 Well, looks like we will get going, I guess, especially as we've got a limited audience here. If you've got a question or you want to make a comment, raise your hand, throw it out, and we will, rather than having this be just a straight up presentation, we can have a little bit more of a discussion. On the other hand, if you want to be quiet, well, I've got more than enough to say. Okay, so the question there was, how good is the indemnification for open AI if you're building some sort of model? That's actually a different question than the one we're going to talk about, but I can loop around to that right at the end. So let's talk a little bit. So that's actually a great place to start at. What am I talking about? Pretty much if you're here, you are interested in building AI-powered applications, or you want to use AI to generate various types of content or help people create various types of content. People are used to thinking that the intellectual property system works for them, like I write a bunch of code, I do some things, I have some sort of content. It is protected by default. What I'm here to tell you is that under the current US Copyright Office Doctrine, a whole bunch of the stuff that you would normally think are copyrightable, licensable, subject to normal licenses, just are by default in the public domain and have absolutely no protection whatsoever. This is throwing a lot of wrinkles into a lot of different issues. And as you can probably tell, I believe that this particular policy is not only completely wrong, but it's completely unsustainable. And I believe it will change sooner or later, but sooner or later when you're talking about a government office could be something on the order of years or a decade. And so let's talk a little bit about, let's talk a little bit, let's get down to the nitty gritty. Specifically, US Copyright Law comes from a section in the Constitution that says it is designed to secure for authors and inventors the rights to their inventions and discovery, their inventions, writings, and discoveries. So this idea that it is supposed to be for authors has been interpreted various times to mean human authors. And there have been some famous cases that have been associated with that. Some of y'all may have seen the monkey selfie case where the monkey got ahold of the camera and took a selfie. And everyone was like, oh, I wanna have this photo because it was a monkey that took a selfie. And some people were like, I am gonna claim copyright on the monkey's behalf. And they went and they tried to copyright it. And the courts came back and said, not only the court, but the copyright office as well, came back and said, actually, a monkey is not a human. We are not here to incentivize monkeys, we are here to incentivize humans. Therefore, the monkey has no copyright. This particular photo is in the public domain. Anyone can use it. Similar things have come up with regard to someone who claimed that a particular book was written by aliens or that a book was written by God under inspiration or various other sorts of things where people have claimed in one way or the other or established that this was a non-human author. There were some artworks that were made by elephants. Again, not copyrightable. And then came AI. So the most straightforward application of this was first challenged by a gentleman named Stephen Fowler who had this machine called the Creativity Machine. And using the Creativity Machine, he did two things. He first of all created what he claimed was a patentable invention. And he claimed and he created supposedly completely autonomously a picture called a recent entrance to paradise. It was like a picture of like a garden or something like that. And he tried to register these in the patent office and the copyright office specifically. The copyright office said, you know what? We already know what to do with this. This was made autonomously by a machine. That's what you're telling us. The machine is not a human. Sorry, not copyrightable. And they went and they challenged that and it came back and said, nope, not copyrightable. But then came Zari of the Dawn. Now, have any of you heard of Zari of the Dawn? Zari of the Dawn was a comic book that was created by an author named Chris Cashtonova. Chris wrote this story and then used mid-journey to create a series of images that went along with the story that she wrote. It was designed to, Chris went through actually hundreds of different iterations. First, trying one prompt, then trying another, then modifying the prompt and using the previous generation image as a seed image and going through hundreds and hundreds of time. And actually created a consistent view and image and set of characters and look that went throughout the comic book that was then cropped and put into context and registered. People started copying the comic book and so Chris registered it. And then when the US Copyright Office said, okay, yes, this is copyrighted, Chris happened to mention on social media, hey, I was able to get a copyright for my comic book that used mid-journey. The Copyright Office freaked out. They wrote a letter and said, please respond to us within 30 days why we should not immediately cancel your copyright registration. That's the moment at which I got involved with Chris in the Zariya case. We wrote a response to the Copyright Office saying, here is what went through. We designed and went through all these iterations. We created all these different images and we put it in the context of this copyright book. You should recognize that all this stuff was Chris's work. The Copyright Office came back and said, well, we agree the words were written by Chris Castanova. Those are copyrightable. The selection and arrangement of all these photos was done by Chris, that is copyrightable. But each of the individual pictures, we actually don't believe that Chris had anything to do with the creation of them. That was actually completely random. And because it was done by the computer, Chris may have suggested to the computer what should have been done, but actually had no control over what was