Hello, and welcome to the NVIDIA AI Podcast.
I'm your host, Noah Kravitz. For many of us, the word Reuters means news.
But Thomson Reuters is more than just one of the world's best-known international news services.
This Canadian company also works across knowledge industries, including legal, tax and accounting, and fraud and risk.
And as any information-driven enterprise should be these days, Thomson Reuters has been thinking a lot lately about AI. generative AI in particular.
Here discuss the transformative impact that generative AI will have in the future of work and is already having on how his company works, is David Wong.
David is Chief Product Officer at Thomson Reuters, And lest I kind of try to pick things apart and get things wrong with the details of how Thomson Reuters works these days, and how big of a part of your company news is or isn't, I'm going to do the smart thing and turn it over to you, David.
Thank you for joining the NVIDIA AI Podcast.
We're excited to hear about what Thomson Reuters is these days and... how AI is playing a part of it and will be in years to come.
So thanks for coming on the show. Yeah, thank you for having me, Noah.
And that's right. I am leading the product organization here at Thomson Reuters And you already shared quite a lot up front about what we do.
But let me just maybe share a few of the numbers, some of the key facts.
Absolutely. Yes, people do know Reuters as one of the biggest news agencies in the world, but that's only 10% of our business, 90% of the Thomson Reuters business. is serving professionals.
So those are lawyers, tax experts, accountants, auditors, risk and fraud professionals.
And so that is the vast bulk of our business.
The reason why AI is so exciting for us, and I'll get into it as we dive into the topic today, is because when you look across all the products that Thomson Reuters offers, they generally boil down to solving one of two problems for our customers.
They either help with an information problem and information retrieval problem.
We provide some of the most comprehensive databases to lawyers and techs. professionals out there, products like Westlaw, for example.
And so an information retrieval problem Or we help customers to be able to provide assistance in creating written work product.
So helping to draft a contract, to be able to draft a brief, to be able to complete a tax return, to file a statutory filing, things like that.
And what is generative AI good at? It's good at retrieving information and at writing and producing written content.
And so... It's very imperative for Thomson Reuters because it's what we do.
And if we don't incorporate generative AI...
We think that other solutions will come out that will do a better job.
And so it's very important for Thomson Reuters.
Absolutely. What's the best way to get into this?
We've talked on the show and certainly as we're as we're taping this, you know, this year has been the year of generative AI.
So the things we kind of went over, I think folks out there might, depending on their experience level, might associate some of these ideas of summarizing text and creating generating new text with uh certain brand names let's say right but there are a lot of different uh ways to go about this, a lot of different models out there, a lot of different ways to implement large language models and other generative AI techniques.
So maybe you can tell us a little bit about what goes on at Thomson Reuters.
Across the company and in your role as CPO, are you... training LLMs?
Are you building end user software? What does the product portfolio look like?
Yeah, well, maybe I'll start and talk a little bit about our customers first.
And one of the reasons why this customer base is, I think... so eager and has so much opportunity to adopt AI solutions.
So yes, there's been tons of activity in the marketplace around generative AI.
There's been a number of research reports.
There's one which has been circulated quite a lot from Goldman, which tried to estimate what is the the amount of disruption by different industries, different types of jobs.
Other economic research think tanks have done similar work. legal lawyers were at the top of that list.
I think it was number two out of 40 in terms of amount of potential disruption.
And I think it comes down to the nature of work for a lawyer or for a tax professional.
Because if you kind of boil it down and I wasn't trained as a lawyer.
I was trained as an engineer. I just got the pleasure of being able to spend a lot of time with lawyers.
The work involves... Substantial research, substantial amounts of reading, synthesis of facts and logic and concepts, ultimately resulting in the production of work product, which is generally written.
And so if you think about that, the life of a lawyer as best as a non-lawyer can is It already suggests that generative AI is likely to play a role in many of those tasks.
And so that's sort of a very roundabout way about answering your question.
Because what we do is we try to do. I think it's important to interrupt you.
I think it's important for people who don't know what, you know, you haven't walked a mile in lawyer's shoes.
I don't want to turn this into the anti-Shakespeare defending lawyers podcast.
But, you know, if you haven't done it, you don't know.
I've been working a little bit with a legal tech firm. uh, some legal tech folks this year.
