Conversations on Applied AI

Ryan Phelan - The Impact of AI on Patents and Intellectual Property

September 14, 2021 Justin Grammens Season 1 Episode 25
Conversations on Applied AI
Ryan Phelan - The Impact of AI on Patents and Intellectual Property
Show Notes Transcript

The conversation this week is all about the intersection of Intellectual Property and Artificial Intelligence! I'm thrilled to have an expert in the fields of Artificial Intelligence, Patents, and Intellectual Property on the program. This is an amazing conversation and one in which I learned so much in just a short period of time from our guest, Ryan Phelan!

Ryan is a registered U.S. Patent Attorney and partner at the intellectual property law firm of Marshall Gerstein & Borun LLP. He works with everything from startups to Fortune 500 companies to develop and protect their innovations and businesses with IP.  He holds an MBA from Northwestern University’s Kellogg School of Management and is a former technology consultant at Accenture. He is also the founder of PatentNext, a Patent and IP law blog focusing on Next-Generation and New Age Technologies. Hs is also an adjunct professor at the Northwestern University Pritzker School of Law teaching coursework on Patenting Software Inventions in the United States.

If you are interested in learning about how AI is being applied across multiple industries, be sure to join us at a future AppliedAI Monthly meetup and help support us so we can make future Emerging Technologies North non-profit events! 

Resources and Topics Mentioned in this Episode

Enjoy!
Your host,
Justin Grammens

Ryan Phelan  0:00  
You know, there are certain ways to show that software, including AI is is patentable. And what we want is to show an improvement to an underlying machine. If we can show in our claims and in our patent, when we write it that there is an improvement to the underlying device, let's say, by improving its predictive nature, or perhaps the AI related software, the invention, the inventive portion of it allows the underlying machine to use less processing power or less memory are, there's some type of distributed nature to it. We can start we can describe all of that, in order to demonstrate that improvement and to show that it is a technical invention and something more than just doing it on a on a processor like the Supreme Court said that we shall not do

AI Announcer  0:49  
welcome to the conversations on applied AI podcast where Justin Grumman's and the team at emerging technologies North talk with experts in the fields of artificial intelligence and deep learning. In each episode, we cut through the hype and dive into how these technologies are being applied to real world problems today. We hope that you find this episode educational and applicable to your industry and connect with us to learn more about our organization at applied ai.mn. Enjoy.

Justin Grammens  1:20  
Welcome everyone to the conversations on applied AI podcast. Today on the show we have Ryan phailin Ryan is a registered US patent attorney and partner at the intellectual property law firm of Marshall Gerstein, and borun LLP. He works with everything from startups to fortune 500 companies to develop and protect their innovations and businesses with IP. He holds an MBA from Northwestern University's Kellogg School of Management, and as a former technology consultant at Accenture companies routinely work with him in a variety of technical areas and industries, including medical devices, artificial intelligence, machine learning and the Internet of Things. He is honored to have been published in several prestigious IP publications, and has been invited to speak at many IP conferences. He's also the founder of patent next, a patent and IP law blog focusing on next generation and New Age technologies. And finally, he is an adjunct professor at the Northwestern University Pritzker School of Law teaching coursework on patenting software inventions in the United States. You're one busy guy, Ryan, thank you so much for taking the time to be on the show.

Ryan Phelan  2:18  
Thank you, Justin, pleasure to be here. Thank you for having me. 

Justin Grammens  2:21  
Awesome. Well, so I know much of the topics today, are we going to be around artificial intelligence and intellectual property? But before we get there, I'm usually curious to ask our guests, you know, what was the path and maybe your interest growing up that got you into this profession and where you are today?

