---

Video Transcript

Auto-generated transcript. Timestamps are approximate.

[0:00]

What’s the most common legal mistake you see that founders and CEOs are making? >> This idea that they don’t need to engage with regulatory council and folks that are really monitoring. >> So you might think you’re a SAS company, but you’ve become a last company. Liability as a service. >> Most people use chat GPT at this point and they’re thinking everything that comes out of that model is accurate. Who’s liable if it’s wrong? >> It’s a tool, not a creature. We’re not perfect. The technology won’t be perfect and we shouldn’t let the perfect be the enemy of the good. the fact that data is everything right now, but if the wrong data is in there, people are going to get denied. >> If they’re doing a violation that’s that’s broad enough and it affects individuals, then you’re looking at a potential class action. >> When we start relying on AI, specifically when we think about in the context of intellectual property, it definitely creates a slippery slope in terms of accountability. Air Canada, it had a chatbot that gave somebody a different refund policy. And Air Canada tried to make the argument that they weren’t responsible for what the chatbot said and the judge said no. >> And if you’re trying to actually be strategic and sell your company, they’re always going to come back and say,

“Where’d you get the data?” >> In the due diligence process, if you’ve cut corners, the acquiring company will find it. >> What’s more dangerous, bad data or overconfident founders? >> An overconfident founder with bad data, >> right? I think the overconfidence probably comes from the bad data. But I think you know the technology is I always say is only as smart as the people that are developing it. >> Humans when they do work they have an a percentage of error. Why do we believe that AI is supposed to be perfect? Welcome to the AI CEO podcast where the future of business is built today. I’m Sim Alexander, founder of Virgin AI, growth strategist, co-chair of DC startup and tech week and your host. In this podcast, I sit down with CEOs, industry experts, founders, and investors to tackle the big, pressing questions on your mind about AI. No hype, no jargon, just insights you need to stay ahead. Now, let’s dive in. [Music] The AI CEO podcast is brought to you by Virgin AI, helping midsize businesses accelerate AI adoption safely. Ready to

lead with AI? Visit Vgent.ai to learn more. Welcome to today’s episode of the AI CEO podcast. My name is Sema Alexander and I am your host. Today I am switching it up for you guys. We have a panel of two guests that I’ve known actually separately. They don’t know each other well, aside from the prep that we had for this call, but they’re two lawyers, two incredible beings who really are passionate around AI advocacy, the next phase of business and AI. First I have Amber Smith who’s the founder of Amber Smith Law. Um and then we have Cameron Powell who’s the founding partner of Deep Law. Both of them are big ecosystem folks in in this area. Um I met both of them through uh actually Amber you and I met through the startup and tech week and you’ve actually spoken there and moderated panels and I mean I’ve just uh really appreciated a lot of your insights in the space and Cameron Powell and I met at the Generative AI collective through the work Katherine McMillan is doing uh

which has been an very amazing robust community that’s focused on AI business and humanity and I just really when you and I chatted it was very apparent that you’ve been in the space very deep in emerging tech and startups with your career and some of the insights that you shared. I was like, you know what, this is we need this conversation badly on this podcast. Legal dynamics and AI are something that is an ongoing uh conversation especially as technology continues to emerge in the AI space. So really excited to have you both on this podcast and start this discussion. So, um, I’m going to start off by asking a question and then, you know, you guys can take a second to to maybe just give a brief minute or two overview of yourselves if that works. Uh, so I know both of you are focused on both sides, I guess, of the battlefield, IP and innovation when it comes to this space. So, what’s the most common, I would say, legal mistake you see that founders and CEOs are making today when integrating

AI? Ladies first. I mean, I’m just going to say, >> okay. All right. Well, thanks so much for having me, Sema. Um, as you mentioned, we’ve had a chance to meet through this DC DMV eco innovation ecosystem, and it really has been a pleasure to see how you have particularly helped to ring the alarm around artificial intelligence and and responsible innovation, which I think is something that I’m absolutely a proponent of. um as the founder of Amber Smith Law um or in in totality the law office of attorney Amber Smith plc I serve creative startups nonprofits across industries primarily in the sports entertainment fashion tech and wellness industries um and so in this way I really work with a lot of clients who are tech enabled and so with the advent of an artificial intelligence it wasn’t so much about helping them build AI powered tools if you will that other companies would incorporate but really figuring out how to responsibly incorporate artificial intelligence into what they have already developed. Um but then of course I do work with some folks who have gone through the tech stars right are preede are looking for investment in building with AI. Um and in this way I would say perhaps one of

[5:17]

the biggest mistakes right is this idea that even in the absence of certain guard rails here in America that they don’t need to engage with regulatory council and folks that are really monitoring the the global landscape when it comes to artificial intelligence regulation. Um, and so earlier this year I had a chance to travel to the UK, participate in some conferences that specifically speak to the intersections of IP and AI. Um, and so really just making sure that my clients are aware that, you know, whereas we may be behind in certain respects here in America, that the globe is very much concerned with how we’re protecting data, how intellectual property um, is being developed with the use of these tools. And so really just making sure they’re engaging with council early and often around these issues um to be able to predict where regulations are going here in the US. >> Yeah, I this is something that I think we’re going to get very deep in into this conversation because people just they’re they’re asking for guidance and we don’t have it here in the states a lot of times and then there’s you know state versus actual federal versus what the administration is saying. So I think that is absolutely what a lot of people are saying is look at what Europe is

doing. Look at what other you know other parts of the world is doing because you want to be more proactive than reactive in a lot of this. So looking forward to going deeper in that discussion. Amber welcome Cameron. How about you? >> Well I uh I started out my professional career as a lawyer and then I took a couple decades off of that to get into tech startups and uh I was in marketing communications. um CEO of a a healthcare diagnostic company. So I I know the imperatives of the minimum viable product, product development, the the the policy choices and the trade-offs you make. Um and you know the what interested me so much about AI was all of the policy implications, all the people implications. Um the fact that because it’s human mimicking, >> it can infect every single area of an organization whether you know it or not. >> Right? And so when you ask what what’s the single biggest mistake that I see when I think about the the background that I’ve seen of inside companies, it’s simply that a company that doesn’t get

legal vetting of its uh intended AI projects could well instead of be instead of automating innovation, it could be automating liability. >> So you might think you’re a SAS company, but you’ve become a last company. You’re you’re liability as a service. Um and and you don’t even know it because again it touches every area. You could have employees and contractors importing it into the building, right? I mean people ask if if is is AI like the new fire and I say yes and it’s already inside the building. >> I love that SAS company instead of a SAS company or a last company liability as a service. Wow, that’s a good line right there. You know, one of the biggest questions we get at Virgin and just in general is, you know, in three years I want to adopt AI. I know it’s super important, all the things, but in three years I don’t want to be in the news. And so, both of you have just shared, you know, the some of the starting points of what a playbook needs to look like for those individuals who really understand that this is emerging. There are implications. You know, there are

things that we need to be doing responsibly, but we want to innovate. This is also one of the biggest call it a gold rush but for the right for the right reasons innovation can drive things that we’ve never imagined before right so um before we get started into the deeper conversation of AI and legal uh and business you know one of the things Amber you and I were chatting about this uh I thought was very important and I think it’s a great conversation to have because a lot of legal firms themselves are figuring out AI right you are in a highly regulated environment uh but for those who are trying to guide CEOs and founders. There are many legal firms that haven’t even figured out their own AI adoption strategy, right? Which then leads you to believe like, well, if you know, if you guys aren’t using it, haven’t figured it out, how can you guide me? You know, and I know it’s a chicken or the egg thing and obviously, you know, different industries have different implications, but can you guys speak to how you are using it? You know, we have one little caveat there. You know, we have seen in the news how lawyers are starting to get disbarred because they’re trusting in

AI, LLMs, you know, large language models like chat GPT and cases are coming out of the blue that they’re using for research, but it’s not true. All right. And so, you know, in the vein of not being able to trust everything, uh maybe Cameron, we’ll start with you kind of share how you’re using AI today and you know, any thoughts around that? Yeah. Well, I you know, for me it’s it’s a natural fit because um for a long time I’ve essentially been a writer. I mean, that’s fundamentally my my temperament, my identity, and that involves lots of revision. Um and AI calls upon everybody now to have the same kind of skill set that you’re now a proofreader, you’re an editor, you’re a reviser. And so, when you ask how I’m using it, I mean, it’s carefully. My my first job as a as a young lawyer, I worked for a federal judge who did not read the opinions we wrote for him before he signed them. >> Wow. >> Now, I helped him get off the bench after that, but it was a real lesson in how not to be a public servant, but also how not to abdicate your responsibility. Now, so there’s different ways that attorneys can use the, you know,