going into it. As a result, they said, well, none of the images are copyrightable. If you can't tell, I'm a little bit salty about this because they got it 100% wrong. But why is this so important? Because generative AI is just a tool. It is an extremely important tool. In fact, it is the first general purpose software tool that we've developed. But unlike previous general purpose technologies that have replaced physical work, generative AI replaces mental work. And so accordingly, we should expect disruption in fields that rely on mental capabilities, even as we see greater levels of creation and creativity. This puts creators, copyright owners, in the case here, coders at severe risk, not only because AI systems are being trained on works that are already created, but specifically because creators are being heading into a period of disruption that they have not seen before. And they are doing it without having the typical tools that they would have for any other type of creation, specifically the ability to have enforceable copyrights in the works that they create using AI tools. This is the greatest risk to the copyright system and to the economic force that it enables. I know that some people may be like, I don't like IP, I don't like copyright. But the fact is that even if you are a complete open source proponent, which I am to let you know, open source actually relies on the underlying legal structures of copyright to make it work. If you don't have that, then all of the licenses that you like to use simply don't mean anything, they're just words on a page. This one, I like, when I think about this, what is going to happen to all of these industries, I think about actually the history of letterpress printing. Now, you all may not know this, like currently when we think of printing, we think of computerized printing. But for a long time, posters and things like that were created using a letterpress, which was, in some cases, they would use these blocks, this was very much like Gutenberg did it. But for posters and things like that, they would actually carve out the images out of wood or rubber or something like that. And so you would have these very charming, sort of slightly different images for each one of these. You'd recognize letterpress printing if you saw it. It's that sort of old-timey look. From the perspective of the letterpress industry, the rise of computers in computerized printing was a disaster because from the, from the, in the 70s, from the 70s, when letter, there were like something around 11,000 different letterpress printing shops in the United States, it has now gone down to maybe 300 or 400. Just like an order of magnitude or more drop in the number of companies that were doing letterpress printing because it just wasn't as efficient. But the question is, has the number of things that have been printed gone down? And the answer is, absolutely not. The amount of printing and printed material has skyrocketed over that same time. Why? Because computerized printing actually made it possible for more people to print more things to actually have more creativity. What it did is it disadvantaged a certain type of very bespoke print, bespoke printing. Out of curiosity, what has happened to letterpress printing? Well, it has actually come under sort of a revival as an artisan type of printing. It has become an art where they've emphasized the human aspects of it. In the same way, AI is going to lead to a huge reduction in the number of people who are working on certain types of intellectual outputs. But that doesn't mean that we're going to have fewer outputs. That means that we're going to have way, way more. We're going to have many more because it's going to be cheaper and easier for people to create than ever before. But unlike the letterpress example, all the people who shifted from letterpress to computerized printing, they didn't stop being able to enforce the copyrights in your posters. You still couldn't copy someone's poster if it was printed on a computer. But that's what the Copyright Office is doing right now is they're saying, hey, if you create this image with AI or you create this book using AI, all the stuff that you used, that you did using AI, can be copied by anyone in the world. It's not protectable. But the thing is we know that AI is going to be out there. It's going to be used everywhere. And how do we know is because we are already starting to live in that future in the world of code. If there's any place where AI tools are starting to become more and more used, it's actually in the world of code. People, developers are using Code Whisperer. They're using Copilot. They're using Tab9, some of these other tools. And in my own practice, it has gone from, within six months, it's gone from, we are investigating this to see if it's even a possibility to, we are working out how we are going to roll this out because the changes and the increased productivity are astounding. It is going to be there. But all of a sudden, all of these things, your open source projects, your commercial projects, all of these things are now going to be lacking some of this essential protection, your ability to license it effectively. But we know that these things are positive and they are taking over in spite of these problems. Why? Because these tools are so effective. We're actually learning that AIS assisted coding tools enhance but don't replace human expertise. If you've, just like if you put something into chat GPT and you've said, tell me about X, Y, Z, maybe it'll do an okay job, it'll sound very competent, but if you know anything about it, it probably won't get it right. It probably won't have the quality that you need. Same thing with code. You need to still have an understanding of what's going on in the code in order to use these tools. It just makes things more efficient. The second thing is that if you're familiar with the economic concept of induced demand, it means that when you make something cheaper, people tend to use more of it. When you make generating code cheaper, people tend to generate more code. And that means that's a good example, for