EP 92: AI in E-commerce - Navigating the Legal Maze

This is the Everyday AI show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business and everyday life.

When you create content on the Internet, who should own it? Who should have access to it? We're going to talk about that and a lot more and even just e commerce and AI in general. But I'm extremely excited for today's episode. As a former journalist and as someone that's been producing content for 20 years and now that talks about AI every day, I think about this a lot. You know, should, should we be protecting our content? Who owns it? You know, will just bots be creating content regurgitated 10 times over? I don't know. So my name is Jordan Wilson. Welcome to Everyday AI. This is a daily live stream podcast and we have a free daily newsletter helping everyday people understand and use AI. So today's topic is a big one, just talking about different legal regulations when it comes to AI. We have a guest who is extremely qualified to talk about this. So we're, we're excited about that. But before we get started, as a reminder, if you are joining us live, number one, thank you. Let me know where are you joining us from? Where are you joining? I'd like to see this. You know, yesterday I think we had people from 10, 10 plus con, 10 plus countries chime in. And I'm asking that because our guest is joining us from the west coast bright and early. 5:30am if you're on the podcast, don't worry, check your show notes. A lot of great resources there and you can click and come join this conversation. We're going to be talking about this in the comments on LinkedIn, but let's get started with some AI news. So first story, there is a battle brewing around Nvidia chips and who gets access to them and who doesn't. We talked to this before on the show, but Nvidia chips, their GPUs essentially power this generative AI movement. So a recent story is showing that now Nvidia must obtain the approval from the US Government before exporting its highest performing chips to, to certain buyers in the Middle East. The end goal purportedly is to stop these high powered GPU chips that power generative AI from ending up in China. So kind of like a workaround. So, so looking at some of these countries in the Middle east, next chat GPT competitor just launched in China. So we talked about this a couple months ago when it was first announced, but the chatbot Ernie Earningbot is now Available, available globally. And it can understand both Chinese and English. So this is from two Chinese tech giants, Baidu and SenseTime, and they essentially just launched their chat GPD style bot to the public today. And it can allow users to do everything from summarize documents, get medical advice, create videos, you know, what you would normally expect or want out of an AI chatbot. And of course, chat GPT being blocked in China. This is kind of, kind of big news because, you know, we kind of just talk about, you know, a handful of large language models for the most part. Chat GPT, Google Bard, anthropic cloud, Microsoft, Bing chat. But you know, we might be talking a little bit about Ernie Bot now in the future. Last but not least and a very relevant story for today, an AI copyright battle is getting massier. So we've covered this on the show a couple of times. But now the inventor of an AI image, an AI generating image platform called Creativity Machine, he has lost a court case that essentially said he cannot copyright his images. But now Stephen Thaler is trying to prove that his bot is sentient and it can have its own independent thought, therefore it cannot be copywritten. Messy. It's messy. So speaking of legal messes, we have a great guest today. So as I bring him on, very excited to welcome to the show, Neil Peretz, the head of legal for WooCommerce, joining us from the west coast. Neil, thank you so much for joining us.

Glad to be here. In the background, I'm rereading the recent Thaler decision.

Yeah, it's gosh, I could just talk about that for probably an hour plus and all these different things. But real quick, Neil. And if you're not sure, Neil, serving as the head of legal for WooCommerce parent company automatic. So if you use WordPress, I think we said what Neil, WordPress powers like almost half of the entire Internet by now. Right.

And WooCommerce actually powers 23 plus percent of all the online stores. Wow.

Wow. So Neil has a great perspective. But real quick, Neil, just, I mean, because aside from, from your current role, serving as head is legal for WooCommerce, you have a pretty extensive background in just building and growing AI companies. Talk a little bit about that in your experience.