[10:31]

genative AI in particular. There’s legal research, there’s contract drafting, there’s summarization. I started with uh I I suppose the legal research side using some of the the more uh retrieval augmented you know type type systems like CA case text and co-consel >> that are not they’re not doing any hallucinating >> they are finding you good sites right um I then became an adviser to a startup that got into the dis discovery space >> and what’s really a gamecher there and I used this in a in a big trial last year where we won an very unusual case against the government. Um, the government dumps 700,000 documents on us, many of them having several pages each, right? >> An individual cannot afford to have his lawyers go through all those documents. You can. And so, you either have to triage and say, “Let’s look at the ones with just his name in them.” One problem. If you’re um in subject of a conspiracy trial, it doesn’t matter if your name’s on it. If you’re found to be a conspirator, you’re liable for what these guys were doing whether you knew

about it or not, right? So we used in the company is Discover AI. Again, as a disclosure, I’m I’m a I’m an adviser. But it wasn’t just that we’re using it as a discovery tool. But when you’re using generative AI to look through a corpus of documents, you can now do something we could never do before, which is have a conversation with the documents. You can ask it questions, you can strategize, you can be creative. Compare that to discovery the way it has been done, where the vast majority of documents in a in a case are irrelevant. So guess what happens to the lawyer, right, trying to go through it. >> They’re stuck reading everything and wasting a lot of time when they can be focused on things that matter. >> That’s hugely boring, right? If you talk about about employee engagement, >> think about a security guard guarding something where nothing ever happens, right? >> That’s a lawyer on discovery. >> Yes. >> Almost everything is irrelevant. So it’s just boring. You’re like, “Oh, that’s not engaging and that’s not engaging and so forth.” We were pulling up, we had witnesses on the stand from the government in our in our trial and they’re saying things and I’m typing in find me documents that

refute this statement. There’s no way I could have found that with a keyword tool. >> Hold me one second. I want to pause there. So is that allowed today to have an LLM I guess or a rag model or you know retrieval augmentation sort of model that has this context in it the 700page document. Let’s just use that as a particular example to go with what you were doing. while you’re on trial, you can research that document when someone is on on the stand. >> Yeah, >> you can do that. >> I mean, I don’t think any court’s spoken to it. Um, there’s there’s not really a risk. I mean, the the the tool brings up the documents that it think answer the questions and you have to click on them and confirm. But >> at that point, you may grab your laptop and go up to the lectur and start cross-examining and saying, “Well, sir, in this document, didn’t you say this?” >> Correct. And I’m sure that even in those cases, they might even be able to prompt you with the right questions to ask, right? like you know I’m just trying to get to like the use fullon use case here because you just shared things that a lot of lawyers would not do. >> So I guess and I want to ask this to you and then Amber we’ll move to you. >> Why did you get so comfortable so

quickly? What are the guardrails that you’ve seen put into these products? And I know you also have the startup hat on so you understand some of this stuff better than most, right? But what was that like uh moment in time for you like I trust this. that says I can know that this is mitigating the risk. Well, I think it was having a more comfort with with technology and and and what I mean by that is not just an interface like how to get how to find your way around an interface, but how are how’s technology developed? You know, what are the what’s the the product development process? And so, what can you expect to find? And I had been involved in the process of building this tool and I’d been testing it and giving them feedback and saying, “Hey, this is wrong and whatever.” It’s it’s sometimes it gives an answer and the answer is like, “Well, that may not not be correct.” But the way you find that it’s not correct is you can click. I said, “We need you need to give me a link.” So I can click and I can confirm what it just said. >> I love it. >> And so once I know that I can confirm and it’s going to pull up a document in the database, not some random thing on the web, I I know that it’s trustworthy. >> So the guardrail was validation

>> and traceability in the documents directly. That’s great. Amber, I’m going to move to you. Can you kind of share your experience and perspective on this? >> No, absolutely. And you know, I echo a lot of what um Cameron shared, but I think he specifically pointed to this idea of abdicating your responsibility, right? And I think when we think about um the law from an industry perspective, this is a self-governing industry, right? When when lawyers are brought up on ethical violations, they go before a panel of other attorneys, right? And so I think there’s this heightened sense of awareness in that we have to regulate ourselves in the way in which we’re utilizing these tools, especially to the extent that we’re advising others on how to do so responsibly. Um, and so for me, I think that some of the use cases he mentioned, right, supporting us in doing research, but also recognizing that, you know, to the extent that you’re using tools that are reliable, they should be providing citations and you still have a responsibility as that supervising attorney to go and doublech checkck and make sure that the facts of of the case that they’re citing are relevant and actually comparable to the matter that you’re dealing with. Right? Those are

[15:47]

the responsibilities of an attorney that no technology is going to absolve us of. And so I would say I use AI almost like a law clerk, right? I’m the senior attorney. I’m the final I’m I’m the one who has put the rubber stamp on it before it goes out the door. So nothing that leaves my desk isn’t going to be subject to the same type of scrutiny that I would have been subjected to at my when I you know when I was at a big law firm straight out of law school. Other use cases that are low risk. I think that’s also important, right? When we think about AI, there are varying levels of risk in the tools that you’re using. So closed source tools are great, right? ones that you know aren’t going to allow for whatever output that you’re developing to then be utilized by the entire public right the things that I that we’re generating with some of the tools he mentioned co-consel right case text or in my case Lexus plus AI right which are very reputable legal tools are only going to be available to others who are paying for these services right who have already verified that these are you know reliable um legal databases that this information is being drawn from. So would would you say that that a lot of

these tools are already vetted because they’ve been in the market. You already know that they’re trusted by other lawyers. So that makes it almost like a a validation point for you. >> No, absolutely. Absolutely. I would say that the the first tool that I felt comfortable using were ones that that were developed from software, right, and organizations that I was already um invested in, right? So even the website that I host my my firm’s website from Wix, right? These different platforms, they’re developing AI tools. So you can imagine I’m a little you should feel more comfortable using the tools that you’re already paying for access to um that are affiliated with these um reputable organizations. So I would say outside the legal research context, marketing, right? That’s something that I think all attorneys, especially those of us who have smaller boutique firms, right? Those are that’s lowhanging fruit to help them help you with copy editing, right? And and reviewing, proofreading, uh rewarding something to be more persuasive, right? Those are things that I think we should feel comfortable using them for. And so I think these instances of attorneys who are citing cases that don’t actually exist, those are outliers, I would argue, um in the grand scheme of things. But on the other end

of the spectrum, there are firms that are telling young attorneys coming straight out of law school, “We don’t use AI in any way.” Meanwhile, they’re going home and using chat GPT on their personal computers, right? >> It’s called shadow AI. It’s going to happen anyway. You cannot tell anybody not to use AI at this time. Obviously, there’s rules and regulations that can be added to it, but so to that end though, you know, you both come from boutique firms. What do you think the difference is when you have a larger firm? Because this also relates to highly regulated, you know, larger environments when it comes to business in general, right? So, it’s one thing when you’re kind of in control of your situation when it’s smaller. There’s another thing when you have hundreds of people working for you. And so, is there a difference? We’ll just keep it in the legal context for now until we get deeper into the business side of things. Is there a difference for the larger firms to think about these things versus the boutique? >> I think so. I mean, I you know, they they typically have much more uh national even international reputations. So, the the cost of a misstep could be could be even bigger. Um you know, a small firm can often be prioritized by

clients who say, “I like the fact that you’re nimble, you’re agile, you’re using lean methodologies, um you know, you you’re responsive and so forth.” um the larger firm, their clients often choose them exactly because they’re so stable and they’re slowmoving and they’re oified and they’re hidebound and you know it’s it’s like watching the government at work. Uh sometimes it’s just everything takes an act of Congress and and and at that size that’s probably not a bad idea. >> Yeah. >> I would say probably the biggest um distinction would probably be the fact they have other departments that include non- lawyers, right? So you don’t have folks that have sworn to an oath in terms of how they’re going to regulate themselves in their business in the in the way in which they operate. And so just making sure that they have policies and procedures in place for those individuals, I think it’s probably even more important. But of course, to the extent that I’m outsourcing some of those things, I still need to be cognizant of what the contractors I’m working with um are using in terms of policies and procedures around AI. >> Awesome. I think that could be a whole podcast in itself because I’ve just heard so many varying things, but both of you brought up some really great

points here. So, I want to transition to this whole AI boom, right? And the reality is there’s lots of pressures that CEOs and founders are now feeling more than ever before because this is not just an AI, you know, boom. There is there’s quantum and robotics and, you know, blockchain and multiple different technology shifts at one time. But obviously with AI, it’s probably the biggest moment in time for companies to look at it from an innovative landscape, also from an efficiency landscape, knowing that it’s the first technology that’s going to impact both blue collar and white collar workers. So lots of boards, lots of emerging tech startups are in that pressure moment of how do we use AI? So there’s this fast pace of like we need to get something out, you know, from automation to maybe enhancements in product or whatever it may be. what is sort of some of the uh challenges that they may be having that may overlooked they may be overlooked by legal right now because they’re just trying to go >> well you know it starts with the data um

[21:02]