And so I have a little bit of a taste, but yeah, it's a lot of reading, a lot of synthesizing and a lot of writing, as you said.
That's right. And it's incredibly important that you have the right information, that you have the right facts, that you're interpreting it correct, because it's in many cases, life and death situations that are being dealt with, or if not life and death, very material issues.
And so across our portfolio, maybe just to focus on legal, but many of the same parallels exist in tax.
What we offer are products that help provide the information.
So we have digital solutions which allow you to research, search, query, as well as the history of law, as well as cases within the different jurisdictions that we support.
We provide practical guidance. So we have lawyers on staff that have written practical guidance and checklists, templates, process maps around legal work. and tooling, which integrates with products like Microsoft Word, Microsoft 365, to be able to do the actual work.
So when you're in Word, we have plugins, for example, that allow you to draft and analyze documents.
And so... We have what we call a combination of research products and legal tech products.
And most recently, It's, I think, pretty publicly known now.
We acquired CaseText, which was the first legal AI assistant called Co-Counsel as part of TR, and they just joined our family a short three months ago.
Oh, congratulations. How's the response been from the customers?
And just because we're talking about it, but don't, if there's a better example, don't stick to legal or don't feel that you have to, but How are the lawyers and the legal tech professionals, legal ops professionals?
How are folks? I don't know if the right word is adjusting to or responding to.
You know, I think of it when I have conversations with my non-technical colleagues leaning friends, I think, in terms of the evolution of autocomplete when we've gone from spellcheck to finish the word, to finish the phrase.
I feel like more and more lately I'm seeing it and it's an adjustment, right?
So how's the customer response? Yeah, it's a great question.
Let me start by sharing a bit of the anecdotal viewpoint, because I've been speaking with law firm leaders and technologists all throughout this year about the trends they are seeing within their own firms.
And one story that really sticks out to me is that lawyers... are almost demanding the solutions that this is one of the few technology changes where it's driven by lawyers and associates, partners and associates that are wanting to get access to the technology.
So the tech leaders at these firms can't keep up.
And they're just trying to say, how on earth am I going to respond to this demand?
They have compared it to when the BlackBerry came out. which seems like a million years ago.
So what, 2005? I think around that you've got one behind you, I think.
I know, I was going to say we do audio only, but I've got an old BlackBerry up on my shelf and a former life I spent a long...
A lot of time with phones, we'll leave it that way.
But I associate BlackBerrys with lawyers.
So there you go. Yeah, it was transformative when the BlackBerry came out because of the productivity that BlackBerry's offered to lawyers.
And... the same kind of trend is showing up right now.
And it's a little bit, you know, just a shift of tax.
It's not quite as fervent on the tax side, but there's still a lot of demand as well.
And so that's actually, I think, really exciting because it's not that common that legal tech is demanded by the front lines.
Usually it's something that, you know, the technology teams say, hey, why don't we use this?
It'll make us more efficient. Now it's coming from lawyers directly.
Now, all that being said, it's not, all of an excitement and positive sort of story there is also trepidation that's what actually drove us So Thomson Reuters, we launched a future of professionals study, which surveyed legal tax and risk professionals across the world to understand what was the sentiment in a more scientific way.
So we did a comprehensive survey of 1,200 different legal, tax, risk, and fraud professionals across the different markets that we serve to see how do they feel?
What's positive? What's less positive? What can we do to best support them through all this change?
Would you learn what were the takeaways?
Well, the high level. Sorry, we don't have all day.
I'll give you the top line. Otherwise, we'll spend the next 20 minutes just talking about survey results.
Exactly. But the top line is that, one, generative AI is viewed as being the most transformative trend in probably the past decade.
10, 20 years for the legal profession. It was pretty unanimous across different types of lawyers, whether you're in a firm or if you're in corporate counsel or if you're in the government.
The way that it will impact the way that lawyers work was across three different dimensions.
Number one, there was a strong belief that it would make lawyers more productive.
So there was a strong, strong belief, yes, this will make me more productive.
It will let me do work faster and better.
Number two was there was concern and a degree of fear around safety and trust to those systems.
We've all heard about the... The famous now chat GPT failed experiment in trying to prepare a legal brief and it hallucinated fake cases.
Right. And then the third is a significant amount of debate, I would say, around the change in the business model. lawyers, particularly in firms, classically, they charge by the hour.