Ryan Phelan  2:34  
Sure. Well, first of all, even from a very early age, like so many of us, I had a passion for technology, anything that you know, blanked or omitted an interesting sound, I'd be looking at it taking it apart, and things like that. So I'm certainly one of those types of people. Throughout my life, I've always targeted that type of stuff went to college, I was interested in electrical engineering and computer science, primarily from high school days, this would have been in the 90s, when web browsing and HTML and stuff like that was kicking off. So basically followed that lead into college, did a little IE and then went full tilt into computer science after I learned how to program in code. And so from there, things get increasingly technical and more interesting for me, for sure, it sounded like you ended up at you know, Accenture, I guess what, what was your role there? Sure. I was a financial IT consultant at Accenture where I was working on Wall Street, some of the time and in Chicago in the derivatives market, others of the time, flying back and forth from Chicago to New York, doing a lot of very interesting trading algorithms for clients, such as goldman sachs and like that. And so usually, FinTech is probably one rung below, perhaps rocket science in that, you know, it has to be very precise, and very fast and efficient, because you're literally talking about the difference between several milliseconds costing, or making hundreds of 1000s of dollars, depending on the size of the trade. Things are very technical, object oriented, multi threaded, processing, you know, all of these concepts. It was kind of like the big data before big data was the word. And so those are the types of things that I was doing at Accenture using my computer science degree that I got in college. It was a fun time. Excellent. Yeah, so I've worked on a number of large scale systems. And I think you have a new appreciation for, you know, when you're using them as user on the outside having to understand a lot of the complexities that happens, you know, on the back end, even something as you might say, as simple as an auction site like eBay, you know, back in the day, just the amount of throughput and traffic that the sites have to take on and you're right, everything is very time very precisely that one mistake or something like that can cost a lot of money. So really, really fun stuff to work on for sure. challenging aspect of the world. 

Justin Grammens  4:48  
At some point you must have gotten involved in law then obviously, right so it was that a passion of yours for like growing up or intellectual property, I guess, how did you work through this transformation?

Ryan Phelan  4:56  
Sure. So at Accenture as a consultant, of course, you're going to be exposed. Two legal issues that come up from time to time. One of those was IP. And it was fascinating to me seemed to overlap with no IP, how can you protect these interesting systems that you're developing? Let's say, you know, on wall street or otherwise, because, you know, like you just said, Justin, there's a lot of complexity that goes on behind the scenes of the various programs that are developed nowadays, and especially was true back then. And, you know, the thought was, like, you know, wow, look at all this time and effort that's going into developing these complex systems, you know, Surely there's an interest in protecting this work product in so you know, the answer to that, of course, in the US, and outside of the US, too, is intellectual property, including patents, trademarks, copyrights, things like this. And, you know, having seen the other side of the table, you know, ask questions about that. I became curious, and you just started looking at law school. Interesting. Cool. And how long did it take you to get through that? So I was a bit different in my approach to that I did northwesterners, at least at the time, they had a different joint program called a JD MBA program, which is exactly what it sounds like. It's a combination of your you get your JD and your MBA, all packed in, though, within three years, typically, to get a JD, which is a legal degree in the US, it takes three years alone, an MBA takes two years, but Northwestern had figured out a way by combining some summer material and and having some other classes overlap with credits that you can achieve both degrees. Within that three year time period instead of was a busy a fun time where I took off my engineering slash computer hat for a little while and, you know, went into a new direction in the JD and MBA space, I wanted to get the NBA in case I, you know, for whatever reason, I wanted to go back to Accenture and keep wearing that same consulting hat, the MBA would would help kind of, you know, break some some ceilings as far as you know, trajectory on the consulting path was concerned. But at the same time, it would allow me to explore the IP on the JD side, which I certainly did. And since being sitting here as an attorney today, we can see, you know, which one of those was more interesting, although I still enjoy the business side as well. You still dabble in code at all, you just still write things are kind of largely out of that. I absolutely do still write code a lot. In fact, the last time I wrote code was this morning, I woke up after having written code last night to run something that I wrote in debug it this morning. So you know, not surprisingly, the code that I write is still in the financial realm. I like to program computers, mainly using Python. Nowadays, it's it's such a nice and simple language to do financial related projects, such as trading my money. And so I have several different projects, right now that I write, mainly communicating with interactive brokers to trade stocks and options, strategies. So that's, that's one of my hobbies that I do when I can. It's still fun for me, I like it keeps me up to speed on the technologies that are coming out which there are quite a many days. Yeah, for sure. You sort of live at this area, I guess, in the middle between kind of the same area that I love is, you know, artificial intelligence and Internet of Things. And, you know, you're obviously coming at it from an IP standpoint, I think most people listening to this maybe don't know what it takes for something to be patentable, I guess, what are the I guess? What sort of break it down into sort of law school 101. But when we're talking about intellectual property, I guess what is form intellectual property? That's maybe like, maybe that's my first question. Is it just something as simple as trademarks and patents? Maybe there's more to that? And then maybe a follow up question is, yeah, what what makes something patentable? Sure, I'll tack lives in reverse order, and kind of flow through it like that. So you know, first of all, the four big IP rights in the US are patents, trademarks, copyrights, and trade secrets. And so they each protect different things. So a patent will protect an idea. It's usually the broadest coverage for IP copyright will protect expression of something that's quote unquote, in a tangible medium, recorded in a tangible medium, you know, such as coin written on a napkin is a very simple example, or, you know, some code stored on a piece of RAM or ROM chip, it's very narrow in that it protects only that specific expression as written down, a trademark will protect the source of a good or signal such as the Apple logo is very famous. And when you see the Apple logo, you recognize that that is a product or something from Apple, which in your mind, in my mind has, you know, a certain quality or type. And then finally, a trade secret is something that protects information that is kept secret by a person or an entity such as a company, as long as they keep that material secret, then they can hold on to those rights. And each of these different forms of IP has their own statutory law and case law that has been developed in the US where you know, if you abide by those then you can

