value, you might want to obtain patent protection on this even if it could be protected as a trade secret. Because obtaining that patent could be valuable if you're going to get acquired or if you want to license the technology. It might protect you even against potential investors who you talk to and won't sign a nondisclosure agreement. But it might provide value to those investors or future acquirers because you have a patent that they can then license or enforce against their competitors. So wide variety of considerations to take into account.

Erik: And if you compare an AI patent against, let's say, a patent for a drug, kind of the molecular structure for a drug, how defensible do you find these patents to be? In terms of a competitor finding a workaround or a similar solution that avoids infringement but basically solves the same objective, do you find a significant difference in terms of the ease of backwards engineering for patents?

Robert: Not really any different for AI than for software. Generally, it's always hard to predict the future when you've created some new method and you ask yourself, is it possible to create another solution that solves the same problem in a different way? It's really hard to answer that question. Because you can't predict whether someone else will come up with something inventive that solves the problem in a way you can't imagine right now. But we all do our best. And sometimes it's more obvious than others that what has been created is pretty groundbreaking and pretty broadly applicable. Maybe if it solves a problem that people in your industry have been working on trying to solve for a while and no one has come up with a solution and now you have, you might have a pretty good idea that this is going to be something that's broadly useful for a while and that's going to be hard to work around.

The thing I would say that's more generally difficult about AI patterns is the fact that we know that today's models are really what's called inscrutable. We don't know exactly how they work. We know the general idea of how a neural network operates. But if you take something like a large language model like GPT, no one can really look inside it and examine it and tell you how it does what it does. So although it's possible to do something like patent a model, if there's challenges, then how do you describe what that model is in a patent application? How do you describe a system when you don't know how it works? Now we know from the history of chemical patents where often people create some new chemical composition or material by experimentation, and then they know what the resulting chemical does, but they maybe don't know what its structure is or how it works, there are ways to patent those things. And it's interesting. I mentioned a couple of ways in which our previous understandings of how chemical inventing works and how you might patent things in the chemical field are being changed because of the use of AI in chemistry. Now I'm talking about how we've applied patent law to chemistry in the past might be applicable now to AI systems that work in these somewhat mysterious ways, much like as chemicals and other things that are governed by the natural laws of physics. I hope that it makes sense.

(break)

If you're still listening, then you must take technology seriously. So I'd like to introduce you to our case study database. Think of it as a roadmap that can help you understand which use cases and technologies are being deployed in the world today. We have catalogued more than 9,000 case studies and are adding 500 per month with detailed tagging by industry and function. Our goal is to help you make better investment decisions that are backed by data. Check it out at the link below, and please share your thoughts. We'd love to hear how we can improve it. Thank you. And back to the show.

(interview)

Erik: How unified is the patent law globally? I mean, I know you're based in Boston, so you're probably working with a lot of the American technology companies. What does this look like in other parts of the world?

Robert: Well, you know, the law as it's applied to software has been pretty varied around the world over the years. I'd say there's been some degree of unification of patent laws that applies to software over the last 10 years. But of course, AI now has thrown things into a bit of turmoil in the last couple of years, and there's now a lot of variability in how it's being handled. Most of the major countries and regions have started to address it. But yeah, there's a lot of variability. It's causing people to sometimes make decisions now about where to pursue patent protection for AI based on where it's most easy to get patents. But what's difficult about that is, the law is changing pretty rapidly. Because it can take a few years to get a patent. It's a tough bet to file a patent application now where it's easy to obtain a patent and be confident that the same is going to be true when you're at the end of the process, you know. It might be getting more difficult in that country or the opposite. It might be that you avoid filing somewhere where it's difficult to obtain AI patents now. But maybe in two or three years, it will become easy.

Erik: Okay. Great. Well, that gives us a good foundation for understanding the basis of patent law for AI. Let's then talk to the second use case which is, your AI is creating innovations for you and you then want to patent those innovations. I guess there could be the extreme case where the AI is basically running the whole innovation process itself, and then the less extreme case where you come up with some percentage. You say this is 50% human, 50% AI or plus or minus. How does the law look at that scenario?

Robert: Yeah, that's also still very much in flux. But I don't think it's as complex a question as a lot of people think it is. I wrote an article a couple of months ago, again, analogizing AI to how the laws of nature will work. If again, going back to chemistry, if you take a couple of elements and you combine them together, you apply heat or water to them or something and you get a new composition out, you may not know how that occurred, right? The laws of physics applied to them and combined those chemical constituents together into a new material in some way that was not under your control. I mean, you put them through a process. Again, I just simplify and saying applying heat to them. But then nature did its thing in some mysterious way to produce a chemical composition, which you can then patent if you can test that new composition and show, oh, this is a new alloy that's stronger than any previous metal. But you can prove that by experimentation. You can get a patent on the metal. And what do you have to describe in the patent? You have to say, well, I took, I don't know, iron and tin and put them together. I applied this process to them. Then the laws of nature did their thing in some mysterious way, and I got this alloy out. And even though I don't maybe know exactly how that occurred, I know here's the alloy. Here's what it is, and here's what it can do. And that's why I'm entitled to get a patent on it.

