AI in the Courtroom: Navigating Legal Landscapes with Gareth Stokes

0:28Welcome to the AI Chat Podcast. Thanks. Delighted to be here. So super excited to have you on the show. What I'd love to kick this off with is I'd love to ask you a little bit about your background and your journey. I'm curious, was AI and kind of this tech side something you always knew you were interested in growing up? Or is this something that, you know, you kind of discovered in college or something? Tell us a little bit about your background there. Yeah, sure. So, well, the first thing I have to admit is exactly how long I've been doing this, which makes me sound terribly old in the context of all these bright young things who are sort of getting started with AI at the moment.

1:04But I've been a lawyer for 21 years now. And I started out as a technology lawyer way, way, way back in 2001 because I've always been really, really interested in computing, really interested in what was then the kind of nascent, you know, dot-com boom and all of those sorts of things. And tech transactions over that period have evolved enormously. The first point at which we really started to see AI becoming a material part of deals was doing kind of large outsourcing deals with a lot of the really big tech vendors. And we suddenly, in around about 2014 or so, started to see various vendors starting to talk seriously about transformation projects and outsourcing involving AI in some form.

1:54We'd had IBM's Watson kind of win Jeopardy in 2011. It took really until sort of 2014, 2015 to start seeing that kind of technology commercialized in outsourcing deals. And that's really the first contact I had with AI on kind of client-facing mandates. Okay, very cool. Super interesting. So of course you had, like you mentioned, a really incredible kind of career in this space. What kind of motivated you to co-lead DLA Piper's global AI practice group? um partly being super interested by the technology so um one of the things that led me to be a technology lawyer in the first place is i'm kind of a kind of turbo nerd as well as as a lawyer okay um and i could see the impact that lots of the technologies were starting to have very very early on um and the impact was going to be outside of you know the simplify, standardize, automate transformation journeys that outsourcing projects were typically on.

2:58And I could sort of see that coming very, very quickly indeed. We were then invited to get involved with a bit of advice for the UK's House of Lords around a report that was published around about 2017 called AI in the UK, Ready, Willing and Able. That was the sort of the kernel of starting a much wider look at the AI landscape and I started then reading lots and lots and lots of the kind of archive papers that were out at that time because there wasn't really a tremendous amount of user accessible country outside of the academic space and the timing was almost perfect because just as I was starting to do that Google kind of published the the seminal attention is all you need paper about transformer models and i spent a long time trying to get my head around that because i'm a lawyer not a mathematician um but eventually it started to penetrate and i started to be able to have some uh some conversations with some of our clients who were then looking at investing in um our organization some of our clients who were looking at building some of these tools themselves and some of our clients who were looking at you know whether or not these things were you know threats to their business model so we started having lots of conversations around around that time and of course things just snowballed even by 2020 2021 the people who were in the know could sort of see see the tidal wave coming and then Pandora's box was opened in November 22 when chat GPT just exploded onto the scene.

4:41And so luckily we were already there with a global AI practice that had been going for a couple of years by that stage. So thankfully kind of caught that way. You're all, yeah, you're all lined up for it. That's super awesome. Something I'd be curious to ask you with kind of the background you have and whatnot, what are some of the common legal challenges you encounter when you're working on AI projects? Yeah, so I suppose this depends on who you're advising and what they're looking to do, because an awful lot of people are in the position of being customer organizations saying, we are not AI experts, we're an insurance company or a bank or a life sciences company or whatever it is.

5:21How can we take advantage of these technologies? How can we benefit from these fantastic productivity increases that everybody's talking about that are going to be delivered by AI? How can we sort of make some of our processes more efficient? How can we deliver these insights? And when they're doing that, it usually starts off with a top-down approach if people are doing it properly. So looking at things strategically, what is it that we as an organization would like to be doing with this? How would we like to be doing that? What are the top-down choices that we need to make around our values as an organization?

5:56And how we can implement AI in a way that is consistent with those values? And so once you've got that strategic approach right, you can then move down to the kind of tactical layer of saying, how now are we going to convert that strategy into a series of policies, a series of governance frameworks, some committees that bring in all the stakeholders, contracts, et cetera, and then operationalize that. Because it's all well and good for someone like me to write into an AI policy that we will do this ethically or do this in a way that's unbiased. but then someone's got to build a system that meets that test someone's thinking to be able to operationalize that yeah sort of proved from a data science point of view that this really does match with exactly what we're doing in that space and so that was a gap that i think daily piper had recognized relatively early and one of the things that we've done is in addition to building a series of legal experts who know what the kind of the nascent laws are that affect AI around the world because it's sort of you know evolving landscape we've also got a series of kind of lawyer data scientists who are able to help kind of prove that the algorithms that are being put together prove that the you know models that are being trained and the data sets they're being to use to train them do match up with those you know tactical level policy statements that say this will be unbiased or this will be ethical, et cetera.