achieve IP protection for your product or your service or you know what have getting into patents specifically, for example, we have in the US 35 USC or United States Code or title 35, which covers patents. And if you look at the statute language in Section 101, which is one of the first sections in that and title 35, it tells us what in fact, is patentable in the US. And it's it's four broad categories of things. And those four broad categories of things are processes, machines, manufacturers, or compositions of matter. And so, in the computing world, three of those are interesting, and one is not so interesting. So a process, of course, is, you know, what we would think of as a running computer program in the computing world, a machine is, let's say, the system, let's say if you have computer code on a particular device, such as like a, an Apple iPhone, or something like that running, we can protect an invention as the entirety of the system. And then an article of manufacturer could be something like a ROM or RAM disk that can hold code. And so that's the third of the things that we can use for protecting software inventions. The last one, the composition of the matter is not so interesting to us, because it usually protects a biochemical type convention.

Justin Grammens  11:15  
I see. So yeah, competition, now you're probably getting into the actual the atoms, I guess, maybe the the pieces of it. When it comes to AI and you know, Internet of Things. There's been a question, or you and I talked about before, like, is this technology even patentable? Is there enough of a differentiator in these spaces that you're talking about? to actually say that yes, this is something that could be patented?

Ryan Phelan  11:36  
Yeah, absolutely. A lot of questions I'll get from computer scientists and others in the computing space interested in patenting AI or other software inventions is, can we patent software inventions nowadays? The real reason why that question is asked is because of a Supreme Court case that came up in 2014. Before this case, it was known at least since 1995, or whatever, you know, from another Supreme Court case that computer related inventions were indeed patentable. And, you know, many, many software related patents were filed and obtained from that case, in 1995, forward to at least 2014, where we saw a reversal of sorts from the Supreme Court in a case called Alice versus CLS bank, where the court came down with a new decision. And for those listening, the Supreme Court, you know, some of those on podcast may or may not know that Supreme Court is the highest court of the land in the US, probably many people knew that. And below the Supreme Court is, of course, lower courts, there is a specialized court in the US called the Federal Circuit, which is the court right underneath the Supreme Court that is specialized in hearing patent cases. So all appeals from District Court, which are the lowest courts in the land, flow up to the Federal Circuit, and this flow up to the Supreme Court. And so the Supreme Court when it comes out in the decision, that decision is binding on not only the Federal Circuit that hears the specialized patent cases, but also all district courts across the US in all 50 states. And so it anytime there's a Supreme Court decision, you know, just like as we see on the news, it has an impact, it sends shockwaves throughout the the patenting community, just like it does. Whenever you know, we're seeing any other Supreme Court decision that comes down this case that happened in 2014, Alice basically looked at a patent that was very broad. That was in the financial world, it had to do with a ledger system where third party intermediaries would provide debits and credits or shore up debits and credits for parties transacting on either sides in a monetary sense. And so what the court said in the claims were very high level in business written in, they didn't mention anything technical, such as a processor or memory or anything like this. They were strictly written in a business sense of distributing funds among in between the parties for purposes of facilitating that distribution between the intermediary and the court took issue with that, and basically said that these types of claims are abstract. And that's the big word, they're too abstract. We can't have these abstract claims that preclude others in the computing space from doing something similar, because all of these claims are just simply claim, you know, traditional business methods that say nothing more than do it on a computer. And so therefore, the court did not want to give a patent which is essentially a monopoly. All patents are monopolies granted to the owner for 20 year term to exclude others from practicing that invention, if they so want. So the court basically said we don't want to give somebody that power to exclude others from practicing this type of invention with such abstract level claims. And so all that said, from that point, district courts reading this Supreme Court decision analysis started taking an ax to some of these earlier patents than the computer base that were broadly written and could be determined to be abstract and invalid and if your patent is found in the abstract, that it can be found invalid and therefore would be rendered useless by virtue of that, that's kind of a change in 2014. From based on that Alice decision that we have to be careful about, however, said that, you know, since 2014, we're now sitting here in 2021. There are many, many court decisions that have gone the other ways, and we now your ways of making claims not abstract or in other words, patent eligible, we can discuss ways of doing that in this podcast.