I mean certainly the you choosing the use case is is the first the first barrier right you may be just choosing a a really bad use case but assuming that that it’s valid it’s good business school goal and you have the data for it and then the question becomes where are you getting the data um is it clean is it unbiased is it anonymized? Do you have permission to use it? >> And if all that’s in place, and again, this is why I think you have to have a a legal vetting, then is the output um reliable? Um is it is it defamatory? Is it making promises binding you to contracts in a chatbot? Right? So, so you really have to understand the the output. And then there’s a human element of oversight. The the whole idea that that um AI is going to replace people. people often don’t think about well whether there’s a person doing that job or an AI doing that job it still has to be supervised you know you you you can’t just let it run run a muck right so um we still have to have humans in the loop particularly when there are important decisions at hand like rights access you know health care housing jobs

discrimination things like that you you’ve got to have humans in the loop >> right >> it’s so funny we talk a lot about that word humans in the loop. Most people who are not in this space are like what do you mean humans are going to be in the loop right? They don’t that concept is very specific to the people who understand where this industry where this technology is headed and obviously I think most people in this community understand that now you know but you know there will be a time you know where humans may not be in the loop and and that’s another part of our legal discussion here when is it okay and validated to have things in automation side but what I’m hearing from you Cameron is that you know at a certain point that analysis whatever comes out of that AI needs that validation with a human so that that mitigates the risk >> and particularly I mean you know it’s it’s it gets to the risk levels you know if it’s if it’s affecting rights access safety like this is the EU regime and it it makes sense even if we haven’t adopted it I think courts will adopt it that’s where you need to have the most due process the most transparency the

most explanability explainability which is I can explain how our AI made this decision >> that’s the core of due process I don’t think courts are going to jettison in that just because AI says we’re a black box. We can’t do it. That’s that’s always going to be part of it, right? You you’ve got to be able to explain what you’re doing. >> Yeah. >> Uh can I ask you something specifically? Look, Chad GPT or any LLM now, like you know, you can put in some information. It may not be right. And most people, normal people, business people, right? Sometimes they think it’s the gold standard. And so they’re going to use, you know, I know there’s like different models and even that is not a very common known fact. Like there’s perplexity in Gemini in Claude and Chad GPT. Most people use chat GPT at this point and they’re thinking everything that comes out of that model is accurate. They just don’t know what they don’t know. So if they use that and put it into context into one of their products or services or solutions thinking that this was accurate because Chachi PT said so it used to be like Google said so right

>> who’s liable if it’s wrong >> no I think you know I think this is especially important to understand I think and it speaks to this idea that folks are generating liability for themselves to the extent that they’re relying too heavily on what is being produced by large language models especially those open- source languages language models that might be scraping the internet for their information, right? There’s not a lot of factchecking. There may be no citations associated with the the actual output and as a result, you know, they’re relying on hallucinations. That’s that’s the terminology in terms of of what happens when AI produces something that is factually incorrect, right, but otherwise um is convincing, right, to the lay person. And so I think it is about making sure that they are that they are fact-checking that they are um ensuring that this isn’t merely a hallucination that they’re relying on. But then I think specifically when we think about it in the context of intellectual property, right? If you’re taking it a step further, not only relying on it, but you’re then seeking some sort of legal protection, right? Whether it’s a trademark, whether it’s a copyright, a patent, or whether you’re seeking to protect it as some form of a

trade secret, right? how legitimate those rights are stems from the amount of human creativity, right, that was involved in the development. And so when we start relying on AI, we’re now having conversations around agentic AI, right? These autonomous models or what have you, you know, you’re really putting yourself in a predicament where you’re developing things that won’t be eligible for protection in a way that makes them valuable, that allows you to license them and monetize them long term. And so, um, I think that’s one of the major concerns that I’m seeing. >> Yeah. But is there a case then to be made? So there’s there’s the use of open- source like to your point, right? These are models that have billions of inferences inside and you know for most AIs if they can’t answer they’re going to answer it anyway and sometimes that’s where the inaccuracies do come. But in terms of fine-tuning models and creating your own closed source using your own context and content that’s more uh is that more of a a way to mitigate risk is that way more accurate that they can use like you know

[26:20]

I’m trying to find that fine balance for these for these founders and CEOs and business leaders. Cameron maybe you can take that one. Well, the internal um internally sourced AI, I mean, it it doesn’t have as much noise from the outside. So, it really has the potential to be more accurate. It doesn’t remove your your duty of care to speak in the in the language of negligence to to check the outputs to make sure they’re reliable to monitor it o over time, right? AI, like a lot of people developing products in general, suffers from Dunning Krueger syndrome, right? This overconfidence. And you asked, who’s going to be liable if I publish something? It’s me on borrowed confidence, right? I don’t have that confidence. Um, it’s it’s AI has this brever sort of uh pro style that makes it seem really like gosh that’s really persuasive. But it’s a trick, right? I mean, any any really good writer can make you believe just just by the the incantatory pros they use. And AI is an above average writer. >> It’s not original. I don’t want to read it at length because there’s no distinctive voice there. When I talk to people about about output and and what

what what what Amber’s saying is exactly true. Um the copyright act refers to an author. Courts have interpreted that to mean it has to be a person. >> Similarly, the the patent uh statute refers to an inventor. >> Okay? The courts have said that means a person. Trademark’s different. Trademark doesn’t care who wrote it. Um you make you need to make sure you’re not infringing somebody else. you need to make sure that it’s unique in various ways, but it doesn’t care who who came up with it. But I think the analogy that that I would that I tell lay people to take away is what I call the red line test. And this is why sticking your head in the sand and having a policy of telling your employees or your contractors, don’t use AI. That’s about as effective as telling teenagers not to date. >> Yeah. >> Right. With the same kind of problems that can >> I’m talking to my daughter right now. She better not be dating. I’m just saying Serena. No, I’m just kidding. stay away. >> You you can you can tell people not to use it. First of all, they may be using it without knowing they’re using it.

They’re using a co-pilot on their home computer or whatever. >> And they’ve been using it since Amazon days and other like, you know, they in different ways, you know, but yes. >> So, you know, in a in a contractor, you can’t be controlling what they’re doing. The safer route is to say, “We know you’re going to use it just like we know you’re going to date. Here are some ground rules.” One is when you give me your final product of the blog post, let’s say, or of the memo of the white paper, I want you to show me the original draft you got from chat GPT. In other words, it’s okay to use it, but I want to see the original. And so I don’t have to do as much work. I want you to do a red line >> between the original and your final because the only thing we own copyright on, >> roughly speaking, is the red lines. >> What did you add to it? What did you move around? What thinking did you put in there? what’s your unique perspectives, your thought leadership that is not something that you can just get by prompting AI >> and and you know we’ve got a whole generation of of people in school now that are turning in AI generated content let’s say and that’s probably the appropriate word maybe slop and you know there the the novelist Don Dilo once

said I often don’t even know what I’m thinking about a thing until I start writing about it writing is a concentrated form of thinking so if They’re skipping that whole part of revising new ideas, going for a walk, having an epiphany, going in the shower, having an aha moment, coming back, revising. That’s the product that we like to read, right? That’s what makes us go, “Oh, wow. There’s some there’s some real thinking that went on here.” >> Yeah. >> So, if you’re using the first draft, you’re you’re limiting copyrights. You’re you’re not developing your own distinctive voice. You’re not learning how to write and reason and do the concentrated thinking that readers want to read. And I would say you’re probably going to have dementia sooner in life than prior generations because we’re outsourcing too much of critical thinking. >> I have so much to ask you in that and get you all’s perspective, but I want to start with the legal part of that conversation. Let’s just use a case in point as a book, an author of a book, right? If they’re using AI today, uh, you know, to help support them in writing the book, obviously they have to

have some perspective or uniqueness because, you know, they’re writing a book. Hopefully not everybody does, right? But say they do. Is that something that they can um legally say is their book? Do they need to say that they’ve been using AI to help support it? What’s the transparency law right now? >> Well, the copyright office does expect you to disclose things that were generated by AI. And this has come up a lot in the image context where people have applied for example to um for a graphic novel to get a copyright, but they’ve said these images I generated with with chat GPT or Midjourney or whatever. >> But that’s theirs. I mean, I know in the rules of chat GPT and and MidJourney, if I’m prompting an image, it’s my license. Is that correct? Well, it depends on the terms of service with the vendor you’re using, but the the larger issue is it you might own it in a sense, but the copyright office won’t let you copyright that image. So, if someone does the same image in their graphic novel, there’s nothing you can do about it. Now, there may eventually be some protection of really elaborate prompts, right? But we