Sure, Dan. To be clear, it's not all generative AI, which is probably more so the topic today, and that gets people excited. I think the AI that I worked on at the other companies is probably far less publicly accessible. Yeah, one company was, it's called Alia Financial Technologies and it was one of the first ones to use artificial intelligence, which we called machine learning, was sort of a subcategory of AI to underwrite credit for consumers. And the goal was, hey, could we figure out a more precise method of determining who might actually be able to repay without blowing themselves up and without missing payments in the context of consumer credit? And if, if you could better assess this, you could offer credit to more people there because there are a lot of people out there. Our hypothesis was that they're not well represented in terms of their reality by the data that might be going into the credit bureaus, which then fed into fico. And people always ask, what's your FICO score? We came up with an ALIA score that proved to be much more accurate. And we would be. We were often able to find people who were not well valued by fico, but based on our analysis, actually could and would repay various forms of credit. And they needed credit and they were being turned down for it. And so we partnered with, among others, PNC bank, and we put out billions of dollars in credit to people that again, FICO would rate lower. And these people had great performance, actually. So our algorithms did work. But of course there was a lot of, a lot of data sources put in and a lot of tuning of our models. So I had to go around and meet with all the bank regulators and try to explain to them why they shouldn't be afraid that we're involving basically AI in our, in our credit underwriting process because there are laws that try to protect against discrimination, for example, in providing credit. And they thought, well, AI, maybe there's all sorts of secret discrimination that goes on there. So on the public policy side, the interesting aspect of that role was trying to get others in the regulatory, basically community, which could include regulatory folks working at the banks comfortable with the application of AI, which already was proving that it was yielding a better performance.

Right. And you know, you bring up some interesting points that, you know, there is a big difference, I think, even between this kind of recent generative AI boom and kind of the more, more quote unquote, traditional AI that a lot of us don't even know, but we've been probably benefiting from it for many decades, you know, but I know a lot of, even our audience, Neil, is building generative AI products right now. Given your background, kind of, what was your biggest surprise in building an AI company?

So the company I mentioned a moment ago, Leah, is a fintech company. I was not the co founder of the company, I was a senior executive at the company. But I didn't start it, and I wasn't really driving as much of the product efforts. I was more so on the legal side, trying to figure out how do we make this all legally compliant. After Alia, I went and started a company called Contract Wrangler and that I was the founder and the CEO of the company for quite some time as well. And there the goal was, hey, can I use, you know, what we call AI generally in what we call specifically machine learning, to read legal contracts and find all the key terms in those and alert people. Because what I noticed working in house as a lawyer was a lot of times people negotiate very heavily on a contract and both sides chime in. You could spend weeks, maybe months fighting about various terms. And then when people are all done, what happens? The contract ends up, like in a file cabinet or Dropbox or box or the F drive, and that's it. They're on their very way and they're off to the next thing. And the reality of what happens when you sign a contract is that's when the contract begins, not what it ends. Somehow they thought the signing the contract was the end. We're all done. There's no work to be done. When it's the opposite of that. The contract is actually designed so people will take on certain responsibilities and people have certain rights. And if you don't know your rights or you forget about your responsibilities, you get in trouble. And, you know, a simple one that comes up for many of us as consumers is auto renewables. Hey, stuff renews when you didn't want it to. You know, maybe you wanted to change. Think about some of the consumer services you might subscribe to, cable tv, cell phone. If you just noted when something was going to end an auto renew and you call the provider and said, hey, I noticed the terms coming up soon. I'm thinking about switching. Oh, we got a bargain for you, right? You'll get better prices.

Oh, that's the worst. It's the worst.

It's the simplest thing. So. But then the question is, okay, could I make a machine that goes and figures out some of these key terms and characteristics of the contracts and then notifies the right person and says, hey, you have. We'll call it an opportunity or a responsibility here because they get essentially almost free money just by calling up and, and acknowledging that there's a renewal coming up maybe, and asking for a price change or getting other bids. And there's lots of other things. You want to go find a contract. Interestingly, when you discuss contracts with non Lawyers, they'll say to you, yeah, I get it, why contracts are important, but really only three or four sections are important. And this is a widely held sentiment. The part where they differ is they all disagree on which of the three or four sections because everyone's got their own categories that they're very interested in. If you talk to the IT guys, they'll say, oh, all that matters is the uptime requirements.

Yeah.

If you talk to the finance people that say, well, all that matters is, you know, when I get paid and do I have any significant liabilities where I'm going to indemnify someone else? Or maybe the lawyers would go ask about that, what, what kind of warranty is included or not included with this kind of thing. And the salespeople would want to know, you know, when is it going to renew? So I go call them up and sell them some more stuff. Finance might want to know which of our contracts allow us to raise the support price every year as prices go up, as we're seeing now in inflationary environment. So the goal was build a machine that can go figure all this out. And auto renewals is a simpler one for us, to us consumers to understand. We actually do get indemnification of the clauses shoved in our consumer contracts, but the reality is we have less bargaining power to do something about it. Right?