Justin Grammens  15:27  
Interesting. So that that software in general, like you were sort of talking about, as I'm thinking through this, like, Can we tie it back to anything related to artificial intelligence, you know, specifically, underneath a lot of this stuff is our algorithms, right? That's kind of what's sort of the crux of what's going on Google, Microsoft, you know, Facebook, everyone's trying to build the best algorithm that they possibly can to make their machine learning as best as it possibly can. I don't know if I have a question here yet. I'm just sort of like maybe regurgitating or trying to, you know, understand that in 2014, kind of these these changes happened, where it was like, we had things that were deemed patentable in the past, were likely going to be harder to be patentable, right? Because in general, they were really sort of cutting off other inventions. Is that sort of true to say?

Ryan Phelan  16:11  
Yeah, absolutely. And so there's been quite the uphill battle from from then in 2014, to now to figure out how you know how you go about patenting software related inventions. And you're correct, of course, the AI, because it's a software related invention, we have to think about AI in those terms, as well. And I'll back up a little bit. And I'll say that the the court, including the Federal Circuit, which is kind of that Junior court, right below, the Supreme Court has issued some cautionary statements, saying things such as this Alice case with abstractness could be a real issue for future technologies that, you know, we want to make sure that we allow our inventors in the US to protect what they specifically called out as being artificial intelligence. So there's a real concern, even by the courts to somehow make sure that new inventions AI is a protectable asset in the US, despite the Alice decision, and the Supreme Court has punted a few times on decisions that could have clarified the matters, because with the Supreme Court, what it does is it it can only hear adversarial matters, meaning that two parties are in a dispute of some type, and they render a decision. And then that decision can be used as a backplate, to allow other parties to understand what the rules are, how that rule would be interpreted so that the Supreme Court has not yet done this since Alice, which is frustrating, Congress has thought that it would act in certain circumstances to overrule Alice by making a statutory change, but they have yet to do so. All that said, you know, there are certain ways to show that software, including AI is is patentable. And what we want is to show an improvement to an underlying machine. Okay, so if we can show in our claims, and in our patent, when we write it, that there is an improvement to the underlying device, let's say, by improving its predictive nature, or perhaps the AI related software, the inventive portion of it allows the underlying machine to use less processing power or less memory, or there's some type of distributed nature to we can describe all of that, in order to make it to demonstrate that improvement and to show that it is a technical invention in something more than just doing it on a processor, like the Supreme Court said that we should we shall not do. Gotcha.

Justin Grammens  18:29  
Sure. that begs the question I think you and I have talked about is, you know, could an algorithm actually be an inventor. So if there's some, this algorithm kicks out something that maybe humans haven't seen before, could an AI be listed as a inventor on a patent,

Ryan Phelan  18:43  
not under current statutory law. So this actually came up not too long ago, where there was a dabbas initiative by a professor Steven Fowler, in the US, and he invented a creativity machine. And his creativity machine would come up with what we think of as inventions in the patenting room, which the standard for determining whether someone is inventor is if they can contribute it to the conception of a claim when it refers to a patent claim. And the patent claims is just nothing more than a listing of what the invention is made of. And so for an AI related invention, or a software related invention, it would usually touch on the steps of the software that's running, like, you know, perhaps some information is received over a network. And then the information is used to compute, you know, maybe the information is training data, and the training data is used to train an AI related model. And then the model is used to make prediction or something like this. And so, you know, the, the inventor would be someone, a person that conceived of those particular elements of that invention. And so, what Steven Fowler did is he, for the first time ever indicated on his patent application that dhabas a machine was the inventor usually you would look Human as an inventor, but he listed a machine and this created issues at the patent office here in the US. He also filed in other countries in jurisdictions, including the UK and in Europe, and others. And basically, what all of these countries, including the US said is they look to their statutory law. And the way that the law is currently written, is that it's it includes words such as whoever invents, or the person that invents and things like this. And so the patent office must follow the that code that we talked about earlier, title 35 of the US code. And when they looked at that, they said, well, the statutory laws written clearly contemplates a person or human inventor, and so therefore, your application is defective by virtue of not listing human. And so they requested him to actually list one he did not citing reasons, you know, that he could not list it because he did not invent it. He wasn't an inventor. And so therefore, they denied his application in so what myself and others in the legal community think now is that if there is to be a AI inventor, or an inventor that could at least be listed on a patent application, and require a statutory change, which would have to come from Congress in order to change those words in order to allow an AI TV listed as an inventor. So right now, things are still in flux. With that, there's certainly been questions whether that should be done and to what extent, and other countries have come out the same way, including Japan, China, and Europe, all have similar statutory code that precludes AI related inventors. And that's merely a function of the laws, how they have been written, usually, the law lags technology by a good 10 to 20 years, which is the case here.