[31:39]

know that since since every prompt generates a completely different image anyway, uh it might only give you limited protection. or someone could reverse engineer your prompt by saying, “Here’s an image that I’d like to steal. Generate a prompt that would generate this, and I know that when I put it in there, it’s going to make it’s going to randomize it a bit, and it’s not going to be too close anyway.” >> Yeah. >> Right. >> Amber, I want to hear your perspective on this, too. But like one of the things that I know this next generation is going to face, and this is just goes back to the critical thinking and sort of the the evolution of where we’re headed because of this new these new toolings. So, I had um Peter Swain on. He’s an AI futurist and he was telling me about his daughter who’s like 9 years old and he does lots of experiments with AI and he asked he showed her how to use MIT Journey right and then he was like I just wanted to show her and then let her do her thing and she did her thing and then she came up to him and said hey daddy look what I created >> just that word I right and he’s like you know what that was her ownership because she prompted the image in her eyes she

created that image It’s the new way that people are going to think about art or you know you know and so there is this fine line and this is you know like we could talk about this for hours with critical thinking and where you know the next phase of how people are going to learn and will they get dementia or will their brain open up to other things because now they have these tools like we did with the calculator that allowed us to do other things. So well I I I wanted to share that because these are the nuances. We don’t know. We’re all kind of assuming what’s going to happen. We have ideas from the past and what’s happened. But Amber, anything on your side on this patent on the copyright side and that’s just one like prompting and genai is just one concept. I want to go deeper in innovation, but let’s start there with your perspectives. >> Yeah. No, I think this is extremely important to discuss, right? again for the reasons I mentioned and and and Cameron speaks to this idea that within the laws are embedded these concepts of inventors and authors but even with the trademarking where you don’t necessarily need a human to create it. You can imagine if the tool that you’re using to

develop some sort of a logo slogan or name is scraping others copyrighted materials, right? Then that output is now infringing upon someone else’s intellectual property. And so the trademark office is great at, you know, uh, refusing an application because there’s a likelihood of confusion with other existing marks, right, or other aspects that demonstrate some some level of unfair competition. And so I think one, it’s it’s parents job, right, to help instill within their children this concept that this is an AI assisted, you know, creation that you may have developed and you deserve some level of applause perhaps for the creativity in terms of how you prompted this. But in the absence of law of a legal framework that would allow for you to protect that, it’s disingenuous to say that you completely created especially if you’re not sure on the data right that was used in the development of that AI model that you’re that you’re generating content from. So I think that’s that’s a major concern. >> Again, I always say innovation has implications and I I know again ChatBT has been accused of using things that were scraped that they weren’t supposed to be scraping, but that goes back to my

point of like I wouldn’t know that. Of course, I do just because I’m in the space, but most people are going to be like, well, how why is it my responsibility the way Chad GPT got their information? I’m just prompting it because this is a tool that that is allowing me to do so. >> And so, I think there’s this goes so much deeper, right? And and I know this is like to me like a legal I don’t know conundrum like you know I’m glad we have people like you guys to figure it out for us but in their interim there’s going to be a lot of people that there’s there’s a lot of gray lines >> there’s an awareness gap I mean if we think if uh >> if we don’t know that how the LLMs are getting their text or how the the the image generators are getting their images then we’re not really thinking about oh um maybe I’m infringing in in my output here somehow. Another thing that we do is we assume that it’s fair, right? We don’t we don’t think about bias. We just sort of assume it wouldn’t be biased. But we’re already seeing, you know, mortgage uh scoring apps are are systematically, you know, biased against African-Americans, right? Amazon tried

to build a a resume tool that was um penalizing resumes that said women’s chess club on them, you know, that kind of thing. So there’s there’s not much way to get around that except through a gradual process of education of what it is and what it isn’t. And one of the things that that really struck me year and a half or so ago is Jiren Laneir, the futurist. Um I think it was he who who said one of the key things that users of a of generative AI in particular have to remember is that it’s a tool, not a creature. It it doesn’t understand what you’re saying. it doesn’t get you. Um, it doesn’t understand what it’s saying. And um, that that not knowing that it doesn’t know stuff, but rather is doing an extremely sophisticated form of next word, next idea, next paragraph, prediction. It understands the structure of a memo. It understands the structure of a legal brief. I’ve I’ve had chat GPT uh, write me some very sophisticated

[36:56]

legal briefs. Here’s a complaint. Here’s a very long prompt. Write a motion to dismiss. And it it’s an astonishing output that puts the billable hour at risk. And that’s that’s going to make Weslaw and Lexus uh make a make a run for their money. >> A run for their money. >> Well, which is going to happen in so many industries, but absolutely. >> But it, you know, you still have to understand, I mean, this is why in the early days in particular, it couldn’t it couldn’t add and multiply properly, >> right? you know, it it could do certain things. >> Natural language processing. It wasn’t meant to be a calculator. >> Yeah. I mean, there were certain areas where it was accurate because it had more examples. Like it could multiply things by pi really well. If you asked for something to multiply by 3.14, it was really accurate. >> But with anything else, it was just sort of guessing in a sense because it’s doing this next word prediction, not mathing. And the math wasn’t mathing. So I I think realizing the limitations of it, I mean, this is why I talk about what I do as AI use and misuse law. If if you don’t understand the limitations,

you’ll ask it to do things it can’t do. And that’s how you get into not just legal trouble. It could be ethical trouble. It could be reputational damage to yourself. >> Yeah. I’m going to before we get in innovation, let’s sit on this one for a second and go deeper with the misuse of AI, right? With the data not being good. I mean we’ve learned from most cycles that especially I mean you for different underserved communities for data not having enough data for example I think Amber I was telling you this I you know when when it comes to medical data and specifically the ranges that when we go to doctors and we see you know your cholesterol should be here your you know whatever your lipid should be at this place a lot of that data is from European white men it doesn’t have black and brown data I know a lot of people working on that especially with the world of AI and quantum to be able to facilitate you know um bring those numbers to make them whole right for the right communities so because we know and

then you mentioned the mortgage piece like that’s redlinining and now how do we avoid that cycle again and I would love your perspectives on this this is not an easy one this goes back to education it goes back to value systems it goes back to it so much. So, Amber, maybe you can start with that. >> Yeah. No. Um, this is absolutely, I think, in a lot of ways, you know, the million-dollar question, right? How do we uh avoid or mitigate discrimination, quite frankly, um, as we’re developing these tools? And my background is in international human rights prior to uh, venturing into commercial finance and banking and corporate law. And so, I’m constantly thinking about marginalized communities um, even the environmental cost of AI, right? That’s another conversation to be had, right? these tools are utilizing lots of resources, right, to generate these, you know, miraculous outputs, right? And so there’s that environmental and human cost from that perspective. Uh but then in terms of the data itself, you know, we’re thinking about these more highly regulated industries, you know, we we

have to we all are being challenged to put on the head of data scientists in this moment. And so within those specific industries, right, those folks need to be in healthcare, in housing, in banking, right? they need to be committed to making sure that they’re not simply utilizing data that has been used by by you know older technologies or what have you that have reproduced some of these biases and so >> I think whether that looks like better data collection right and going out and collecting more information now right utilizing various resources to you know to ensure that you have cleaner data sets is going to be more and more important than ever >> I I like how you mentioned that we’re all data scientists now. But I want my community to understand that it’s not when I say it’s not that complicated. It’s not that simple either. But what I’m trying to say is just be smart, right? It’s like bad data in is bad data out. You have to think about these things now. So when you’re creating any sort of innovation, automation, right? Like sit back and say

where did this data come from? A lot of companies, especially if they’re bigger, already know that their data internally is screwed up because their systems aren’t talking. That’s just the fundamental stuff. But what is missing and what is the implications of that? How can we be more responsible? Is there syndicated data to help support the gaps that we might have in this data? Because in the world and and Cameron, you brought this up early. I I talk a lot about this too around, you know, in the world of AI, the fact that data is everything right now because data is what’s going to drive the insights. Data is what’s going to drive personalization. data is what’s going to drive precision. That could be in decision- making. So, to your point, yeah, you know what, they’re approved for a mortgage and they’re not. It could take a few months right now to get all the paperwork and all that stuff. Eventually, between blockchain and AI, that decision could be made in two minutes or less, right? But if the wrong data is in there, people are again, same situation as per previous are going to get denied, right? And so I think of as a CEO, founder, business leader,