7:24And I think that's a really important piece of the puzzle because too often there's a gap between the kind of operational layer and the tactical layer. Okay. Yeah. That makes a lot of sense. And definitely sounds like something, you know, people have to consider and to work into their strategies. Something else I'd be curious about asking you about is, you know, a lot of people kind of mentioned the importance of, you know, upfront data work. In your, based off what you've seen, what are some of the best practices for data governance and AI projects? You know, specifically when we're looking at that, the risk side of it that I'm sure you deal with, you know, in the legal front.

8:05Yeah, and there's an awful lot to that because there's the sort of, there's the data governance in relation to the organization's own data. So there are already, and there have been in many, many different parts of the world, a series of legal frameworks that govern how you have to kind of collect and use data for various different purposes. So some of that will be kind of personal data or PII type rules and regulations that exist in various different parts of the world. And an awful lot of what organizations, particularly enterprise customer organizations, want to do with AI tends to involve processing personal data in some way anyway.

8:43There's then also a whole series of laws and regulations that are sector specific to some extent. So if you're doing AI and life sciences, you've got to make sure that you're complying with all of the various different regulations that exist in that space. And you're likely to end up building something that is, you know, a software medical device in regulatory terms in some sense. Or if you're deploying it in the financial services sector, you've got to make sure that you're complying with all of the oversight and control regulatory frameworks that exist in that sector and so on and so forth. So there's how you control your own data.

9:15And that's a big challenge. But at least organizations feel like that's their own. They own the IP in relation to it. They're confident of its sort of veracity. They may have the mother of all tidy up jobs to do it, so they're honest in relation to sort of actually getting that data into good shape. But that's their data, and they sort of feel like they can put their arms around it. The challenge with AI is that we're also seeing people coming along and saying, well, now I can use these pre-trained tools that exist in the world. Some of those are, you know, access via APIs because they're controlled by some of the big name AI providers.

9:51Increasingly, we're also seeing people being very, very excited about some of the open source and open access large language models that are out there that they can then host and run themselves. But the question in relation to all of those things is, okay, great, but how much visibility do I have in relation to those sorts of tools? How much confidence do I have in how those are going to be used? And so it's really about sort of data governance in terms of your own data, understanding what the potential data landscape risk is in relation to those pre-trained models, and ideally putting it together and doing something that allows a kind of best of both worlds solution.

10:31And that might be prompt engineering or using existing large language models to kind of allow you to talk to a document or talk to a document set or something of that type. There's a lot of solutions in that space. In some cases, people are very, very sophisticated and they actually want to unlock the model and do actual training runs to truly fine tune a model to be better for their needs. obviously that's an awful lot easier with you know the likes of a falcon for a falcon 180b or a falcon 40b or a llama 2 or something like that they have open source open access models that kind of training is a lot easier there are solutions out there from the enterprise players that allow a kind of fine-tuning layer on top of a locked large language model to be developed as well but that's that's an area of an awful lot of interest for the more sophisticated organizations who are saying how can I have the best of both here?

11:28I feel really confident about my ownership of this data. I feel like I can contractually get comfortable with the risk that exists around some of this other data, or I can take one of these models and host it myself and do fine-tuning runs and feel confident about the quality and ownership of that model. And that allows me to then feel good about the situation I find myself in and the risk that I've got around that data landscape. Okay, very cool. Yeah, I think those are definitely some really important considerations and things to look at. Something I'd be curious on asking you about is, you know, when you're kind of in this space and you're, you know, consulting and helping people specifically in kind of an AI focused area, what are some of the key elements you think that, you know, organizations are overlooking right now when, you know, maybe that's preparing data for AI models or for other things?

12:21about what are some areas that you see that people should be aware of or thinking about? Yeah, there's a few common challenges. So I think that one of them is people thinking that the current state of affairs is relatively static and they don't really need to be worried about the changing regulation that is coming down the tracks. And that's not just, we hear an awful lot about things like the EU's AI Act for intelligence, which is sort of the first big regulatory framework that's coming down the tracks in that space. But I think that there is going to be a lot of regulation in other parts of the world coming down the tracks very, very quickly indeed.