example, a good example is ATM machines. When ATM machines were first brought out, people thought in 10 years, we are going to have no bank tellers in the United States. Turns out, we actually have more bank tellers now than when they first rolled out ATMs. Why? Because it turns out that the things that people were doing, first of all, it made it so much easier to get money out that people started doing it more, they started using the banks more. And what's more, what the tellers did shifted from making change into actually creating higher value stuff. So you actually needed more tellers, it was induced demand. The third thing is that these AI tools that tend to automate the least helpful, the least exciting pieces of development. And I think that this is going to be true in just about every place. One of my rules of thumb is that if a computer can do it, the computer should do it because humans make crappy computers and computers make crappy humans. And so if it's something that can be automated, you really want the computer to be doing it because they'll do it cheaper, faster, easier and better than a human. Now, there are offsetting negative effects, but a lot of these, for example, you're needing to go back to SNP scanning to make sure that there's no accidental copyright infringement. You're also needing to run with some of these issues because hey, you're to deal with some of these issues of the copyright office, not recognizing the copyright. And we can talk about that a little bit later. But AI tools are actually amazing and they're going to be everywhere. The question is, what are we going to do about it? What's the underlying legal doctrine? Well, the good news is that we've all actually heard and dealt with these arguments before. Well, maybe not us personally, but us as in this society, the United States did. Specifically, everything that we're hearing about AI assisted art, AI assisted writing, AI assisted coding, just about every part of it was previously argued with regard to photographs. There was a great, when photographs, specifically the garotypes first, but then photographs later, when it was first invented, people insisted that photographs were not art. In The Pencil of Nature, William Henry Fox wrote, the term photography is so well known that an explanation of it is perhaps a perfluous and that some persons may still be unacquainted with the art, even by name, the discovery of it still being of a very recent date, a few words may be looked for, general explanation. And made suffices say that the plates of this work were obtained by the mere action of light upon sensitive paper. They have not been formed or depicted by optical, they have been formed and depicted by optical and chemical means alone without the aid of anyone acquainted with the art of drawing. Is needless to say that they differ in all respects and as widely as possible in their origin from the plates of the ordinary kind, which out there exists in the United skill of the artist or engraver. They said, this is a good artistic aid, this is a scientific aid, but this is completely unlike art. There was another one where, a journal called The Crayon, Rembrandt Peele said, a portrait is satisfactory insofar as the painter has sympathy with the subject. Now, in a daguerreotype, there is no such sympathy, there's no such medium and the face comes before us without passing through the human mind and brain to our apprehension. This may be the reason why a daguerreotype, however beautiful or accurate, is seldom satisfactory or agreeable. And while we acknowledge its truth as a fact, it always leaves something for the sympathies to desire. It is not art. This was so entrenched that this idea that photographs were not art, that it was actually, people started to recognize that there was some value here, though. And so they didn't want people to lose that value. They thought that this was perhaps historical value or scientific value, but in 1865-ish, they actually added the word photographs to the Copyright Act. But still, people didn't think that it was going to be copyrightable until an 1873-ish case called Burrow Giles Lithographic Company vs. Serenity. That was the first Supreme Court case to declare that photographers were authors and that their photos were protectable despite the photography mechanized process. The dispute in Serenity was about this photograph. It was a photograph of Oscar Wilde. Burrow Giles had taken this photo and had made 85,000 copies of it and sold it. And Serenity was like, can't do this. This is my copyrighted work. And so Burrow Giles made two different arguments. They first said, this is the mere output of a machine. It isn't a writing. Number two, it is completely unoriginal. Copyright is about originality. You just took a picture of what was there. And you didn't add anything to it. The Supreme Court said, well, we, they said it wasn't necessarily the photo that made it copyrightable, but it was all the other stuff around it that they did that the photographer did. They said, the plaintiff made the photo entirely from his own original mental conception to which he gave visible form by posing the said Oscar Wilde in front of the camera and selecting and arranging the costume and the draperies and the other various accessories in the photograph, arranging the subject as to present a graceful outline, arranging and disposing the light and the shade, suggesting and evoking the desired expression and from such disposition arrangement and representation made entirely by the plaintiff, he produced the picture in suit. It wasn't the picture, it was the stuff that he did before he made the picture and in around the picture of the decisions that were the copyrightable bit. But this left open the question, what about other things that didn't necessarily have all this arrangement? Well, the next word was actually in Blisztain for V. Donaldson. This case was actually, is mostly known for this idea of high art versus low art. This was actually some circus pictures like this one, this is one of the ones that was a subject in the case. And in that, they found that the