[42:11]

innovator, be smart. These these concepts, data science, obviously it’s a very complicated field. I’m not, you know, I’m not negating that, but the concept of just how should we be thinking about data? There’s some fundamental truths and that’s kind of where I wanted to go with your comment, which I loved. Cameron, any thoughts on that on your side? Yeah, I would say do not ignore all of the research on what having a diversity of thought and experience can do for a team, a board, a company. In all of those situations, a people who come from different approaches, different backgrounds, different experiences, you know, even in the invention context, you get a physicist, a doctor, a lawyer, a teacher, a welder in the same room. their patent ideas will be better at the end of the day, right? Intellectual ventures has has shown that sort of model. So, and it it it it doesn’t have to be uh something that um uh discriminates against anybody

to have diverse thought and experience. I mean, I view myself as having diverse thought and experience and it’s it’s not some sort of um let’s look at my skin and determine that. But that’s how you avoid because you have if you have if you bring in people with different perspectives. You avoid avoid that mono focus that lets Facebook, you know, ruin the 2016 election in in in some respects, right? Because they just didn’t see the whole picture. They didn’t see the holistic picture of things. You have, if you have a diversity of thought and experience, you have someone who uh studied redlinining in college and has an idea of how data can be corrupted. and let’s look at where we’re getting our data and let me let me look at the assumptions within that using my critical thinking liberal arts skills to understand what went into this into this piece of data. Or or the same with, you know, if if all of your data um about u mortgage risk is 95% of it is white suburban homeowners, then you can’t be surprised that everyone

else is judged a high risk because you’ve just defined the norm using a biased not a large enough sample size and not diverse enough sample size. So, and this is I think where companies need to be very careful because if that’s how your company did business in the past and then you’re just using that and enabling AI on top of it because that’s and using the same formula, you are perpetuating the problems that we’ve been seeing in society scale at 100%. And honestly, that’s what AI can do and that’s where responsibility is such a a big proponent of all of this right now more than ever it’s ever before in my perspective. The other thing I wanted to share your concept of diversity of thought like you know even in the beginning of this conversation between both of you the the fact that you were like you know what we’re using these tools these legal tools and and we’re doing it in this way and here’s why but Cameron you specifically brought up that you know I understand product development I understand the emerging tech cycle I know that allows me to understand where AI is right now which

then helps me understand how to mitigate risk in this process again that’s a d not a lot of lawyers have that right? I >> I have different ideas from any group of lawyers I sit with. You know, they I’ve worked with a lot of people who from law school a long time ago. I won’t out them with how long they’ve been out of law school. Um but we get in a room and we start talking about things and they have a lot of knowledge from having done only law for all these years that I don’t have. And I’m bringing in technology. I’m bringing in talent considerations. I’m bringing in uh the flow state at work and how teams get into flow and how they become productive. And so it’s just a different perspective, right? So there’s a diverse thought and experience based on my background. >> Yeah. But I think all three of us hold that because when you started, you started talking about, you know, I’ve been in tech and fashion and and rights and like you have this very eclectic but beautiful background. you know, the the types of industries I’ve worked in is probably every from financial services to smart cities to web 3 to AI, but service-based companies, you know, like for me that allows me to sit here in

front of you guys and have this conversation. Next week be in front of the crew, which is like construction and real estate for women. Last week it was from the World Trade Center Institute with multiple industries talking about the same thing, but being able to be relatable. >> And I think >> you’re a generalist. Yeah. Right. That’s how you you you’re more creative, right? Specialists are known to be brittle in their problem solving efforts. >> They’re very brittle because if they run into the first brick wall, they often kind of fall apart. The generalist has figured out ways to maggyver his way around that. >> Yeah, I think we all have our level of expertise though. So, and I do think in the world of AI, that last mile expertise is going to matter more than anything. It’s going to be very interesting. Go ahead. Did Did you have something you wanted to add or >> Well, I mean, I just completely agree. I think that I also have um you know I think having introduction to a industry that wasn’t disrupted by technology has really served me well. You can imagine I came out of law school a couple years before the pandemic and so disruption has really just been core to what my experience has been like and navigating um the legal industry or what have you.

[47:28]

And so I think naturally I was drawn to artificial intelligence and other technologies that are disrupting various industries. And so I think that’s created a particular perspective, but you can imagine, you know, as Cameron mentioned, for those who maybe have been a bit more set in their ways, there’s still something they can contribute to the conversation. And so I think it’s about having these intergenerational conversations, these intercom community conversations, um, so that we can really come to the the best solution for our collective problem. >> I love that. >> So everybody loves use cases. Everybody wants to know who’s doing it right, who’s doing it wrong. legal is one of those things that I think this is very pertinent to our discussion and conversation. So I would love to hear from both of you. You know, who are the couple sort of cases that you’ve seen that like they’re they did it right? It could be your clients. It doesn’t have to be. And then who are the couple case studies that you’re like, wow, like they really messed up, you know, and and learnings from that. Well, you know, I if you if you look at the um the risk hierarchy, that’s that’s one way that

some of my clients choose things. They’re like, “Okay, should should should we be playing around? Do we want to sort of play in the in a very high-risisk area that could be high reward, like we’re dealing with safety, health, access? Do we want to be sort of in the middle where we’re recommending and nudging, but we’re not making decisions that affect anyone’s livelihood and people could ignore it?” or or the the the lower level at least in terms of the risk of of problems with customers is internal productivity and cost savings, right? So so sometimes they’ll just choose a risk a use case in one of those risk areas. um if they choose something like um customer service, it really helps to have it well trained on all of your customer service interactions in the past to have it reviewed, to have it tested, to have people, you know, bang away at it for a while and and give it throw it curveballs to see if it’s going to agree to sell a truck for a dollar as happened in one chatbot where it it offered to sell somebody a vehicle for a dollar. Um or Air Canada, right? It had a chatbot that gave somebody a different um refund

policy on a bereavement flight issue >> um than they had posted elsewhere on their website and Air Canada tried to make the argument that um they weren’t responsible for what their chatbot said and that somehow the user should have known about this content elsewhere on the website and the judge said no that’s not how it works and so they were held the the the policy that the chatbot gave was held enforceable. Now, there’s other issues, you know, a chat box. >> And that makes sense. That makes complete sense because that’s where I was getting the information and I have it, you know, documented. >> Yeah. I mean, if your employee gave that information, you’re liable. Doesn’t matter if it’s somewhere else on the website, it sells it says a different thing. >> Yeah. >> So, so that’s an area where, you know, you see people that are they’re kind of rushing to market. >> They haven’t really vetted whether the data is good. Um, and what sorts of things it can it can represent. I mean it might be better again if it’s retrieval augmented generation that it’s only giving answers to to things it knows the answers to. It’s not generating answers by predicting words. It’s saying is this in my database? No. Let me kick you up to the next level and

you can talk to a human. Now >> I just want to um pause there because people don’t know what our a rag agent is which you just explained. And so if you’re creating something that’s custom for your customer service and you’re inputting it with data that you have, your context, your content is different when then you’re using it to having a large language model generate that information for you. And there’s ways to even incorporate both, but have guard rails uh in place and guardrails can be prompts and specific things. This is what it should say. I don’t want any financial advice here. I don’t want this to you know hallucinate in this way you know and so this again it mitigates your risk so I just I like to help people really understand what these things mean the legal research tools use that for example they say don’t go ranging across the internet to do this legal research here are the cases use these and only these cases don’t be predicting words you found on random websites >> and and so you get really good product >> yeah you could even think about like your intranet how many companies have internets that never really get touched

or they’re just like dusted, you know, but like how much content and knowledge transfer is missing because people don’t even use their own internal data, right? But now you could have conversations with your own internal data, right? Any use cases come to mind, Amber, for you? >> Yeah. No, I I I definitely think that, you know, those clients of mine who are utilizing it from a marketing perspective, right, who are analyzing um either content they’ve already created and allowing AI to help them develop short form content for different platforms. I see that as being a really beneficial lowrisk use case um across industries, right? That that likely won’t end you up get you in trouble because you’re prompting, you know, these tools, whether it’s descript, right, or some of these different places that will take thems out, right? Um and really provide something that’s going to allow for you to continue to generate um work. Um I have clients who are developing tools um particularly in real estate right that are creating AI powered hubs for folks that are um helping to you know streamline that process of buying and selling real

[52:41]

estate right and so you know folks that again have that engineering background um and are able to actually incorporate that um into what it is that they’re they’re doing doing their due diligence I think in how they’re incorporating AI but then on the back on the other end of that spectrum you I think as you mentioned, it’s folks that are really relying on the output of these tools in exchange for making sure that there’s that last line of defense. Um that they are in the loop and and what they’re sharing with clients that they’re actually going out and relying on. Um you know, it’s just it’s irresponsible. Yeah. >> Right. Just in general. Um, and I think it’s a reminder that at the end of the day across industries, there’s only so much liability you can absolve yourself of when when others that you’re providing services to or products are relying on what it is that you’re you’re you’re charging them for. >> Yeah. So, let’s talk for a second. I mean, we’ve talked I think the generative AI a lot, right? That’s the prompt something to get something, the content, the images, all you know, um, documentation in certain aspects of it, right? Innovation’s a different ballgame

and here’s an opportunity to innovate like we’ve never invaded before. Right? So when you think about medical and financial services like some of the regula regulated uh areas that have you know privacy specifically um challenges right or uh HIPPA concerns um but they’re the probably the biggest areas that can use a lot of this AI to innovate in ways that we’ve never been able to be done before a financial planning agent obviously it’s fintech too but to create those allocations for you look at like the actual where the market’s going and potentially make those decisions by the second or millisecond, right? Like like traders do today or if you’re in medical you’re like physically like I’m a big believer in precision medicine, you know, and like you know my body is very different than your body than your body, right? And the things I’ve been impacted and affected environmentally or otherwise like that can now be created into a model of me for my health, right? These things are really potent. They’re groundbreaking disruptive ideas. There’s