Yes. So one thing, one thing I wanted to talk about, Neil, and I heard you say it there. So you said, you know, talking a little bit about your background and contract, but saying, you know, if you don't know, you said, if you don't know your rights and responsibilities, you're in trouble. And that got me thinking to where a lot of things are at currently. And this is, this is a big topic, but I've gotta, I've gotta pick your brain about it. So speaking of rights and responsibilities, you know, it's been in the news a lot, just the legal, ongoing legal battles over generative AI. So whether it's, you know, the output from chat, GPT images from companies like, you know, Mid Journey, Dolly. Right. Just what's, what's your hot take? Like what's, what's going to happen with, with all these legal battles? I know we can't, you know, it's so much gray area and it's probably going to go on for a long time. But big picture, what's going to go on in this fight over, you know, content created by these different AI systems?

Well, I think a lot of folks don't understand how they work. So that drives them to certain assumptions that probably aren't true. And those assumptions then drive their vision of who should pay who. They don't understand. At least with the generative AI models that are using transformers, they don't understand that essentially it's a probability game that they're playing as to, hey, if I type X many tokens or X many words, if I type X many words, it's going to suggest, here's what the next word should be. Actually, images is kind of a variation of that where it's what should the next pixel be, if you will, to achieve a certain goal. And that's a probabilistic guess that they have on their part. Obviously, the more they can understand what preceded the blank spot that they're supposed to fill in, you know, the better they'll do. If I said, you know, eating, and you'd fill out who knows what, you know, cereal. If I gave you the phrase in advance, it was dinner time, so he was eating. Well, then you wouldn't say cereal, right?

Oh, I would. Gosh, I. Small, small tidbit. I eat cereal. I used to eat cereal probably about two to three times a week for dinner. So my, my chatbot would be telling.

The truth on that exact. This actually gets to user generated content as well. Right? Because, you know, if we trained on all your content, it would say serial afterwards. But odds are low actually that you searched at your neighbor's houses that that would be the main item on the table. Yeah. So, you know, let's imagine you wrote a great book about your life and it was, you know, every page of the book, you described how you ate cereal for dinner. Great. That content's out there. Imagine that, that, you know, work is put out on the Internet in some publicly available manner. A GPT type engine sucked it down and it'll have, you know, one little data point that says one guy somewhere. You know, for him, we fill in the word serial when it comes when presented with the phrase what people that asked what people ate for dinner. But you know, if you went and grabbed the blogs of all of your neighbors, you'd get a far lower incidence of cereal coming out there. So it's going to say, statistically, what should I put out there? And statistics say, no, it's not cereal.

Yeah, that's a, that's a good point. That's a good point. Statistically, I guess I'm, I'm weird in that regard. Hey, as a reminder, if you're joining live, we have Neil Parrots, the head of legal for Woocommerce So Neil, like we talked about at the top of the show, you know, the parent company Automatic owns a lot of, a lot of different companies. But you know, WordPress, Tumblr, you know, so, so the parent company has so much ownership of just everything that's out there on the Internet. And again, I invite people, if you have questions for Neil, please, please get them in now so, so we can throw them up. But how it just to me, I mean should, should like is this something that you guys are talking about internally? Is this something that you wrestle with just ownership of content, given that, you know, automatic and WooCommerce and WordPress own, you know, own quote so much of the end content or I guess the users who are generating it own it. But is this just a conversation that you guys are tackling with internally and where do you think this discussion will eventually land?