advertisement could still enjoy copyright even though it wasn't quote high art. But in sort of talking about why it was still copyrightable even though it wasn't high art, he said, Justice Holmes said something, he said, it is the personal reaction of an individual upon nature. Personality always contains something unique. It expresses singularity even in handwriting and a very modest grade of art has something irreducible which is one man's alone. So in Holmes' view, he said it wasn't actually about the quality of the art. It was the fact that a person had put something in it and he said it was impossible to actually put to have any sort of art, any sort of expression that didn't have some representation of the artist in it, of the writer in it. Now, this was, this particular line of thinking was picked up in a 1922 case called Jewel of Circular Publications versus Keystone Publications. This Jewel of Circular was about a director showing photographs of like trademarks for various jewelers. But in part, there was a discussion of both serenity and blistering and in that case, Judge Hans said, brogiles left open the intimation that some photographs might not be protected. But I think that even as to these blistering rules because no photograph, however simple, can be unaffected by the personal influence of the author and no two will be absolutely alike. Moreover, this all seems to me besides the point without regard to the degree of personality because photographs are in the copyright act regardless of the degree of personality which enters into them. So they said this, any photograph also shows at least some trace of originality and that is enough for it to be copyrightable. And so this over the next 40 to 60 years, this became the rule. It was mostly, it was next discussed in extensive fashion in the Geis case. This is the case about the Zapruder film of the Kennedy assassination. People said, hey, this is news and all you did was put your camera in the right place. You weren't actually creating any art when you recorded this. And so we want to be able to use it. According to the judge's reading of the then current Copyright Act, said Congress has explicitly made photographs the subject of copyright without any limitation because no photograph can be unaffected by the personal influence of the author. And they quote a serenity, quoted blistering, quoted jeweler circular and then went to what you may not know is one of the sort of the preeminent treatises on copyright and in more on copyright. And it said, this is the prevailing view. Any photograph or almost any may claim the necessary originality of support of copyright merely by virtue of the photographer's personal choice of subject matter, angle of photograph, lighting and the determination of the precise time when the photograph is to be taken. So this takes us back to Zaria. The Zaria decision only quoted serenity. It said, hey, you didn't do enough stuff around the creation of these images. And what's more, we understand that there was actually some sort of randomness associated with the generation because it didn't come out the same way each time. Because we're going to actually decide that you didn't have enough control over what was coming out, you simply made suggestions. And so it was actually entirely the output of the machine and was not, and was not the result of any creative input on the part of the author. There are many, many problems with this. The first is, it is ridiculous on its face. If you were to look at all the different pictures and you were to say, and you were to see they had consistent characters, they had consistent lighting, they had consistent style. They had all this consistent stuff. And you were to say, this was completely random. That just doesn't even pass the smell test. It's ridiculous. But it's more, they got the legal standard completely wrong. The legal standard is not whether someone has complete control over the output. Because if you think about a photograph, do you have complete control over what's going to happen out in the field when you press the trigger on the button on a camera? No, you don't. Something can come in, a bird can come in. In fact, some of the, some very famous photographs have been created because someone was just watching and they happen to catch a really good moment. In fact, if you think about sports photography these days, what do they do? They take like thousands of photographs like it over a course of milliseconds and they're not choosing each one of those instance. Instead, they're going back and they're like, which is the instant that I happen to capture that seems like it expresses the thing that I want? Generative AI applications, anyone who's used any of these things knows that at minimum the prompt and gives some sort of minimal input. The amount of necessary creativity that you need to support copyright is bare minimum. But if I were to take a picture and I was to take a and create an identical image using mid-journey, for example. Even though I'd spend far more time and effort tailoring the mid-journey image, currently they're saying that one little snap of the photo is worthy of copyright and the thing that you created via mid-journey is not. This, as soon as this gets into to a judge and the judge is able to say, look, this is not actually the right structure. This is not falling through on the underlying result of the underlying purpose of copyright. They're going to need to be forced to change that. But that's going to require getting something through the copyright office, going through multiple rounds and perhaps all the way up to the Supreme Court before we actually get to that. That's going to be three, four, five, 10 years minimum. At that point, my personal feeling is that they will be required to change this because this idea that the coders, the musicians, the artists that will all be using AI tools that none of that stuff that they've created is actually copyrightable, that it can be copied by anyone, that is going to create such a disruption that they will be forced to change it. But that's going to be something that's coming