And if you're let's say, significantly more efficient and you don't need as many hours, what does that do to your business model?
And so that's a pretty dramatic shift that is being contemplated.
Is there more to that? Was there more, this is when we go top level, as you said, now we can dive in, right?
Well, what did the data say? Are people in favor of a different model or are we kind of headed that way anyway?
It's a conversation I've had in, you know, if you think about, there's probably a better way to say it, I think about it in terms of you know, knowledge industry work or work that depends on language a lot, right?
So legal profession, anything involving writing, marketing, that kind of thing.
And thinking about the same thing, what's your unit of work that you're measuring on?
And that, I think, opens up conversations.
And I'm curious to see if either in your study or just in your own work sense, you you've thought or what you've thought about this.
I'm sure you have the role of the worker.
And I don't know as much about how law firms and, um, accounting firms are set up in sort of a more marketing creative line of work you might have a shift from somebody being a writer or a graphic designer to kind of more of a creative director role. where you can deploy these virtual tools to do, you know, whatever extent of the work for you.
Is there a similar kind of thinking that you're seeing in some of these other industries likely They go in text?
Well, there's a whole lot in that question.
Yeah, yeah. No, I'm just trying to know.
This is what happens. Oh, no. Let me break it down into a few different areas.
Well, first off, the survey did show that there was a debate.
So... We asked viewpoints from the response to say, are you optimistic or pessimistic? about the transformation.
Because you had almost universal agreement that it was transformative. the nature of the transformation, there were varying viewpoints.
And it ranged from people being so optimistic that this was going to be usher in a new era of nirvana for professional work that is going to solve all the ills of the past.
You also had some very pessimistic respondents that said that this was an apocalyptic situation, that it was the end of the profession as we know it. and everything in between.
What I've heard from firms on this is it's a little bit more measured, which is that the nature of work is likely going to change incrementally because tasks will get replaced, not jobs.
Everyone generally believes that. jobs are not going to be replaced tomorrow by generative AI, that it's going to be augmentation and that individual tasks will get replaced.
The big question is, as more and more and more tasks get augmented or accelerated or replaced, by these AI tools, what ends up happening?
And that's where I think there is a big question.
There are implications on talent. So how do you hire and train new lawyers and new professionals in the future?
There's implications on the billing models and the business models.
So Do you shift from, you know, an hourly basis to a fixed fee basis or start to charge for technology or things like that?
And there are implications for career paths as part of it. what does it look like when you go from associate to a partner or from an analyst to a partner within a within a law firm or an accounting firm.
So a number of different impacts. I think the one which that I personally at least subscribe to the most is that it is very likely that professionals that use AI are going to replace professionals that don't use AI.
You know, it's almost like going back in time and imagining, what if you hired an accountant that didn't know how to use spreadsheet or Excel?
Right. They probably wouldn't do the job that you needed to today versus someone who did all their bookkeeping on giant books by pencil.
So there's probably going to be that shift.
I think there's also likely to be a shift towards more value-based business models.
So think about even today, when you go to account and you ask for a tax return to be completed, you generally don't get charged by the hour, usually get a fixed fee.
Right. Because it's now quite efficient to be able to do a tax return.
The software is there. Even when an accountant is doing a tax return, they have lots of support, lots of software around it.
And so I think more kinds of engagements are going to become fixed fee or value-based instead of hourly because you'll be more efficient.
You'll be able to produce it more predictably.
I also think this is more of a broad statement about the industry, that it will create more work I think if you look at economic history, anytime there's been a dramatic increase in productivity, it's generally created more activity for that industry.
And so I suspect what you'll probably see is more legal work done more efficiently, more accounting work more professional work out there and potentially more easily accessible.
And so imagine with AI assistance, any person, any consumer might be able to get high profile legal assistance for a fraction of the cost because the AI can help to make people more efficient.
I think that's ultimately a really good place to get to because access to justice is a perennial problem.
Absolutely. As long as it doesn't spur an increase in the number of frivolous lawsuits, wow, we're all good.
I'm all for it. My guest today is David Wong.
David is the Chief Product Officer at Thomson Reuters.
David studied engineering and physics at the University of Toronto undergrad, and then moved into product leadership roles working at a few different companies in technology, including Facebook, before joining Thomson Reuters about three and a half years ago now.
Is that right, David? Yeah, it's just past three and a half years.