It's very much analogous to how AI is working these days, right? Again, I'll use ChatGPT even though it's not the most common thing that's used for inventing, which is you take some text. You put it in, and then you get some text out. You don't know how that's happening. And as I mentioned, even the creators of ChatGPT may not know exactly how it's happening. It's somewhat analogous to the laws of nature. But we have rules and guidelines from the whole history of chemical patents that apply to help us understand under what circumstances can you get a patent in a situation like that. Where you as the human didn't really guide every step of the inventive process. There was something else, either nature or AI, that was churning to produce the output from the input. And we know how to write such patent applications, how to interpret the patents when they're produced in that way. So that's what I would say, that this is not a deal breaker for getting patents for AI. We can apply a lot of what we know and have done from the field of chemistry in the past to AI.

But the question that it does really raise, AI is making it possible to produce inventions a lot more easily than in the past. You could say that it is effectively boosting the inventive skill of inventors. I mean, this was a main theme of my first book back in 2009. Again, I used ChatGPT as an analogy even though, again, it's not really what's being used primarily in the inventive process. But it's been said that emergence of ChatGPT has boosted the writing skill of people worldwide. If you are a poor writer but you're able to express generally what you want to say, and you put that into a prompt to ChatGPT, it will produce average human output. So it's effectively boosted your writing skill. Well, that you can think of AI as being able to do that, right? If you know how to command an AI system to experiment, to produce a new type of material, or to write software, or to design a circuit or something else, the AI can do a lot of the grunt work, the simulated experimentation, and often produce either a final result, or a bunch of ideas, or candidate, or prototype inventions that you can then evaluate. So either it can boost your skill, or it can at least reduce the time required to invent really significantly. And so that raises the question: if AI is making it easier to invent, should the bar for patenting be raised conversely? So there's a lot of justification for that in the history of patent law.

The last thing I'll stop and say is, I think that's generally true that we should be taking that into account. As AI becomes more widely used in the inventive process, then we should be looking to that and say, well, the bar is now raised for patenting. But at the same time, because people are learning how to use AI to become better and better inventors, it's also I think enabling people to boost themselves over that higher bar. So there's kind of a cat-and-mouse game going on, as there always is when new technologies come out where it'll be harder to get patents on things that can be easily or trivially invented using AI. But as people become really savvy about how to use AI in the inventive process, those people who get really good at it will be able to create things that are sort of inventive enough to get over the new higher bar for patenting.

Erik: Yeah, that makes sense. I mean, are you seeing — I'll use the word entrepreneur for lack of a better word. But are you seeing entrepreneurs set up these patent farms where they just use AI to automate the patenting process? I guess, as much as possible, you also automate the legal process of obtaining patents if you're really trying to do this at a high volume. Not with the objective necessarily of selling products but with the objective of having attacks on the companies that downstream might want to use those patents. Do you see companies putting this type of business model into practice?

Robert: You know, I haven't seen people do it as what you've called a patent farm, but I can imagine that it's probably either happening or going to be happening. In the past, there were some companies that did this in the pre-AI days. They got together solely to invent and to obtain patents and then to sell and license those patents. What I'm seeing more of is existing companies who are already developing products and services. Then they're now incorporating the use of AI to boost their innovation process, to enable it to happen faster, to create more possibilities.

The thing you mentioned about on the flip side, the defensive side, which is looking at patents and using AI to evaluate how could we, you know — there's a mousetrap. I often use that simple example. And it's patented. Okay. Let's use some AI to come up with another mousetrap that has the same benefits but that doesn't fall within the scope of the patent. We call that designing around the patent. When I wrote my first book in 2009, I spoke to somebody who in the mid-2000s already was using AI for that purpose. Because you can imagine, with an AI system, and I'll refer to things like evolutionary algorithms. But you give them an objective. You give them a purpose. Very much like with ChatGPT, you give it a prompt which is like the high-level goal you want it to pursue. With the way you would use an AI system to design around the patent is, you'd say here's the mousetrap. Here's the functions it performs, and here's the specific physical components like a spring or a latch that it uses to perform those functions. Now what I want you to do is explore the space of components that can perform those functions and find me some that perform the same function without using those components.