Justin Grammens  21:41  
Yeah, interesting. It's gonna be interesting to see how it plays out. One could argue, well, whoever programmed the algorithm maybe would actually be the inventor of it. But some of these things are, you just don't know where they come from. Right? I mean, I would think it'd be difficult to point to an inventor, if it's actually the machine generating the results, I think of something like GPT three, which is basically, you know, it's creating its own words, right? It's, it can actually write a poem for you, and say, you were to try and patent that poem, for example, you know, I know, it's, you know, apples and oranges. But whatever the output is, how did that come from? Right, there wasn't a human behind that, as a human. You know, there's a bunch of people at Google that put together this algorithm to make this thing happen. But it'd be very difficult to assign that to a living and breathing person, I think.

Ryan Phelan  22:23  
Yeah. And that's the debate is, you can have an invention, an invention that has joint inventors in it, meaning, you know, more than one. So therefore, the question becomes, you know, if you have at least one human inventor, can you listen AI as like a second that are? That question is still up for grabs. But you know, that certainly puts one foot in the realm of Okay, so that's possibly statutory. And so there's actually a very interesting case, that's informative for what's happening here. And it actually occurred in the 1800s, the late 1800s. And it had to do with cameras, and I think this would be interesting to talk about with the audience here. So I'll make an analogy. So, you know, in the late 1800s, you know, photographs were becoming more popular. And this was a copyright related case, but the question became, well, if you're a person using operating a camera, you just push a button and a picture, a portrait is generated, you know, well, who actually owns the copyright to that? Is that the camera, the camera, do it? Or do the person pushing the button, do it or something like that? And so, of course, nowadays, it's fascinating to think about this, because we don't think about we think about the, you know, the photographer, right, that takes the photo, that's the copyright holder, no question. But back in, you know, 100 or so, years ago, there was a similar question as to you know, hey, a machine is painting this picture, you know, is it really a machine or is the person and so, what the Supreme Court said in that case, that became a Barrow geils basically is that the person could be considered the copyright holder, because he or she was responsible for moving the camera around to select the theme, the possible lighting, you know, and all of these other creative aspects that go into you know, capturing a photograph. So, here a similar argument can be made in that computer scientists or other other person, you know, in this field, could you know, for an AI related invention could select training data, right, and depending on how they supervise that training data for a supervised learning type methodology, or you know, invention, then you know, that person would be the inventor of you know, whatever model or you know, whatever device that use that model would be because they had hand in crunching or applying, manipulating that training data before it occurred or whatever process you could also use the same type of analogy for unsupervised or reinforcement learning as well. So, you know, those types of things if you wanted to talk about the difference between, you know, fully machine invention versus a, you know, half and half human plus machine invention, those types of things. Come up with you're really quite fascinating that that we're at this point now that that can be done.

Justin Grammens  25:04  
Yeah, absolutely. So it feels to me like, it's really the creativity or the person that actually, like you said, created the training data and ran it through the machine. They're the ones who can legally own this patent. And really, the computer is just sort of a machine in this, it's it's like, it's like a car, right? The car doesn't go anywhere. It takes a human to get in it and drive it around and point in there in the in like, the right direction is that that's kind of what it feels like the sort of similar analogy.