Yeah, it's a conversation we have a lot internally in the legal team and then it's conversation that we have a lot externally where there are content creators who speak with us. Our mission is to empower the content creators. So in the case of WordPress.com people have content up there that they're targeting in various manners. It could be through a blog that actually maybe is co branded by us because it's our, one of our lower cost plans or could be something that, or just behind the scenes. And so if those folks want reach, which they usually do, that's why you're publishing publicly like that, then our job is to help them get reach. And part of getting reach is actually being able to be consumed by machines of various sorts. You know, if we step back, just think about search engines, right? Odds are high if you wanted your content to be found, you would like to have your page indexed by a search engine that leads it to be more likely to be found by others seeking out information on people with, I don't know, green sock fetishes or something like that, if that's what your blog is about. So you would like to be indexed by search engine. And so we have various setups where we try to facilitate people's content being consumable. And that includes being consumable by automated sources because people want that. And that historically has been how you, for example, end up being indexed on a search engine. The most likely way we have had people query is there a way to shut that off? And to some extent we don't control that because these crawlers, or think of them as scrapers that go out and gather content, they don't necessarily Knock on the door and ask and say, hey WordPress, will you go invite me in? They're out there grabbing all sorts of publicly available content. Now if they get the content through some type of machine to machine feed, well, you know, it's a little bit easier. But really they're set up to go consume content with what I don't. If you, if you ever heard the phrase headless browser, but that's essentially what they use. They're using a headless browser that by headless it means that they're not actually looking at anything visually. They're just saying, give me the HTML code and I'll decipher it myself. It's a lot easier if you could just suck down code and you'd have to wait for it to render on your screen. Don't go around and do that. Suck down code, suck down code, suck down code. But to the person running the website, the browser showed up, you don't know what the end user sees or doesn't see. If it's headless browser, just something else showed up. Now there's, there's clues what types of browsers are out there etc and you can do other things to try to manipulate your site to foil the machines if you really wanted to.

Right.

But generally if you're a publisher and your goal was to put out content, make it publicly available, that's our job is to help you do that@WordPress.com and then, you know, similarly at Woo, we, we power more stores than anyone else on the Internet. Not all of them are hosted by, by automatic by the way, because Woo is open source software as well. But we do run woocommerce.com and there are many, many stores that we host and people like folks to know that their store is out there and exists and, and can be found. Yeah, so they want to be an influencer. Correct, if you will. So generally if people's job is to make themselves publicly available, we're there to facilitate that. Now then they get this question of saying, what if people don't want to be found for some reason by machines, they're okay with, you know, being on the public Internet or you know, if they're behind a paywall of a sort where there's exclusive access, we have options and we have tools to enable people to do that, ranging from the traditional, you know, you can buy a certain tier of access on my website to actually the way even the Web three options like token gating. If folks want to get all excited about Web3, we can say fine, you can token get your website because we have the biggest partnership ecosystem of really anyone who puts out either stores for Woo or, or websites or WordPress of other plugins you can get some of which are many, many, many, many, which tens of thousands of which are available from third parties that work with us. Right. So every which way you want to slice and dice your website for access. Yeah, it's available.

Yeah.

For WordPress and Woo.

Yeah.

And that'll keep, that would keep a scraper out potentially.

Right. But you, you, you do bring up a very important point because you know, you mentioned, you know, Even just like WordPress, right. Like I've used WordPress for I don't know, 10 to 15 years. I love it, you know, but you know, even WordPress right now they, they have their, I believe it's the Jetpack AI Assistant, right, that helps you exactly, you know, generate AI kind of AI content. And you know, this goes to Andy's question here, like who in the end or is there even an answer to this right now? If you use generative AI, whether it's the, you know, WordPress jetpack assistant or chat GPT, who has rights to the output? Is there an answer?

Yeah, Andy's question is a good one. So part of what we started on the conversation of saying if I already have content, what do we do about, you know, it being scraped and, and what's, what are some WordPress considerations on that? Andy's comment is then if I, if I use, I think this question is if I use AI to generate some content, who owns that? Which actually gets into the recent failure decision and other decisions about that, which is that generally the courts have found that if, if I want to own something, one way to own is through copyright law. And copyright basically gives you the right to decide who gets to copy something.

Yeah.