13:00So from a US perspective, we've already seen Senate committees interviewing CEOs of very well-known AI companies. We've seen summits being held with both political and technology players being invited along to those summits. We've seen lots and lots of industry commentary around the need to give reassurance to customers around some of these sorts of things. And some of the big players have even gone so far as to come out with, you know, statements that are intended to give comfort around IP issues or statements that are intended to give comfort around data issues or whatever they might be. So there's a tremendous amount of changing regulation there.

13:49And that, I think, means that large customers can sort of take where they are now and look to the future. And if you're a provider, and we work with an awful lot of the foundation model providers and organizations that are building these things as well, it's important to sort of be able to have conversations both in public and sometimes you know via other channels to sort of try to have some degree of sensible dialogue around the direction of travel with regulation because you and everybody that you might potentially consider as a customer in future are going to get caught by this stuff so i think that being prepared for what's coming is a really really big part of this and the organizations that are ahead of the curve there are the ones that will you know save money and not have to do remediation jobs in future and they will you know they're baking in benefits today that they'll be able to continue to get the benefit of in an even more accelerated form tomorrow and those who are you know in cautious they're the ones who are going to end up kind of you know doing this job twice right you You know, there's definitely a lot of incentive to get this right first time if you possibly can.

15:06Okay. Yeah, that makes a lot of sense. Something you kind of touched on that I'd like to, you know, double click on a little bit is like, how do you see the relationship between intellectual property and AI evolving in the coming years? Right. You mentioned some companies, I think specifically Adobe has said, if you're using some of our image generators and you get sued for something, we'll cover the lawsuit. So there's kind of like that angle. There's people like OpenAI that just kind of have like Dolly and they're just like throw it out there. And they're just like, I mean, it's kind of like a use at your own risk in a way.

15:37Right. How do you see this evolving in the future? It's a really interesting one because there's obviously a lot of cases being pursued in various different forums at the moment. So we've got a few examples of class action suits that are out there, particularly in the United States, where people have made a whole series of different claims around the extent to which copying may or may not have happened in relation to training materials that are alleged to have contributed towards various different training models. and I you know there is I think a degree of of naivety amongst some of the people who are looking at these questions at the moment and I kind of touched on this at the beginning trying to get to the heart of exactly how these systems work and what it means to train an AI and what an AI model actually is in terms of the collection of weights and biases that make it up these are not familiar concepts for lawyers who are more used to the sort of technology copyright litigation cases that are sort of more akin to Napster and the sort of out and out piracy of large volumes of media.

16:54The other complicating factor is that actually these are cases that in some instances are being brought against organizations that are both AI model providers on the one hand and rights holders in their own right on the other hand. So there's a massive incentive actually for a lot of these big organizations to get things right from an intellectual property perspective. So I'm really, really interested to see how some of these cases will play out because I don't think it's going to be very, very clear cut. I think there's going to be a lot of interesting decisions by people who probably need to be brought up the curve fairly quickly in education around how these things work in order for judges to hear the cases properly and to have the facts at their disposal when they're making these decisions.

17:39And then we've also got some cases out there where we've got large rights holder organizations pursuing claims. Those are easier to see being pursued in a really clear cut way because there's a much more, you know, there's a single claimant usually. They've got a very particular position that they're going to be able to pursue. They're usually able to sort of be relatively well funded and, you know, they'll have gone for a sort of a law firm that they've specifically chosen in order to represent them on those grounds. And those are the cases where I expect the issues to get flushed out and be cleared up much more quickly and much more clearly.

18:19and once we've got those positions we will then know what it is but i think one of the interesting analogies here is when photography was brand new originally there was no copyright in photographs at all there was a sort of suggestion that you know in order for copyright to exist there has to be some act of human creativity involved and people thought that well you know pressing a button on a camera there's no creativity involved there so how can that possibly attract copyright so the very first photographs were deemed to have no copyright in them and it wasn't until people started to sort of particularly pose models with particular lighting and particular dress and all the rest of it there was deemed to be enough creativity involved for you know copyright persistent photographs you know there was a there was decades of development of law around how just something as straightforward as photography was treated by the law so you think of you know how much more complicated ai technology is than you know an old silver halide photographic camera in the 1800s it's going to take some time for the law to settle down in these areas and it's not going to be clear cut and all ai is not trained in the same way and all data sets are not licensed in the same way so just because one case goes one way it doesn't mean all other cases are going to necessarily follow suit there's there's so much of this that turns on the facts yeah i think it's such an interesting space because you know i definitely feel like there's this kind of idea of like ask for forgiveness not permission and people are just you know blasting out these ai models that they've scraped the entire internet for.