Ryan Phelan  25:30  
Yeah, that's exactly right. And in fact, you know, if we were, let's say, you and I were going to talk about, you know, patenting your AI related invention, you know, one of the things that we would talk about was like, Okay, well, what's, you know, is this a supervised? You know, most inventions are supervised, in most AI related technologies, or inventions or applications are supervised anyway. So we'd probably have a discussion about what your training data looks like, what are you using? What's your data set? You know, how many inputs does it have? You know, how many columns, you know, what are your hyper parameters, stuff like this? Like, how are you pushing this into the, you know, some type of AI related algorithm is this a neural net, things like that. And so, once we have that down, that would be an element of our patent claim. And therefore, you would automatically be an inventor, because you would have told me at least one element of the claim, and you would have contributed to that, which is the magic formula for naming you as an inventor. Other steps in the claim process. And there's kind of a magic 123 process for claiming any AI related invention, at least in the supervised learning category is that you want to describe, you know, that your training data that you're using, you want to describe how In fact, or train your model. And then you want to also describe how you're using your model, in some type of predictive or classification sense, for some type of in use. A very popular one is an autonomous driving vehicle. So you'll have a model, or you know, one or more models that are trained, that are put in type of some type of autonomous vehicle or robot that can be used to steer or drive or direct that robot, you know, an environment. And so that's pretty much how your claim would look. And you can use that same claim strategy to describe your improvement, which we talked about before. So we can satisfy the Supreme Court's test in the Alice case. And so once we kind of wind all those things up, you know, where you're the inventor, plus where, you know, we're able to demonstrate the improvement, and we're appropriately describing the AI related invention from, let's say, a supervised machine learning aspect. You can do the same thing for reinforcement, or again, or unsupervised learning example, as well. But once we have all those things lined up, then you're in a good spot for patenting your AI related invention.

Justin Grammens  27:44  
Yeah, absolutely. It feels like, I guess just so just step back for a second, you know, as you're an attorney, and you're probably watching all the patents that are being issued, you probably have, you know, clients that are submitting stuff and working on stuff in this sort of AI IoT world feels like better data will lead to better models. Are there other areas improvement or invade era advances like in general that you're seeing? Or is it really like people just sort of making these models better and getting better data? Are those the types of patents that are happening these these days? Are those are the types of ultra property claims that are happening? Or are there other areas where like, now, this is really interesting, people don't know about this. But since I'm sort of watching all this stuff going on, there's another area where we're sort of seeing a lot of big leaps and changes in technology. And it could be an AI Honestly, it could be an Internet of Things. It could be in any other area. But I'm just sort of curious, based on your perspective of what you're seeing going on in the legal landscape. What's sort of coming across your desk that you're like, wow, this is kind of cool. And I know you can't talk about you know, anything that would be under NDA or things like that, but I'm just wonder if you have any sort of like, general observations?

Ryan Phelan  28:46  
Yeah, absolutely. And that's an excellent question. And I'll answer a very high level for the reasons that you laid out. But basically, I'll say something like this, like 99% of the AI related patent inventions that we see that other firms have seen when, you know, talking to other patent attorneys at other firms, our use of existing AI related algorithms and open source software such as Google TensorFlow, or Facebook's AI torture, Python sky kit learners, something like this use of those to do something with so most inventors are not interested or at least, you know, their inventions do not involve going underneath the hood and tweaking the AI, you know, algorithm itself. They're more interested in using the off the shelf, open source packages in order to do something with And so, again, a lot of times you'll see that the real novelty in the inventions is the specific selection of the training data and how the the model is trained in you know, one of my favorite definitions of machine learning is from young mkuze Facebook's AI hit. He defines it he says machine learning is the science of sloppiness, which I absolutely love. You know computer science myself, because pre AI days you know, you would write a program Going to be very procedural, you know, you'd have step one, step two, step three, bunch of if then else statements, if you have like an expert system that, you know, went through all these different permutations of data flows and flowcharts, and stuff like this, in order to get to a decision, now, you know, fast forward where we get, you know, more powerful machines and computers and bigger data, and we're able to do AI on a massive scale, not that AI wasn't, you know, around before, it's been around since the 1950s. But we've now arrived at a point where we have the compute power to do it. But now, you know, we can kind of flip instead of like some, you know, flowing program that I just described, now you have, let's just take all the data, push it through an AI algorithm come up with a model. And we're going to use that to make decisions instead of having some very specific high risk data flow diagram of how things are going to be determined. And so you know, hence, the the definition of the science of sloppiness, you can be a little sloppy, and you say, Hey, I got all this data, I'm not 100% sure what it's going to explain or what or if this will actually be predictive. But you can keep training, keep training, and keep tuning and keep tuning your your model until you get something that's highly predictive, or something that's very useful for classifying or weather like this. And those are the types of inventions that we see now that 1% of invention. The other type is the stuff where they go underneath the hood, they say like, well, the current algorithm is not as effective because it does this or we need to add something to it. And then you start talking about typically augmenting some mathematical stuff that's happening in AI training, maybe there's a back propagation step, maybe there's, you know, some type of different, you know, weighted step, maybe there's adding more hyper parameters or something like this. And so, you know, those types of inventions, while clearly in the AI space, because you're actually touching the underlying AI algorithm, there's are more rare, more people are just using AI, as you know, you would use a processor, for example, to do any software related program, every software related program pushes instructions through a processor. And so people are doing, you know, the same thing with open source AI platforms in that similar capacity.