Gotcha. So as we kind of look ahead and as we're recording this, we're getting into that end of year.
Spotify Wrap just came out. I'll just leave it at that.
What do you see on the horizon for your work, for Thomson Reuters' work with AI, generative AI.
And generally speaking, if you want to go there to kind of the future of work, as we've been talking about a little bit, kind of let's say over the next, you know, one or two out to maybe five years, I think.
And if you don't want to go five years, I understand it's tricky business. these days, especially predicting anything in the future with AI.
But, you know, maybe it's just a continuation of what you were talking about a moment ago, but where do you see things headed?
Yeah. Well, I think the next one or two years... I think we have a pretty good idea, at least within Thomson Reuters.
So... We have made a pretty broad set of announcements to the marketplace of what we're going to be creating and what we're going to be building.
This year, we announced that we would introduce generative AI into all of our flagship research products.
So the research products that we offer for... case research for legal in the U.S.
Westlaw, practical law, as well as our tax research product, which is called And so we have been furiously working on introducing generative AI capabilities into those products.
I think in a previous podcast, you had a discussion around retrieval augmented generation.
I kind of think of us now as like the retrieval augmented generation company because...
That's what all those solutions are, right?
It's a huge database where we retrieve information and facts and we synthesize, read, and we respond to tough questions complicated technical and legal questions based on that information.
And so those are going to be released over the next few quarters.
We are also working furiously to introduce and expand the legal AI assistant to the marketplace.
And so I mentioned before co-counsel, co-counsel, which is the legal AI assistant that was developed by case text.
Right. We are integrating and enhancing and expanding co-counsel across our product offerings.
So we want to make co-counsel available in all of our products, wherever we have AI experiences.
And we also want to make it available where lawyers do most of their work.
So that includes integrating with the Microsoft suite.
And so that's a pretty aggressive roadmap that we have for this year.
Yeah. So yeah, you said the next one or two years.
So we have a lot of work to do. And that's just for legal.
We want to do that in tax and we want to do that in risk and fraud and we want to do that for other professionals.
Does your purview touch the news division?
So my specific purview actually does not touch the news division, but I work with those teams quite a lot.
I realized that except for my reference to, you know, everybody knows Reuters for news.
We hadn't actually talked about news, so I wasn't sure if that's a...
And that's a podcast, a series of podcasts in itself, you know, especially with election season around the corner and in the U.S. and all that.
Absolutely. Yeah. Where I think there are connections is that our teams, because we do a lot of editorial work within our teams, which means that we have to collect... facts and information about how the law continues to evolve.
So the law is a living thing, right? Every day there's new cases which get ruled on that impact The law needs to be analyzed.
It needs to be editorially enhanced and incorporated into our different systems.
So there's some parallels there where the production of news and the production of our content have some similarities.
So we're looking at how we work together with the news team.
But the broader impact of generative AI on journalism, I am not the expert to speak to that.
That is a big, big topic. All right, well, we'll wait and see how this podcast comes out in the wash, and then we can talk about getting one of your colleagues on to dive deep into how do we...
How do we report on and consume the news in, you know, gosh, 2028?
Who knows? So you mentioned the importance, and I think the importance anywhere when you're working with information, but particularly in an area like the law. where material or life and death consequences, as you mentioned, are a real thing.
Generative AI, there's been a lot of advances in a short amount of time all across the board, but as relates specifically to accuracy and to training models, using techniques to only receive information from certain sources, that kind of thing.
How do you, when you're talking about a company as big as Thomson Reuters and areas as big as, you know, research tools for the law, etc., How do you think about trust and safety and ethics and AI, and how do you actually approach it in your hands-on work?
Yes, it's a big question because it covers not just how we build our products, and how we operate as company, but also how we interact with our customers and the industries that we serve.
So what I would say is I'd break it down into a few different areas.
There are the policies and the guiding principles that we have for AI development.
So at Thomson Reuters, we have our AI and ethics principles, which are very similar.
You can probably see Microsoft's version, Google's version, Facebook's version.
They all, I think, share a common lineage and some of the principles there.
That's also related to our privacy and our data protection policies, which generally for legal and for tax use cases are quite high.
So, you know, our clients are Our customers are conservative by nature.
And so we want to make sure that we adhere with some of those standards.