so many more that can fit into these categories. What do they need to be looking out for when there’s like you know I think you mentioned it earlier Cameron around you know rights around PII so your privacy your data like >> that’s a different ballgame. Any thoughts on that from both of you? Well, I think you know the output may be less problematic because you’re dis let’s say it’s drug discovery or something. You’re discovering something, but you’re not just going to accept it and run rush that drug to market. So, there’s going to be a whole process of is this really going to work on this on this gene therapy? Is this really going to happen? And so, you you there’s a whole process of checks and balances there. The the the larger issue probably in most cases is was it your data to be to be using, right? you know, there’s there’s HIPPA, there’s data privacy. Um, a lot of times organization can they they can think that they’ve deanonymized, you know, they’ve taken the the identific identifying information out of their uh patient data and so forth, but there’s

still something there that makes them identifiable. They they missed that fact and now they’ve they’ve they’ve violated HIPPA. They violated, you know, state data privacy laws. So understanding that, you know, getting the rights to the data, making sure you’re not using a scrape and prey kind of methodology, you’re not you’re not violating privacy. And the data, again, the data, if there’s bias, it’s probably going to come out in the in the process of the prediction because it either will or won’t predict something that’s valuable. But you could predict a drug that works for me and that doesn’t work for either of you. And that that could be a real thing because maybe it’s got more data from people like me and it’s it doesn’t understand that our bodies and and hormones and cells and everything are are aren’t completely the same. Right. Right. So I I think in there’s probably less risk there because it’s a more of a closed process, but there’s not zero risk. >> Amber on your side. >> No, absolutely. And so this specifically speaking to those more, you know, like highly regulated industries, right, that do have those regulations in place that,

you know, stipulate how data is supposed to be protected. And so I think it’s it’s making sure that you’re adhering to that, you know, as as Cameron mentioned. And then of course, I think it’s also about, you know, also looking globally, right? Like the work that we’re doing now because of technology, you know, we also have to be conscious of the GDPR, right? the >> TVPR is the marketing like regulations based out of Europe that really focus on privacy and >> data privacy. >> Privacy. Yeah. >> And I think but that’s also where closed source tools come into play, right? Because there’s less need to redact and and scrub the the data sets that you’re using to the extent that you’re not putting them in some sort of a large language model that others will have access to. So I think you know making that extra investment so that you have you know access to large language models that aren’t just generally available I think is another another way to mitigate risk. >> I think the theme I hear in that is data data data data right like and and honestly that is u the biggest play in AI but it also drives brand new business models and so a followup to this is I know that people are going to be able to

[57:58]

license out their unique data sets in the future. It could be a data set of myself which you know we’re working on a product but like there’s a lot of that data can be used now for you know B2B TOC type of opportunities. Is there anything that they need to be thinking about? I mean you mentioned where the data is coming from and getting the rights and the because I I do know folks who are scraping the internet today for publicly you know public data right is that okay like you know where hey I wrote uh a review about this product and now they know emotionally I’m tied in one way or the other to something on my social media to the way I’m responding you know for you know again anything public. Is that okay to do? >> You know, there are vast reservoirs of of data that we’ve all put out there. Organizations like Facebook have terms of service that say that they can at least train on anything that you’ve put out there publicly if you haven’t put friends only.

>> They say, “Well, if it’s public, it’s like the internet, right? That’s that’s something we can train on.” In in terms of companies scraping, you know, there’s there’s there are a couple potential issues. One is what are the terms of service of of the website that you’re scraping. So your scraper needs to be able to read that and if the terms of service say do not do this or if they have robots.txt txt files that are saying, you know, do not do not you scrape this information or if you’ve gotten through a login um and in other words, you’ve scraped where where the own the the site owner has made efforts to keep you away. Now, you could be potentially uh in violation of the Computer Fraud and Abuse Act, the CFAA, right? Which is similar to hacking. It says basically you can’t hack. But um it it if if you try to build a a system that goes onto LinkedIn and scrapes LinkedIn and tries to predict when employees will turn over and you want to sell that to corporations and say we’re following your your employees uh updates to their LinkedIn profile and we’re looking at their posts and we’re looking

at what people are saying to their posts and here’s a x% chance that this employee is going to leave. They they’re probably violating LinkedIn’s terms of service. Um, >> and and if they’re doing a violation that’s that’s broad enough and it affects individuals, then you’re looking at a potential class action and that’s where you can have serious damages. >> That is a huge insight. You know, it’s like again, these are things that we don’t as founders and innovators, you don’t think about. You’re like, this is publicly used data like you know, like it’s out in the ether. Anybody can use it. So thinking about the terms and conditions of the actual site itself is really really important as because for me I I’m thinking about it like maybe there’s like a a first pass if you want the information maybe you can use it but then if you’re trying to actually be strategic and sell your company they’re always going to come back and say where’d you get the data. Is that fair to say? >> Yeah. I mean in the due diligence process if you’ve cut corners Yeah. the acquiring company will find it and they’ll say if we buy you we’re buying all this liability or or we’re buying bi

bias. We’re buying something that doesn’t really work as advertised. Your val we’ll buy you but your valuation’s half of what you’re asking for. Right? So, so this is why data integrity and data dignity are very important to build in from the front and not to try to take shortcuts because somebody whether it’s a regulator, a plaintiff’s class suing you and getting discovery into how you came to your decisions. And if you say stuff like, “Bro, let’s let’s just go ahead with it.” Um, you know, I understand the creative mindset, right? It’s like, “Hey, wouldn’t it be cool if we could do this?” And they’re like, “Yes, that would be cool, so let’s do And it’s like, well, there are other criteria. Yeah. Right. Is it it’s cool, but is it legal? Is it ethical? Do you have permission? Uh is it is it discriminatory? There’s a whole checklist. And that’s why I say you’ve you’ve got to have your your AI projects looked at by a a team of diverse thought and experience. It includes a lawyer. >> Let me ask this one followup here and then Amber, I want to hear your thoughts here, too. for that LinkedIn example, is there a way to negotiate that data with

LinkedIn to say, “Hey, this is what we’re creating, can we like or or is it a non-starter once somebody’s terms and conditions says that you can’t use their data?” >> Well, this is exactly what a lot of the LLMs have have done. You know, Open AAIS deals with Reddit. They they tried to get one with the New York Times. um they’ve they’ve got them with um a number of organizations where they said we we see that you have um terms of service that say we can’t be scraping we’d like this so we’re going to pay you now there you’ve got all the people on Reddit who’ve been creating all that value saying wait what the hell you’re you’re now taking ownership of our of our creations and so there’s there’s rights cascade down and someone’s always going to be left holding an empty bag but but that’s what they’re trying to do. I mean, data is king and increasingly you’re going to have a lot of large corporations >> um and and and and universities that have massive amounts of internal data >> um that that is just um uh a a gold mine um if they can figure out how to

[63:16]

monetize it. >> Yeah. >> Amber, your thoughts? >> No, absolutely. You can imagine as a transactional attorney, as an IP attorney, one of my responsibilities is to support with rights and clearances. So earlier this year, I had a really cool project. A client was developing a podcast for a foundation. And in the production of that podcast, they needed to use a number of different archival uh resources, if you will. And so whether it was from the New York Times or CBS or what have you, and you know, a lot of what they’re finding is on YouTube. And then you realize this is copyrighted materials, and you still have to follow that kind of chain of command to see who actually owns it. And then you read those terms of services and see that they have, you know, a limited um, you know, you have a limited ability to use them outside the context of fair use, right, for educational purposes. But once you start monetizing what you’re developing with other people’s data, that’s when you you want to have some sort of clear express license um, from that owner or else you’re, you know, you’re absorbing some level of liability. Um, and to the extent that, you know, if you’re being acquired by another company or you’re taking on investment, a lot of times what you’re doing is you’re representing and

warranting that you own intellectual property, right? And you own you know the things that um are integral to your business. And so you can, you know, skirt the process and and not do your due diligence, but if anything were to arise, any type of lawsuit or what have you, you would still be liable even if you’ve already sold that company because you represented on the front end of that transaction that it was yours. So I think it from a due diligence perspective, from the idea of rights and clearances, you know, AI is not going to supersede, you know, some of those existing some of that super existing legislation until it does, right? Until there is something Yeah, and I’m also a creative problem solver, especially when it comes to business models and strategy. So for me, I’m like there is syndicated data that you can purchase out there. There is open source data sets that you can actually leverage. I know like companies like Nvidia are creating data sets, right? So it’s like you really there are other ways to potentially go about what you’re seeking. And now that you also understand I mean having a and I you know I’m just speaking from a startup perspective not like a a grown sort of seasoned company but it’s it may be hard