19:47And I'm sure like, yeah, in the future, some people will criticize that or it'll be interesting to see how that evolves. But it would appear, you know, people are just trying to get the technology out there. Something I would be interested in hearing your opinion on, you know, you're over in Europe right now. What are, like, what role do you think, you know, regulation is going to play in the future of AI, especially in terms of data and ethics and all this kind of stuff? Of course, like you mentioned, the EU is one of the first places that is kind of putting forth a AI regulatory framework right now.

20:16This is definitely something we're talking about over here in the United States. But what if you could make a prediction, how do you think that's going to kind of shake out? How do you think this is going to get rolled out? Yeah. And this is one of the this is one of the nice things that, you know, the fact that I was able to have these conversations with colleagues in, you know, the United States or in Southeast Asia, we get to kind of compare and contrast what's going on in different parts of the world. And so one of the things that I've been saying for a little while around this area is you can see the European mentality of, you know, let's go for control and regulation as the method to control these nascent technologies.

20:52In the first instance, you know, we have the EU AI Act progressing through its sort of final trilogue stages now, and we sort of have got a very good idea of what the final form legislation is likely to look like. very similar in texture to GDPR in many ways it's sort of this idea of use case based regulation and fines if you get it wrong very very big fines as well in the US interestingly it's much more about control via class action litigation and that seems to be the method by which people are kind of trying to settle some of these early sort of concerns around different areas of AI use I think regulation does have a really important role to play though because you only tend to sort of have regulation and policies and standards where you expect people to be doing something you know regulation isn't about saying no to things it's about saying how can we allow people to do these sorts of things so the fact that you have regulation actually creates an awful lot of certainty.

22:00It creates a common platform. It creates public trust. So it means that actually getting people to adopt these solutions in the real world will be, I suspect, easier where people feel like, you know, there is some kind of oversight. There is some kind of ethical standard. There is some kind of transparency, et cetera, that is imposed by, you know, the law and as a sort of standard to what the, you know, good graces of the AI providers may choose or choose not to do. and I also think that that common platform will actually enable inward investment in a much more significant way and we've sort of seen this as well you know people often look to the GDPR as sort of a standard that people will say is you know anti-business or creates some you know difficult hurdles for people to get over actually it has generated a massive amount of confidence in how people can go about creating data safe businesses and there's now been a huge industry around that in europe precisely because of that regulatory environment and i would expect the net effect of the ai act and other regulations that follow on from that to be exactly the same there will be lots and lots of different kinds of you know like say open access open source ai generated in lots of other places i suspect that europe will end up having a bit of a lead by being the place where people feel that they have got the ability to generate and market the safest AI, the ones that customers can feel most assured about.

23:34So we'll see how it plays out, but that's certainly my prediction for what the real world impact of regulation will be. And when people talk about the fact that there isn't regulation elsewhere, I also think that's a sort of slightly false challenge because Because the mere fact that there isn't necessarily something that is called the AI Act somewhere else doesn't mean that there aren't lots and lots of regulations that impact how you use AI in other places. It just might be a much more complex environment or something where, you know, there's a much more political component to how that regulation is imposed.

24:04You can use AI as long as you don't criticize the government, or you can use AI as long as you're doing it within the sort of narrow guardrails of these slightly awkward industry guidelines or regulatory requirements or whatever it might be. So that horizontal regulation will also create some much-needed parity across lots of different regulatory environments. It will be easy to go from using an AI model in healthcare to financial services to media and sports or something like that in Europe in a way that in the UK, which is looking at sector-based regulation, or the US, which is also looking at sector -based regulation, going from reusing an AI in financial services to insurance, which is very similar but separately regulated, or life sciences, which is very differently regulated.

24:53that would be a real challenge that would be that would be a lot more difficult potentially yeah i think that's uh i think that's spot on and actually that's really interesting you say that is a take i have not heard before that more regulation will actually spur more business growth but it makes a lot of sense right you i mean even if looking to past like tech you know cycles or whatnot you had you know when the whole web3 thing was in its you know full steam you had all of the top crypto platforms asking for regulation, asking for clarification, the lack of that definitely gave people an uneasy feeling.