But also importantly, the legal industry is famously known for self-regulating through things like bar associations and professional associations, ethics codes, things like that.
And that is where there's a lot of interesting work that's happening today.
And I think the United States is leading the charge here around practical solutions guidance and ethics guidelines and best use guidelines for the use of Gen AI.
And so we're starting to see that, that courts are starting to create rules Professional associations, bar associations are creating rules and there are now ethical guidelines which are starting to be drafted within The industry.
And we try to take an active participant role there because, you know, we're there to serve these industries.
Right. And we want to make sure that we're keeping abreast of everything that's happening.
But I think all three are critically important.
And for us... and the company building for these industries, we're trying to make sure that we're actively participating as part of that work.
It's understood if you want to deflect this question, but not to get meta, but have you gotten into... copyright disputes around training data with lawyers or those conversations.
I would imagine those conversations with experts from the legal community might take on kind of a different dimension than the way I've been thinking about them relative to, you know, if a songwriter's song lyrics get scraped as part of training data, what does that mean?
Oh, I'm happy to speak to it. Because one, the intellectual property of I think standards, rules, precedents, et cetera, I don't think have been fully established.
And so as a result, Thompson Reuters, we've been taking a pretty conservative view on that.
And so, for example, this is one reason why we very much use this retrieval augmented generation approach.
We are trying as much as possible to use pre-existing models augmented with information, either through providing context and prompts or by that fine-tuning. for some of the solutions so that we're not in a tricky situation where we're training with somebody else's data or we're training with our data and having to determine you know, what are the intellectual property rights that are associated with it?
So we've been... a little bit more conservative in some of the products they've been creating.
That being said, we are doing research to understand what are the pros and cons of different techniques.
I think the second thing we've been trying to do is to try to make sure wherever possible to make sure that our systems are very well segregated by a customer so that we don't have intermixing.
There's a lot of concern about, well, if I'm law firm A, Is my data going to be intermixed with law firm B or, God forbid, the client's? that might be opposing each other and things like that.
And so segregation is also very important to make sure that we can be very clear that this instance, this particular user is segregated from this particular user, for example.
So there are a number of techniques we've been looking at there.
But I don't think there is a clear picture yet around what the intellectual property is. landscape will evolve to.
And so we're just trying to prepare for different outcomes.
David, for folks who would like to know more about all of the different things that you have your hands in thompson reuters has their hands in when it comes to transforming the way we work with knowledge and retrieve knowledge and all this good stuff There's a wealth of information out there.
The website is the best place to start. Where should people start?
Absolutely. So I make it, I'm going to make it easy.
It's www.tr.com slash AI. Perfect. Thomson Reuters, TR, artificial intelligence, AI.
That's it. It's just tr.com slash AI. And that's where we have a lot of what we're doing I believe we have linked to our future professionals report through that site.
You can also learn about all the work that we're doing doing there, as well as our trust principles, the data and ethics policies that we have and and everything we're doing to try to make these solutions not just really usable, but also really safe.
Because as I mentioned before, the legal industry is about life and death issues.
And it is a high bar. I've been in a lawsuit and I've been audited.
So I think your focus on trust and safety is excellent in both those areas.
All right, well, David, this has been fascinating.
And I can only imagine that by the time this podcast goes up, let alone we talk again, there'll be more to cover.
There is kind of an interesting... I'm sort of wondering myself, as much as I keep hearing that the pace of research isn't slowing down...
It seems like the customer end of things, you know, There was this mass adoption and interest in the chatbots and ChatGPT came out and not that people aren't using it, But there does seem to be a little bit of a lag because as you've talked about, we've talked about there are so many associated things to figure out when you're using a technology that's this impactful and new, at least in adoption. you start using it and you start thinking about everything from workflow to, you know, the ethics of intellectual property.
So fascinating times ahead to be sure. But Thomson Reuters will be right out in the front leading the way along with the other movers and shakers in the knowledge industry.
And we're grateful for your work. So, David, thanks for coming on and sharing some of what you've been up to and what you're looking forward to and encourage everybody to.
Keep an eye on tr.com slash AI, follow you on LinkedIn, and keep up with the latest.
Yeah, it's a pleasure. Thank you for having me.
And I can't be more excited about the work that we're doing.
It's some of the most fulfilling work that I've done in my career.
So exciting times. Thank you.