to negotiate a contract with a LinkedIn to get their data even though if you have the best idea in the world right versus like a noia or somebody bigger right like so um there are other ways is what I’m saying that are legal um and you can potentially do the same thing right so it’s cuz um I wanted to make sure that we solve for people is there any thoughts on that from you guys. Am I right in that way? >> Generally, yeah. I mean, you >> this it’s just part of the checklist, you know, and it’s like >> if if you have the legal vetting, what it gives you is not just potentially preventing an Air Canada situation or an Amazon resume situation or whatever, especially again if it’s if it’s filtered through that team of of diverse thought and experience and they can there’s different people who have different perspectives on well, did you think of why this might be a problem? Oh, no. our six programmers didn’t think of that. Well, now now you have. But the other thing that you get out of out of a process of of having it sort of a auditing of your AI adventures is that you can get a a kind of a clearance

letter, right? I mean, ignorance is no excuse. Courts under the law of negligence will say, “Was this foreseeable to a person of of in your position?” In other words, if we brought in expert witnesses from other companies in the industry and we brought them in, would they say that you knew or should have known this would cause a problem? And that’s the heart of negligence. You knew or should have known that a harm would would happen. You did a thing anyway. The harm happened. You were the proximate cause and so forth. And so if you have an attorney at least trying to help you with this, >> then you’ve shown that you’re exercising some some due diligence and you’re showing a standard of care >> and it and it looks like it’s not intentional. Say, um it’s the same with IP. Intentional infringement >> gets you triple damages, >> right? If you unintentionally infringe, you can still get damages. you can still lose the case, but you may not have to pay the other side’s attorney’s fees and three times the damages as a as a almost punitive award. So, showing that you went through those efforts um that

there’s a duty of care and it’s so and and a lot of these duties and a lot of people think that we’re not because there’s not federal AI regulation that we’re not effectively governed and that’s a real mistake, right? We have um statutes that say you can’t make arbitrary and capriccious decisions, say about healthcare denials. Arbitrary and capriccious just meaning not rational, not consistent. >> Mhm. >> Um you have contracts that sometimes bind you. I’ve I’ve worked with my my co-founder in deep law had his YouTube account terminated because as far as we can tell YouTube’s AI couldn’t tell the difference between his boring lecture those are here his words >> on on how the adult industry has has actually innovated with things like file compression and and so forth couldn’t tell the difference between that lecture >> and apparently pornography. M >> it heard him talking about adult industry and it it flagged his his account and put an age restriction on it and even though he went to a human then

[68:31]

who said who looked at the video and said no this is fine the next thing you knew the next morning the AI has come along again and flagged it and then he pulls the video down and the next day he’s terminated forever like you’re just you’re banned and it it was it was a huge mistake but >> and he did it to two lawyers. Oh man, YouTube. >> But you know, YouTube, unlike all the other platforms, a few years ago, they made a decision. And if if this is on YouTube, you’re you’re you’re you’re a beneficiary of this decision. >> They said, “You know what? We don’t want to be like Twitter and Facebook and whatever where we can terminate you for a good reason, a bad reason, or no reason at all. And in fact, we don’t have to give you a reason. Elon might have been in a bad mood. You’re terminated.” YouTube said, “You know what? We’re going to give people some reliability. If they come on this platform and they invest and they develop a following, they’re going to have some due process. So, one is we’ll tell you the reason >> and two is we won’t terminate you unless we have, and this is the contractual language, a reasonable belief that you violated our terms. >> So, our argument as novel as far as we

know is AI doesn’t have beliefs. If you look if you look up the root word, >> not yet. >> Not yet. If you look up the root word in German, belief involves emotion. and it’s subjective. We don’t think AI knows what’s reasonable. And even if an AI could hold a reasonable belief, we’re saying YouTube can’t delegate its belief to an algorithm. And furthermore, even if a human was involved in its decision, it’s it’s not reasonable. And so there’s an example where it’s not the EU’s law. It’s not a state law. It’s not some AI specific law. It’s just you have to be reasonable and rational and fair and and that and the law of negligence that’s going to govern a lot of behavior whether people like it or not. >> Yeah, I think that argument will work Cameron until we have AGI artificial general intelligence where AI will become sentient and actually have emotion. So that’s legal. That was my leal moment. But I actually have a question for both of you because there

is this humans when they do work they have an a percentage of error right with work right they just you know they’re not perfect humans aren’t perfect when you even think about like data centers there’s like 99% uptime right like there’s still that 1% era why do we believe that AI is supposed to be perfect >> I would say that and speaking to the to our last point I think it’s also really important just to note that to the extent that clients are using AI, businesses are using AI for the sake of increasing their ability to scale, right? Scalability is is what differentiates between a small business and a startup, right? This ability to go from zero to, you know, um 100, right, in an exponential amount of time. Um but it’s also a part of being investor ready, right? To be able to articulate what your company needs. And so that might look like the funding in order to develop a license agreement with one of these major platforms to have access to their data, right? That’s something you can articulate to folks that are looking to invest in your idea. And so having that clearance letter, having these

memorandums from your attorneys, right? That’s also a part of being a more astute CEO, a more astute, you know, advocate for your uh enterprise. And so I think that’s something important. Don’t look at it as a downside that there is costs associated with having to invest in good data but to be able to say this is what allows for me to have a better understanding of the business that I’m operating and what my business actually needs >> and a better IP because now it’s legally your IP part of your IP >> and that’s what people are investing in people are investing in the intellectual property people investing in the things that you can protect right um and so with that said I think the idea that AI is to be perfect I think because people are are looking at AI to increase efficiency right increase productivity they’re assuming that that means it’s also eliminating that room for error that I think is inherent when humans are the ones fully behind any type of endeavor. Um but I think you know the technology is I always say it’s only as smart as the people that are developing it. And so to the extent that we’re not perfect the technology won’t be perfect and I think that’s an unrealistic expectation to have. Um but we shouldn’t let the perfect be the enemy of the good, right? And so that’s why we’re still developing here that’s why we’re

having this conversation why we feel passionate about incorporating AI. Um, not because we have this unrealistic expectation that it’ll be perfect, but because it is it is beneficial and it’s allowing us to scale our businesses. >> So, you said we shouldn’t let perfect be the enemy for good. That’s your line for this podcast. I love the enemy of good. I love that. Quick shout out to one of my favorite partners, Legacy Sessions, the studio where we tape the AI CEO podcast in the heart of Washington DC. If you’re a founder, business leader, or creator who wants to look and sound worldclass, Legacy is the spot to record. Beautiful sets, prolevel production, and a team that just gets it. Visit legacy sessions.com and tell them I sent you. Okay, so let’s get into another random but fun story. Uh, I just saw that there is a gentleman that was going to court. Okay. And he wanted to, you know, you’re you’re allowed to represent yourself in court. that’s part of our legal right. And so the judge and him started talking. And

[73:46]

the judge was all pissed off and it was one of those like random shows I’m sure whatever but he he generated an AI lawyer chat GPT pass the bar right and he’s like why can’t I have chat GPT represent me right and so going back to Cameron your sort of use case you mentioned you know you had a rag system you know you had an agent that was actually with but it had context and content that allowed you to look at something quickly while you were doing the discovery and the conversation of your of your uh with your client in court. The judge was pissed and she’s like, “What do you mean?” Like, you know, so one, just to get both of your perspectives on that, but do you believe the world of legal is changing? Is that the future? I >> I think so. I mean, there have been estimates that 80 to 90% of um people with civil uh noncriminal legal issues cannot afford representation and don’t get it. the idea that state bar associations can continue to say, you

know, thou shalt not enter. Um, you can’t help other people. And there’s there’s a there’s a First Amendment component, right? There’s there’s a there was a case in New York recently where uh a company wanted to offer a service of of having lay people help people fill out a debt collection form to explain when they would get defaulted. And here’s the issue. There’s a huge number of of debt collection uh suits that get filed against people >> where the person who’s filed against does not actually owe anything. It’s an error or or maybe it’s made in bad faith, but they don’t know about it. They don’t get the notice. Then they go into default. So now the judgment is owed. They didn’t have any process any due process in front of the court. So now they have this default judgment. The collector’s coming after them. So this company wanted to have lay people maybe uh one of one of the plaintists was a minister in a church who would help the cl uh their you know clients or or volunteer basis help them fill out a one-page form with here’s my defense or this wasn’t a fair contract or I didn’t

do business with this company but just to check the boxes >> and they wanted to get permission to do that >> and the state of New York hadn’t sued them but they were afraid of being sued so they said we want an injunction to keep New York from suing us if we do this and the judge balanced the the the sort of need to regulate the practice of law and protect people from unauthorized practice against the first amendment rights of the people who wanted to give the advice and so forth. And he decided in favor of the first amendment >> and said I this is a very narrowly subscri circumscribed issue. They’re they’re not representing them on anything else. It’s free. They’re not offering to to you know represent them in other areas. I’m going to allow this basically. And I I think you know it’s the same with medicine. I mean the the number of people who can’t get good quality timely care because they can’t get e easy or or affordable access to doctors is is just astronomical. So, I think it’ll start with AI being used by legal aid societies by the more