Now there, there are some exceptions to it for fair use where things can be copied against your permission for certain, in certain circumstances. But you know, generally if you own the copyright on something, you can control how it's reproduced. So the recent Thaler decision was one in which someone had a machine said, my machine came up with all this cool art by itself. And the court said, well you can't file a copyright on that. And generally the, our doctrine in the United States is that copyright is something created by human. So if it was created only by a machine, without very much human input to guide it, what you're going to find is you can't copyright it. And it's essentially kind of a public domain offering out there if there was some substantial human input there, then you'd say, hey, it was kind of like, you know, I, I used the computer as a tool. And there was another recent decision I think this was on. It was a recent copyright decision about a series of comic books that were, had art in them and there were some basic prompts that were given and then the art popped out and the court decided no, that or I think it was maybe this, the copyright office decided no, you can't get a copyright on those because there wasn't really much human effort to create it versus let's say Photoshop. Right. Photoshop is using software as well, but there's a lot more guidance that the human provides and stuff that comes out of your Photoshop. Yeah, you could still go have a copyright on that. Now another type of ownership, right, you could have is trademark. With the trademark, it doesn't really matter where it was generated. It matters what use you put it to. Trademark is protecting something else. So I could have something completely computer generated, no problem. If I file a trademark on it, I can own it in that setting, the class for which my trademark applies.

Are you still running in circles trying to figure out how to actually grow.

Your business with AI?

Maybe your company has been tinkering with large language models for a year or more, but can't really get traction to find ROI on Gen8AI? Hey, this is Jordan Wilson, host of this very podcast. Companies like Adobe, Microsoft and Nvidia have partnered with us because they trust our expertise in educating the masses around generative AI to get ahead. And some of the most innovative companies in the country hire us to help with their AI strategy and to train hundreds of their employees on how to use gen AI. So whether you're looking for chat GPT training for thousands or just need help building your front end AI strategy, you can partner with us too. Just like some of the biggest companies.

In the world do.

Go to your everyday AI.com partner to get in contact with our team or you can just click on the partner section of our website. We'll help you stop running in those AI circles and help get your team ahead and build a straight path to ROI on gen.

Yeah, and it, and it sounds like, you know, there's more gray area than not, right? Because even as an example, if someone says, oh, I used, you know, a chat GPT or a generative AI tool to start this, but then I, as the human continued, because in theory, there's really not an easy way to prove what you know, I, I talk about, you Know these AI content detectors all the time that don't work. So, so I mean, is, is, is this just going to always be something that always ends up in the legal system? Since there is no definitive way to say, oh yes, this came from a generative AI system versus this was human generated, probably.

If you want to have your content be impervious to challenges, you'd keep track of the fact of your human effort that you put into it somehow. And then someone accuses you of saying that a computer just did all that, you did nothing. And who knows, maybe we'll end up with humans out there whose job is to add value by moving around some words.

I think that's SEO industry in general, right? Yeah, you go in there, you move around some words, I add a value.

Right. And so that, that'll get to the question of, from the core of, well, how much value did you add? How much value do you need to add to have, be a human work? Or is it just the, the human part of it that is the copyrightable portion?

Yeah, and I'm gonna, so we, we have a question here from Sarhash. I'm gonna, I'm gonna tweak it a little. He's kind of more asking if we create content using multiple blogs and provide with genuine references to the content, can we end up in legal issues? And again, this show isn't legal advice, but I'm going to tweak this a bit because I think it is an interesting question. Should, when people are publishing content, you know, Google has, has recently kind of released this synth ID that kind of watermarks, you know, images. But should publishing companies or even individuals, if they are using kind of copy verbatim from a generative AI, should they be saying. So, I mean, what's, what's kind of your, your thoughts on that?

Well, there's a few different questions built in there. So one is generally, if it's someone else's work and I reproduce part of it, you know, regardless whether it's AI generated or not, is that okay? And the answer is, well, of course, this is, I'm a lawyer, so I'd have to say it depends, of course. So it depends on the context in which you, you're reproducing it. But generally there's this idea that you could go reproduce or manipulate copyrighted works of others for certain purposes. And that's considered the, the term is fair use, at least in US Jurisprudence. So a famous case in this regard is, is involved Google Books pursued by the Authors Guild, and of course if you go to Google Books, there's a lot, a lot of books in there. And the way they got all the books in there is they went and grabbed all the books and scanned them all in. And one would assuredly think, yes, the books are copyrighted. Yeah, they're, they're published books that you pay money for. And in the end the court said, well, it's okay, despite the fact that they're reproducing all these books because A, you're only reproducing a snippet of the book and B, you're actually adding new value, which is you've now rendered the whole book searchable and findable by those in the Internet. Sure, that's good. That's an added value. You've essentially provide a transformative use. It's your, your use of it isn't just, oh, I could read the whole book, because you can't. Your use of it is now a content finding engine. And that really does add a lot of value for, you know, consumers of books. So in that context, yeah, they can copy entire books. But how they deployed it, it was a transformative use. It could be, you know, other transformative uses could be, you know, parody of a sort. Right. It's hard to make a parody if you can't reproduce what you're, or at least a subset of what you're parodying. So we wouldn't have parody if you could say, nope, you can't mention me. How do you make a joke about something you can't mention? Yeah, example. So smaller amounts of content. Content you're talking about. Sure. You could probably reproduce it. There's probably a fair use argument to be made. Of course, the, the specifics matter. You could either go check with a lawyer or at least just do some reading about fair use and what are the components of fair use. So you could think about that for your content. Yeah. And so that, that was, you know, generally the case with, you know, using other people's content, I think. Was there a subset of AI Question about that, though.