Justin Grammens  32:08  
Yeah, makes sense. It feels to me like I've always thought of software as like, just like little Legos, I guess. And you're assembling pieces together. And somebody can make you know, an animal out of Legos and somebody can make, you know, a building out of Legos if they want. And so it feels to me like it's a different assembly, I guess, of the inputs and outputs that you're talking about that maybe make them distinct and unique in these cases, not the underlying algorithms that are used under the hood most of the time.

Ryan Phelan  32:32  
Yeah, 100% agree. One of my favorite toys was Legos. Back in the day, I think of software similarly, especially now, given the many, many different API's, ai being, you know, one of them, the AI package is providing an API into the AI world. And so absolutely, you see people assembling things from an AI package, or maybe an IoT or everything nowadays, instead of building software from scratch, is you're combining a lot of different API's or, quote unquote, stacks, so you can have a platform or device very quickly.

Justin Grammens  33:05  
Yeah. And so I think this thought just sort of came to mind was, you know, I think we're thinking about artificial intelligence. And a lot of these pads are being done in the positive sense. You know, I don't know, you working on IP, but I just don't know if you deal with at all, you can have your finger on anything going on with the deep fakes, right? So any of this other stuff of people saying, Yes, I'm using it for a negative way. But is there? I mean, are there deep fakes that you know, of that are been patented? Or that there's trademarks on or anything like that? Are you? Are you aware, watching some of that stuff at all?

Ryan Phelan  33:32  
I have not done anything that involves the use of the term deep fakes itself, I certainly know what they are. And I certainly have patented and familiar with a lot of virtual reality, augmented reality. And graphical user interface type patterns, I don't think of an inventor would would call their own invention, that deep fakes. But in that same realm, you know, where you're, you're doing things such as your graphical manipulation to map somebody's face onto somebody else's face or things like this. That's all in the same, you know, wheelhouse of AI. And so, Could somebody get a patent on that? Yes, if they did, it would probably be, you know, dressed up more in like the, you know, again, the VR, ar gooey space, it'd be interesting to see how they would use that patent, you know, coming out the gates, a lot of times, you know, for those types of patents that I've seen, especially now, you know, post COVID people are interested in those from a consumer product perspective, such as, let me try on the product, you know, from the comfort of your home virtually, versus having to go to a store and try on something, you know, such as like, a pair of sunglasses or like this. So,

Justin Grammens  34:39  
yeah, exactly. I wasn't sure if there have been any legal rulings that have come down around malicious use of Well, I guess in general, probably anybody can use a patent in a malicious way, I guess. Right.

Ryan Phelan  34:49  
Yeah, they and there's there's a whole body of law on patent misuse out there. For example, you can get into any trust issues if you happen to have a whole patent thicket as we call it, which is just a bunch of patents. And you're asserting news in a way that would cause harm to consumers by inflating prices, things like this, those are rare most of the time, which you will see in the news for patents misuse, or at least, you know, it's fun to talk about as patent trolls, which are individuals that don't make anything, they don't have an entity per se. But they, they buy patents as assets. And they use them to assert the patents against companies that do make products and the apples and Google's and Microsoft's of the world, you know, spend a lot of money defending those types of cases year in and year out. And so, there's a real interesting, I suppose, it's a political debate to be had about, which is an area that I like to stay far outside of, but there is a debate to be had is like, you know, okay, if anybody can get a patent, you know, inventors, that's a good thing. But what happens when they get in the hands of the trolls, per se, which non practicing entity is a polite way to call them, you know, what happens when they get these patents, and now they're, you know, it's no longer the inventor, it's, you know, a third party that owns the patent, because they bought it from the inventor, asserting these things, you know, you know, how do we solve that? And so it becomes an interesting discussion on both sides.