It can be quite good at doing that. Of course, the challenge is sometimes the space to explore is so large that the time it would take to really do that grows. But we're seeing, because of this explosion in power of AI hardware processing and algorithms getting better and better, it's becoming more and more feasible to engage in the use of patents to design around. But when I mentioned the cat and mouse game, well, if you're an inventor then, how would you address that problem? Well, before you even file your patent, you would take your mouse trap and say to an AI system, how can I design around this? And you create a bunch of variations of your own invention. You include those in your initial patent as a way to try to protect it against someone else doing that after the patent is granted. So this cat and mouse type of phenomenon is showing up in all different kinds of ways in the patent world in respect to AI.

Erik: That's very interesting. Just a tactical question on that last point. I guess there's two ways that you could do that. One, you could try to package as much as possible into that patent. And the second is, you could say, okay, I'm a large corporate. I'm going to file for 10 patents instead of 1. I'm just going to have these other. Is there a material difference in the effectiveness of those two strategies?

Robert: Yeah, there's a bunch of trade offs. For the most part, it's usually a lot less expensive to the first thing, which is to make your initial patent bigger. So there's just a variety of aspects of how patents are, what their costs are that make that the case. So typically, it's more cost-efficient to put a ton of stuff into one patent application. But there can be reasons to split them up into many. But you know, one aspect of this phenomenon that AI is fueling is that, traditionally, when you had a breakthrough invention, someone would create it and they had the idea. Initially, that's what they had‚ a new type of micro process or architecture or something. And they file a patent on it. It would take some time for that person or company and their competitors to really think through what are the follow-on inventions that would arise from this. Then that would mean lots of different companies would be getting patents on the incremental inventions, you might say, or improvements to that initial technology.

I think that AI is actually now sort of enabling us to compress that process. That once someone has an initial invention, they have a strong incentive to use AI to invent as many variations of that or improvements of it as possible and then use AI. And even in the process of writing a patent application, using generative AI to flesh out as many variations so that you fill up the space as much as you can in that initial patent around the scope of that initial invention to make it more difficult for competitors to do what normally in the past would have happened over a fair amount of time, some number of years, and by a fair number of companies. So this is really putting the pressure on everybody to patent sooner and more broadly and extensively.

Erik: It sounds like we might be in an environment where there are more intense competition around patents. This, I think, traditionally has been a game where the larger corporates have had an advantage, because they have patent lawyers on staff. And where smaller companies, if they don't really play the game right, can be in a bit of trouble even if they have a good patent if they fail to defend it. How do you see that evolving? I mean, I suppose, to some extent, there might be new tools that are also enabling smaller companies to compete here. But how does a small player compete in this space given the evolving dynamics of mass patenting?

Robert: Yeah, I mean, I think you're right to point to small companies. I think AI right now is really — I wasn't the first one to use this term — democratizing innovation. Because although it's a small number of large companies that are creating all of the current leading models, well, not true. There's a bunch of great open-source models being created by smaller companies and universities. But they're getting most of the attention and certainly the biggest market share with tools like ChatGPT and Gemini and so forth. But because those platforms are available for anyone to use, and because compute power in the cloud now is so available for anyone to use, it creates a real opportunity for small companies to use AI for all of the purposes that we just talked about.

I mean, that's why my latest book AI Armor is really targeted at smaller companies who are just starting out, who are growing. And it leads them through how to use intellectual property strategically for their growth and success and eventually to get acquired or go public or engage in some other kind of successful exit. But I think AI really, although it's true that the large companies are able to take advantage of it at a scale that smaller companies can't, it's also true that smaller companies can use it to magnify their abilities in a way that was never before possible. I mean, I think it's why Sam Altman has wondered aloud, right? When will we have the first single person unicorn, billion-dollar company that just has one person who's really smart about using AI to grow a company to a billion dollar? The same question applies to patenting, right? Can a small company leverage AI, both to invent more quickly and effectively and on a broader scale than would ever have been possible with just a small company before and then to obtain patents and write patent applications and pursue patents?

I think the answer to all of that is yes. So it's a very exciting time for a small company. I think the real challenges for small companies are to find from a business perspective the market niches where they'll be able to succeed, where they will be able to against the larger company. That's a more of a business question than a technological and legal question.

Erik: On the legal front — we've been talking about patents — are there other legal issues where AI is being highly disruptive right now? So maybe I guess the obvious one would be copyright where you can create a lot of content, low cost. And if you can somehow secure the rights to that content, then you might be able to defend territory. I don't know if trademark, if there's any implications there. Are there any other legal domains where you see this as an issue that people need to kind of educate themselves about?