25:25And of course, crypto has had all of its own problems and whatnot. So not exactly correlated that to AI and whatnot. But I think you're spot on where sometimes this regulation, it gives people a lot of confidence. It gives businesses confidence. And I would be really curious to see. Yeah, I mean, that'd be really interesting to see what the investment looks like with that regulation in place. Well, there's one further point on that as well. And it's, you know, if you go from regulation at the kind of global level to the kind of in-house regulation that companies have, what we are definitely seeing is that those companies that have adopted policies and procedures around AI, they have effectively given permission to their people to innovate in a very particular way.

26:10They've sort of said, this is how we want to do it. go and bring us your best ideas and they tend to be the ones who are sort of going and having a whole load of early successes whereas companies that don't have a well-developed policy sheets in place the first thing that people have got to get past with any ai project is am i even allowed to do this at all is this something that this company wants me to do and that's a big barrier that's that you know the lack of a policy is a weird um you know point of uncertainty that people then have to get past. So yeah, it works at the corporate level. It works at the national level.

26:44That's really interesting. One question I'd be curious to get your opinion on in regard to that specifically is, do you think, let's say the EU passes their regulation, do you think that's enough? Let's say opening eyes like, okay, well, we have to comply with that. So now we're just kind of like GDPR, right? We don't necessarily have that in the US, but a lot of businesses in the US, if they're going to do business in Europe, they're GDPR compliant. So it's kind of like the US citizens or people around the world essentially get the benefits of GDPR if the companies operated in those areas. Do you think, let's say like the EU AI Act gets passed or whatever, that those benefits are just going to kind of by default get passed on to everyone?

27:23Or do you think it's going to necessitate still that every country comes up with all their own regulations? um like they will you know so it's an interesting analogy to sort of talk about gdpr because gdpr does have a whole series of explicitly extraterritorial impacts and it was designed that way so that and it's always a point where when you sort of explain this to businesses that don't touch on europe at all the fact that because they're doing business with someone who's in europe or that you know they may end up holding some data that was once in europe they could get caught by this regulation it's always a sort of point of surprise to say the least.

28:03The same will apply with the UAI Act. Again it has provisions within it that make it have explicitly extraterritorial effect and since so many businesses you know you've got 350 plus million very wealthy consumers in Europe that it's not a market that most global organizations can easily ignore. So if you want to be able to access that market, you're going to end up creating AI that's likely to then have potential reuse in Europe, or the outputs might find themselves being used in Europe. And in that case, you've suddenly got yourself caught by the regulatory regime. Wherever it is in the world that you have based that model, wherever it is that you've done your kind of compute runs to train that model, you're now caught.

28:48So there will effectively be this kind of Europe regulates the world kind of roll out as a result of this. Inevitably there's then a kind of question of enforcement. How realistic is it that just because someone is technically caught by the rules that they will meaningfully comply with them if they're based a long way outside of the EU. So I'm expecting that we'll still see an awful lot of people who are kind of technically caught or might technically have to comply but choose not to because you know they don't need to. But certainly any large global enterprise, any organization that is hoping to have the widest possible global market, they will need to sort of be building tools and products that are compliant with this regime wherever in the world they happen to be.

29:34Okay. That's very interesting. Hey, it's been incredible to have you on the show today. As we wrap up, I would love to get maybe one piece of advice from you that you feel like you could give to people that are looking at implementing AI, working in this space, maybe corporations. What's one piece of advice you feel like you could give to these people that are looking at this? I think don't be afraid of putting in place policies and procedures. The more that you do that, the more that you'll end up being confident that you're developing AI tools, AI products, and giving permission to your people to be innovative in a way that is future-proofed, because it forces you to think about who are we as an organization, what do we want to be doing with AI, and how is that going to evolve in the future?

30:18So the more that people do that, the more future-proof they'll be. I love that. That's fantastic advice. Garth, it was amazing to have you on the podcast today. So many great insights. If people are interested in connecting with you or your team, what's the best way for them to go about doing that? So I can be found on LinkedIn if people search for Gareth Stokes or DLA Piper's website is www.dlapiper.com. And we have an enormous number of different AI resources, including a blog called Technologies Legal Edge, which is technologieslegaledge.com. And I have written a whole series of articles expanding on all sorts of different aspects of AI there.

31:01If people would like my opinions for free on a range of AI related topics as well. Very cool. And I'll make sure to leave a link to the website in the description below for the listener. Once again, thank you so much for coming on, sharing your insights. It has been absolutely phenomenal to the listener. Thanks for tuning in to the AI Chat podcast. Make sure to rate us wherever you get your podcasts and have a fantastic rest of your day.