Justin Grammens  36:10  
Yeah, very cool. Just a couple more questions, I guess, as we're getting kind of near the end. But, you know, I like to ask people about just artificial intelligence, sort of machine learning and stuff like that taking over people's jobs in the future. And I've known of a couple people that have created algorithms to look for patterns. So they actually, I think, I'm probably gonna get this wrong, but I know they were actually looking across a wide swath of various patents that have been granted. And then essentially trying to figure out which person or which entity is probably the most patentable entity here. And so it was a pretty interesting project that they've been putting together. It feels to me like, obviously, that is something that somebody could have done in software 1520 years ago, it would have been a lot more difficult, though. And I'm just wondering, from your standpoint, where do you see your job, I guess, the job that you do on a daily basis, enhanced, harmed, maybe it's negligible, but like, you know, did you see your job being impacted at all by this technology coming out?

Ryan Phelan  37:04  
Yeah, I mean, that's a question for everybody. I think jobs that are highly cerebral, intellectual property, being one of them are pretty safe for a while, in my opinion, I could be totally wrong. But I think that, you know, the amount of detail and nuances that go into preparing a patent application to be successful in a bunch of different legal ways, would be very difficult to train a machine to do, like, right now, as we sit here. Now machines, using AI can be trained to do amazing things such as you know, I'm thinking of AlphaGo, you know, being the the go masters of the world, and, you know, other things like this, but, you know, when you think about it, these specifically trained API's are really good at mainly doing one thing, and that's the thing that they've been trained for. So because IP involves doing many, many, many things layered on top of each other, I think it'll be quite some time before an AI system can come out to kind of do everything, for example, you mentioned, you know, searching for, let's say, you know, specific patents, let's say prior art are ever in the value of patents, that's been, you know, attempted to be done, you know, many, many different ways over, you know, decades. And now even, you know, using AI, even the most recent ones that come out, you know, they're still not as good as, as the ones that we've had in the past. So it still remains a very difficult problem. I have thoughts on how, you know, it could be done as well, you know, would it be successful, or more successful than previous projects that have done to do these things? No, probably not. Now, of course, you can automate some things here and there. But building a company around like a single, you know, automation of one aspect of IP law is just too expensive and probably, you know, not worth it. At the end of the day there. You know, I get lots of emails from individuals every day having a new platform that does one thing, but, you know, I really don't need that one thing I need, if you're going to like push the button and have the patent, like spit out of your hair, we're far off from that world.

Justin Grammens  39:07  
What about on the other side of it, you file a patent, and there's a patent examiner that's assigned? Could the examiner or maybe their patent office already does this? Do you think when somebody files something, you think there's a automation process that happens again, just purely speculative, but they must get so many patents?

Ryan Phelan  39:23  
They do. In fact, they're exploring the use of AI right now, and something that they do quite often which is called patent searching. And so what the patent office primary job is from a high level is making sure that you know, no bad patents get out and get issued. And so an examiner, a patent examiner, which is the person that will review any patent filing that you file with the patent office, their primary job is to look at, you know, your claims your patent claims, which will define what you have invented, and to compare that to the quote unquote, prior art, and the prior art is nothing more than previous patents that have already been filed, published. Issue patents and then publications out there in the noon world such as you know, let's say an I triple E publication or a standard or even newspaper article, or something like this. And so what they do is they have two databases that they search called east and west, and basically their searches, and you'll get your search list in your file wrapper, which is kind of a history of everything that has occurred throughout the prosecution of your patent application. And we'll show how the examiner search in right now, when you look at it, it's a bunch of and in ORS like tradition, Boolean searching, there'll be a ridiculously long statement of this term, and that term, or that term, and this term, and you know, all kinds of different permutations. And so what the patent office is trying to do right now is automate that process or make you better by using AI, which will include Of course, you know, some type of linguistic AI, or natural language processing, where certain words are known to be synonyms for others. And you know, they won't have to use the N and RS for specific words, they can, you know, the AI invention will presumably do that for them and spit back prior art references that the examiners can look at. And what they do is they compare the prior reference to your claims. And if there's a match, they're like, well, somebody else has already invented this. So you can't be the inventor of this, because it's already been invented. And so you go back and forth with the patent office to either argue against that, or to amend your claims to add in something new, where you're able to say, Well, okay, maybe that has already been invented. But when I add this additional aspect, then, you know, all of this stuff together hasn't been invented. And so perhaps AI in the future will help examiner's with that process. And I think it will, I think that that's the lowest hanging fruit, and the easiest one to find the patent office and others to target because there's a treasure trove of data. And you know, natural language processing has existed for a while, which can be applied to a text of prior references out there.

Transcribed by https://otter.ai