along in ways. So what does this mean for you all? Well, like I said, a lot of the things that you think of being subject to copyright in the AI space, some of them are still copyrightable. Like if you write the code using PyTorch or TensorFlow or what have you to implement a model, your code, yes, copyrightable, you wrote it. But the outputs of it aren't. What is more, what is also sort of a, I haven't gone into it in this particular talk, but the weights associated with a particular model, there's a really good argument that the weights are not copyrightable. They're not patentable, they're not copyrightable, they're not protected actually by any type of intellectual property at all. The best you could do is maybe keep them a trade secret, but the fact is that by interacting with these models, it actually starts to describe, you're able to extract some of the weights and some of what was in there, which actually over time can lead to the destruction of the trade secret. So what have you got? You've got maybe temporary trade secret status, maybe contracts, but a lot of the stuff that y'all are doing with AI is not going to be subject to typical copyright rules right now. So if you want to build a business around it, you want to build structures around it, you need to start thinking differently about where is the place in which you are providing value because it's not going to be enough to say, oh, this is my code, this is my output, this is my whatever, because those things maybe you can control it somewhat, like I said, through contracts and agreements, but you're not going to be able to really get injunctions to say, hey, someone took my stuff, stop it, because that is specifically a creature of copyright, and that sort of thing is actually not allowed under contract. Now, there was one question right as we were starting, in the case of these outputs, some of these various organizations are saying that they will indemnify you on the output. What does that exactly mean? Well, turns out that even though, and this is another reason, by the way, that the Copyright Office is completely wrong, is that even though they say that there's no copyrightable content in the output of an LLM or an AI system, the output can nevertheless infringe someone's copyright on an input. Specifically, the way that you've gone through the training, it you memorized part of the input, and so it moves it from, you're able to identify, here's an input, a specific work that was on the input, and it was recreated in your output. Now, as part of the code, as part of a distinct part of a picture, a distinct part of text, if you've got copying from the input to the output, that's still a copy even if it's gone through the AI tool. That's still, unless one of the exceptions to Copyright applies, that is going to be an infringing output. So, when they are saying that they are going to indemnify you, mostly they're saying, well, if someone argues that one of these outputs infringes on one of their inputs because of the training of what have you, then they will argue that, but usually those are so tied up in some of the specifics, like you have to use all of their mitigation things, and you have to accept exactly as is, and a bunch of conditions that actually probably are never going to happen. If all those conditions are true, yes, they'll promise to defend you, but normally they are not actually going to be true. So, be very careful in what you're looking at. But what we're going to be doing is we're going to be getting to this point where you're going to have stuff that can possibly infringe, but that is the only possible copyright that it can have in it, at least under current doctrine. And so, if we, like I said, what's the takeaway? The takeaway is right now, if you're creating a business around this, be very under, be very careful in terms of understanding where your value comes from and where the thing that you're selling is, because the typical means of licensing is not going to work, at least right now. We've got three minutes as far as I can tell, so I'll open it up to any questions. The Chris, Chris is an immigrant and specifically an immigrant from Russia and was worried that being involved in any sort of court case would cause issues with getting or maintaining a green card and so chose not to pursue it further. So, the Zarya case is unfortunately no longer active. We're actually looking for and developing some other cases that we think will be good cases to bring up, but Zarya case is happening now. Are there any other active cases that are driving this development forward, or are you? There's one interesting case called, there was a something again out of, I forget the word, it was a stable effusion in general or mid-journey, where it was a piece of art that won an art competition in Colorado and they tried to, I forget what it's called, but they tried to copyright that and the copyright office doubled down on that and said, nope, not copyrightable. There are some other people who are trying various things. So far, the copyright office is still doubling down. It's really going to require going all the way through the court system. Yeah, which is the way it should work, in a way. It is the way it should work. I'm just, in the meantime, it causes a lot of problems. What could be the, are there any risks of sort of negative side effects affecting our work in working with generating things through AIs by it suddenly becoming copyrightable? It depends. If what you're doing is starting to, my take is probably not because what that would do is that would sort of restore people's expected, what people expect things to work in the way they expect them to work. That would be a restoration of how stuff has worked previously. So if you are dealing with things normally, then it probably wouldn't make a big difference. A place where it could cause a problem is if someone was knowingly copying a bunch of other work relying on the fact that it was declared to be public domain and then all of a sudden it's declared to be not public domain, that could be a significant issue. Right. All right. Thanks, Andrew.