Robert: Yeah, copyright is the big one which, as you said, there's two sides to it. There is, if you train a new model or even use a model to create content, if you use DALL·E to create a new image, are you engaged in copyright infringement if that image is a copy of or contains elements that were derived from someone else's copyrighted artwork? That's one. Does it infringe? So I think I'd put the two together. Does the training infringe, and does the use of the model infringe? So those are really major issues which the law is grappling with now. There are some major lawsuits ongoing. The U.S. Copyright Office has put out its own guidance on that. Oh, sorry. Yet another issue then is, can you obtain copyright protection on a work that you create using AI to any extent? Again, you create an image using DALL·E. Maybe you used what you consider to be a creative prompt. Can you then get copyright protection on that image that comes out? That's another thing that is still very — that's a legal question whose answer is very much in flux now. And it applies to all kinds of content, not just artwork. It applies to text and video and audio and all of that right now.

I mean, I think there might be other areas where what we call the inscrutability of AI, the fact that we don't know how the system works, could raise issues in fields where you are required to, let's say, produce a document. Whose contents, whose accuracy you verified? There's lots of legal requirements in different fields and securities law and so forth where you need to verify the accuracy of the information you're producing. And so I think AI raises that kind of question. I think copyright is the really big one.

Erik: Okay. Got you. Robert, I have a rule which is, whenever I talk to a legal expert, I have to get some free advice. So let me give you a case study here that some of my clients are wrestling with. This is the topic of IP ownership for co-innovation. So let's say you have a mature chemical company or a pharma company. They want to collaborate with a technology company to embed their domain expertise in that technology company's solution and then bring that to market. And so you have this case where you have contributions from both parties. Are there any established best practices for how they navigate patent ownership of that solution?

Robert: Yeah, I mean, the first thing is to have a really good contract in place. And I say that because when it comes to who are the inventors under U.S. law, there's a set of rules that apply to that. I don't want to say you can override those rules with a contract, but you can make them somewhat irrelevant. If you have these two parties and they agree by contract in advance whatever we invent together is going to be owned by both of us jointly, or it's going to be owned by one of the two companies, then once that contract is in place — if it's well-written and well-thought out, and both parties are happy with it — I mean, maybe Company A agrees to assign all its rights to Company B. So Company B is going to be the sole owner of whatever comes out of the collaboration, but Company A is going to get a share of the royalties or revenue or something else. As long as both parties are happy with it and the contract is legally binding, then when the patents are filed, they'll be filed under the name of whoever the contract says. And that makes it really fairly easy.

There's a separate question, and the US has its own rules about inventorship, regardless of who owns the patent. Okay. Let's again use this example. Company A and B, they co-entered into a contract whereby all the inventions that they create jointly are owned by company B. When there's a new invention that they create and they file a patent application for it, you still need to identify who the individual human inventors were that contributed to that invention in the way defined by patent law. So even if there were inventors in company A, you have to list them as inventors on the patent even if company B now owns the patent 100%.

And I say that because companies trip up on this a lot. They confuse the ownership with the inventorship. They say, "Well, we own the patent. We don't have to list anyone from Company A as an inventor. Let's just leave them off the list of inventors." And they file the patent. That patent can then be invalidated later for failing to list all of the true inventors. Conversely, this has nothing to do with collaboration. What I see a lot is, you have a company, and the CEO wants to be listed as an inventor on the patent. Or at university, the head of the lab, because that person is always listed as a joint author on academic papers. That person wants to be listed as an inventor on all of the patents that come out of the lab. If you list someone who isn't an inventor, that can be a basis for invalidating later. The patent law has its own definition of who is and isn't an inventor. That definition is very different from the definitions that people use within a corporate culture or academic culture. And you have to use the definition of the law. Otherwise, you put your patent at legal risk.

Erik: Okay. That's fascinating. I think there's already a use case there to use an AI solution to filter through the patent databases and find all the prestige individuals that were listed there that most likely, 98% probability, they didn't actually contribute to this patent. Right? Great. Robert, this has been really helpful. Is there anything that we haven't touched on yet that is important for folks to understand?

Robert: I think we've touched on all of the key issues. I would say, for anyone out there listening who is an innovator, whether individually or a company, I hope this has been helpful to you in thinking about how can you use AI within your own innovative process. And is there any kind of innovations in the space of AI that you develop that are really significant improvements, that give you a competitive advantage and that might be worthwhile seeking some sort of intellectual property protection for to give you a benefit?

Erik: Great. Well, Robert, thank you. Last question. What is the best way for folks to reach out to you if they want to continue the conversation?

Robert: Yeah, reach out to me on LinkedIn. I'm at linkedin.com/in/robertplotkin. Just my whole name as one word. Or you go to my website, blueshiftip.com. That's the website of the law firm that I co-founded, Blue Shift IP. We specialize in obtaining patents for software and AI technology.

Erik: Awesome. Robert, thanks for your time today.

Robert: Thanks so much, Erik. I really enjoyed it.