sophisticated um parties who are lay people but are have a really high reading comprehension level or understand even how AI can be used. I’m sure they’re already using it in small claims courts. But is that and I think what this judge is part of her problem was probably just the novelty of it, just the shock of it. like you didn’t say you were doing this, but I I can see a point in the future where even with AI’s hallucinations and issues, the argument it brings up in my favor as a lay person is going to be better than what I could do on my own >> 100%. >> Why shouldn’t and maybe it can go through a filter. Maybe the court has another technology that reviews it, but but courts already are able to catch false citations. They’re able to determine what’s real and what’s not. They bend over backwards for prosay or unrepresentative parties. I I can see this be part of the future and I think jurisdictions that that have a huge resource issue like California and New York are going to be at the forefront. >> Yeah. I mean, look, this is an entrepreneurial idea. Like, you know, I I always hear hear um Elon Musk talk about how humanoids they’re going to be

more humanoids than people by 2040. And, you know, they’re going to do a lot of labor work. But I see them doing work like this, like, you know, because it’s it’s things that in a way need is just programming, right, with the guard rails. At that point, there’ll be a lot more use cases that they’ll be uh programmed with, you know? I don’t know. Oh, I just there’s I think there’s a future >> that looks like this. >> I will say in this particular instance and I I’m I’m fairly certain I know this was the exact circumstance because in this case there was an AI generated person, right, that that showed up on the screen. I think it was a virtual maybe hearing or what have you. And so I think that’s the distinction, right? It’s one thing to use AI to help you draft a brief, right? as a pro-say litigant, you know, it is your right to defend yourself, to submit documents to the court, things of that nature, as long as you’re adhering to deadlines or what have you. But when you go that extra step and have that person appearing on your behalf, right? Someone who is over >> purporting to be a human, >> right? Someone who’s overconfident, right, who hasn’t taken an oath of

[79:00]

office, right, or or an oath to uphold certain ethical standards, who can’t be sanctioned to the extent they violate any of those standards. it definitely creates a slippery slope in terms of accountability and and really and again regulating the industry of of of law. And so I think in that case I think we’re going to see uh you know an increase in the amount of agent AI agents that are supporting clients and helping to submit documents to the courts. But I I can’t foresee a future in which we would allow these humanoids to actually appear on behalf for all those for all those reasons I mentioned. >> I love the opposite perspectives here. Let’s go. >> And there are areas that that where AI has been shown to be very accurate. I think one of them might be um like eviction law. >> Yeah. >> Facts are often fairly simple. You know, it’s like upload your check here if you paid it, if you paid the rent, right? Or um upload your contract, you know, put in a few facts and then it could generate some things that sends the parties uh more questions. They could answer them. Pretty soon the judge starts to say, “If what I have is

correct, I’m leaning toward the tenant.” Landlord, do you have anything to say or would you like to dismiss this? I mean, why why shouldn’t that be an early line of defense against a well-resourced landlord? >> It could, but I mean, have you been to New York? >> Have you seen the squatting issues? Have you I mean, I just could go on with stories that unfortunately friends of mine who have been tenants and vice versa have been owners just it’s the craziest time. But I I don’t disagree with you. You know, I think there’s a lot of that. So, uh, we’re gonna get into some rapid fire questions. You ready for them? >> Yep. >> All right, >> let’s do it. >> Would you trust an AI tool to draft your next client proposal? Yes or no? >> First draft? Yes. I’m going to revise it. >> Yeah, assuming I’ve prompted it with prior proposals, my pricing model, things of that nature. Absolutely. >> What keeps it safe? double-checking, validating that the information actually corresponds with what my typical pricing model is that you know it’s not overcharging for things or what have you or maybe it isn’t undercharging. I think that’s

important because those proposals ultimately lead to that engagement. So I would say it’s it’s critical that I have that last line of >> defense. >> What’s more dangerous bad data or overconfident founders? >> An overconfident founder with bad data. Right. I think the overconfidence probably comes from the bad data. So maybe we’ll say what came first, the chicken. The chicken maybe the bad data if that’s what’s leading to the >> overconfidence could breed the bad data. Yeah, >> it’s both. >> Okay. AI agents that deal with sensitive customer information. Okay. Now or not yet? >> Not yet. I mean there there’s too many passwords. You’ve got to put the information into a into a system that you may not have the rights to do. um they’ve got to go and log in places and potentially expose the data to outside vendors. Um they’re just I just don’t think they’re accurate enough yet and I don’t think the the legal regime is quite in place. >> Yeah. AI agents to the extent that they’re advising folks on the other end of that that tool. Um yeah, we’re not

there yet. >> Okay. Start with you on this one. One AI tool you actually trust right now as a legal professional. Shout out to Lexus plus AI, specifically their protege tool. It’s it’s something I I do rely on. >> Cameron, >> well, there are a couple. I I dec AI for discovery. I think I I have a good bit of faith in. Um I I I did like Caseex’s um co-consel legal research until Weslaw shut it down. Um I don’t think Weslaw’s uh AI research is as good. And uh um I I actually for a first draft I’m really very impressed by chat GPT Pro. It needs to be reviewed but absolutely it’s become very impressive. >> And the pro is closed source. Correct. >> Um >> behind the pay wall. >> Yeah. I don’t I don’t it it doesn’t hallucinate um case reporters. So if if there’s an actual case that was published in the law books that you see

in lawyers offices, those are real. The ones that I’ve seen it systematically hallucinate are a WLAW or Lexus site where it’s just predicting and it makes it up. >> So I’m actually going to ask a followup there because this is an important one with Chad GPT. Are you using the public version or the enterprise version? Because the enterprise version is a little bit more closed loop. the open version is not right and it could potentially be training the model unless you hit the setting that says it’s not training the model in the privacy which we don’t know if it is or not. >> I I’d love to hear your thoughts on that. >> Well, the free version I I would not use um I mean other than for basic research but if I had to tell it to train on a folder full of documents, let’s say emails, I can’t tell it not to look at my client’s email address and that stuff goes into the training. Um, so I I need to at least use the pro model or the enterprise model to keep it from training on my client’s uh information and keep it confidential. >> Yep. Amber, any thoughts on that? >> Yeah, I typically don’t use TAG GPT um

[84:16]

from a for the purpose of developing work product unless of course it’s around maybe marketing content, things of that nature um or just for personal uses. >> Okay. >> Um but Lexus plus AI, we’re working on that sponsorship. They >> are you listening, Lexus? >> Right. Come on. >> I mean, >> they But no, they they again, this is one of the more reliable tools that lawyers, you know, across industries are utilizing already. And so, they’ve just incorporated these new AI tools and and so they can support with drafting, they can support with with doing that kind of deep dive research as well. So, I think that’s that that’s what I feel most comfortable from that perspective in terms of generating work product. >> Okay. In 2030, the CEO CEO who ignored legal risks in AI will be >> learning how class actions work while on a Zoom deposition. >> Obsolete. Obs, you know, I just think there’s there’s no way for you to really continue to scale in a sustainable way without um really taking into account

the risks, but of course the rewards as well. >> Amazing. Listen, Amber Cameron, thank you so much for this awesome dynamic conversation on legal, AI, business risk, but really innovation at the end of the day without what you guys do on an everyday basis and sort of the insights that you were able to provide. I mean, I this is thousands and thousands of dollars of insights for the last hour and 25 minutes. And I just want you guys to understand that and that’s why I knew this conversation was going to be so important. So, any last words and just let people know where to find you and then we’re gonna close this out. >> Uh, right now, uh, best spot is probably my LinkedIn profile at Cameron Powell. >> Can I scrape that, Cameron? Yeah. Do I have your permission? >> Yes, you have mine. Check LinkedIn. >> Amber, >> no. No. Thank you both for today’s conversation. This was um I think definitely the culmination of you know really diving deep into the world of artificial intelligence, web 3 I think more largely and a creator economy that I think we’re all really um participants in. And and with that said, you know, please stay in touch. You can find me

via my website amberithlaw.com. That’s amvitthlaw.com via socials amberith law and on LinkedIn at ambermith esquire. >> Amazing. So, thank you again for listening in. I mean, if you like this conversation, which I know you did because Amber and Cameron brought so much gold to this conversation, please like, subscribe, and comment. The more you share this podcast, the better we get at bringing great great humans, great experts just like these two and continue this conversation to the journey that, you know, we never thought we were going to be at, but we are there now. So, until next week, thank you. That’s a wrap for this episode of the AI CEO podcast. If you found today’s conversation valuable, don’t forget to rate, comment, and subscribe. It helps me reach more leaders like you and continue bringing in top AI first founders, industry experts, and visionary CEOs to share their insights. AI is evolving fast, and the CEOs who harness it today will lead tomorrow. Stay ahead of the curve, keep learning,

and keep building. Thanks for tuning in, and see you next time. [Music]

Watch on YouTube →