Yeah, you know, real, real quick here. Just as we, as we kind of wrap up, because we've, Neil, we've been top to bottom here. We've, we've talked your background, we've talked, you know, some of the AI, AI powered companies that you've built. We've talked about general copyright. But I'm gonna do this, you know, because everyone's always looking for free legal advice. I'm not gonna, you know, have you, have you give free legal advice. But what Is your stance, if someone is wondering, should I be using generative AI to produce content on the Internet? What is kind of some, some general guidance and best practices or recommendations that you might give to people so they can still be creative and put great content out for others to consume, but using generative AI tools but to still kind of, you know, not have to tiptoe that could I get sued Line. What's, what's the best practice or recommendations?

And, and actually this dovetails with the second part of the, of Saharsh's question, which was he said, you know, should I, should I note something, you know, that was some of the content might be AI generated out there. And I'll say that that is the preferred approach, at least for professional companies creating content. There are certain laws that are now starting to say if it's AI generated, you should go say so and essentially give people a warning. And that actually to some extent could protect you because, I don't know, you probably had it on prior episodes. AI isn't always correct.

Yeah. It lies.

It makes stuff up. Yeah, completely. I've had it. I've experienced it myself as a lawyer. Right. Had it suggest things that don't exist, laws that don't exist, cases that don't exist.

Oh, yeah.

So if you put in the warning that it's generated by AI and especially if it's not been fact checked by you. Yeah, that's probably a good thing for you to go put on there.

Yeah.

So, you know, that gets the question of how do I use AI? So number one is, yeah, you might want to tell people when and where you used AI. Alternatively, you need a human workflow behind the scenes that verifies things. If, if there's any type of factual assertion, for example, you'd want to go verify it. Otherwise people will assume you said it and you'll be held responsible for the misstatements of something that the AI came up with. So that's part of it for using content. Otherwise, I think you could assume that you probably can't get a copyright if AI generated a whole cloth. So if you want to get a copyright, you did part of the work, you'd want to go keep track of that and document it for yourself. And the challenge of this is, we talked about probabilities and what the machine comes up with. The machine can come up with the same result for many of the people. If I have a magic machine that makes images, there's not a guarantee that it's not going to make the same image for everyone. Else. So I don't even necessarily even have a lot of uniqueness when I rely solely on that. And some people really want that. They want something that is uniquely theirs to go protect. So you kind of integrate the AI with your other creative process. Then maybe you'll have something more unique, more ownable would be the way to go.

Neil, you've, you've helped walk us through everything from, you know, your background to how to safely create content, tackled some questions. Can't thank you enough for taking time out of your busy day and early day to join the show. So thank you. Thank you again so much for joining us.

It's fun. Thanks, Jordan. There's a lot more to talk about in any one of those topics.

There's I. Oh gosh, I know. We could go hours, right? So, hey, if you're listening on the podcast, check the show notes. If you're joining live, make sure to check out our daily newsletter that we put out each day. Neil just dished a lot of information. We're going to recap it all in the newsletter, so make sure to go to your everydayai.com Sign up for that. Neil, thank you again for joining us. And we hope to see everyone back again for another episode of Everyday AI.

Thanks, y'. All. Thanks, Jordan. Love the newsletter.

Thanks.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going for a little more AI magic. Visit youreverydayai.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.