Robert Hogarth: Will AI Force Lawyers to Raise Their Game?

WEBVTT
00:00:05.599 --> 00:00:13.439
Hello and welcome to the Garfield Podcast, a series of conversations with the people who are working with AI to improve access to justice.
00:00:13.439 --> 00:00:27.839
In this episode, Garfield founder and CEO Philip Young talks to Robert Hogarth, founder and CEO of Settle Index, about how his AI-powered service is delivering fast and impartial guidance on what cases should be settled for.
00:00:27.839 --> 00:00:34.960
They also talk about some of the latest developments in the industry and whether AI will eventually force lawyers to up their game.
00:00:39.679 --> 00:00:49.200
Welcome to another edition of the Garfield Podcast, and on this edition, I am honored to be joined by Robert Hogarth of Settle Index.
00:00:49.200 --> 00:00:54.719
I won't uh in any way spoil his thunder by introducing him myself.
00:00:54.719 --> 00:01:08.159
Instead, what I'm going to do is I'm going to hand over to Robert just to begin with a few words explaining what he's doing now, and then we can talk about that and then talk about Robert's career because Robert is a very unusual lawyer.
00:01:08.159 --> 00:01:09.760
He's an entrepreneurial lawyer.
00:01:09.760 --> 00:01:11.680
We need more of those in this country.
00:01:11.680 --> 00:01:17.200
And I think hearing about the journey as well as what Robert is working on at the moment is going to be very interesting.
00:01:17.200 --> 00:01:19.280
Over to you, Robert, to say hello.
00:01:19.680 --> 00:01:24.239
Hello, and thanks very much, Philip, for inviting me onto your podcast.
00:01:24.239 --> 00:01:36.480
So Settle Index, I've been working on it for about five years now, and it's always been trying to do the same thing, which is to work out a settlement value for disputes.
00:01:36.480 --> 00:01:42.239
And I've been learning as a sort of startup entrepreneur that you have to pivot all the time.
00:01:42.239 --> 00:01:49.599
And I always say I've steadfastly not pivoted, but in fact, what's changed is how the software works.
00:01:49.599 --> 00:01:58.000
And I've always been of the view that lawyers should do financial modelling when they're trying to advise clients about settlement.
00:01:58.000 --> 00:02:01.040
And actually, you can't get them to do it.
00:02:01.040 --> 00:02:04.719
And after five years, I've decided you can't get them to do it.
00:02:04.719 --> 00:02:14.240
So what we've managed to do though now is we by incorporating AI, we can build the model and actually now value the claim without any help from the lawyer.
00:02:14.240 --> 00:02:20.159
So what we have now is a product that will tell you the settlement value of your dispute.
00:02:20.479 --> 00:02:32.960
So I mean, this is really exciting for me personally because like Robert, I was in practice for many years doing big complicated commercial dispute resolution, both in litigation and arbitration.
00:02:32.960 --> 00:02:37.039
And one of my frustrations was always the mathematical side.
00:02:37.039 --> 00:02:39.280
And my frustration was sort of twofold.
00:02:39.280 --> 00:02:45.599
One, that my, although I love maths, my mathematical skills were not quite up to scratch to do a lot of the modeling.
00:02:45.599 --> 00:02:48.240
And secondly, that no one wanted to do any modeling.
00:02:48.240 --> 00:02:55.280
And people just wanted to sort of, like a lot of humanities grads, to sort of try and muddle their way through to work out roughly what the claims were.
00:02:55.280 --> 00:03:02.080
And I and I spent a lot of time in practice saying to people, if only we had a product that would do this, if only this existed.
00:03:02.080 --> 00:03:07.280
And so when I met Robert and I discovered that he had built the product, it was like a nirvana moment.
00:03:07.280 --> 00:03:09.039
It was like Paradise Beckons.
00:03:09.039 --> 00:03:13.360
So do you want to unpack a bit more, Robert, how the how the product works?
00:03:13.360 --> 00:03:19.439
Um, because some of our listeners will be lawyers and some of them are technologists, but all of them will be interested.
00:03:19.759 --> 00:03:23.840
Well, obviously, Philip, if I tell you how it works, I'll have to kill you afterwards.
00:03:24.240 --> 00:03:26.639
Yeah, maybe at the high level only, yes.
00:03:27.680 --> 00:03:35.759
The math that it does is classic um Daniel Bernoulli 1738, when he said how you value a chance.
00:03:35.759 --> 00:03:38.879
So we do that, but we sort of do it on steroids.
00:03:38.879 --> 00:03:48.960
So uh the problem with complex litigation and even quite simple litigation is there are a lot of variables that have to be packed into the model.
00:03:48.960 --> 00:03:56.240
And obviously, whether you upload it automatically or you build it from scratch, you can do either with our software.
00:03:56.240 --> 00:04:02.400
You have to build something that can just accommodate almost an entire range of possibilities.
00:04:02.400 --> 00:04:08.159
But the maths is basically what are the chances and what what would the value of the outcome be?
00:04:08.159 --> 00:04:13.599
And you take all the value of the outcomes weighted and add them together, and that gives you your value.
00:04:13.599 --> 00:04:26.959
It it sounds relatively simple, but um, my uh CTO tells me I think we've unbelievable lines of code because it is complex to do it in a sort of automated way.
00:04:26.959 --> 00:04:33.519
And what and what we've done, because exactly following what you say, Philip, that lawyers simply aren't interested in doing this.
00:04:33.519 --> 00:04:41.199
We've been going down a route of trying to make it simpler and simpler to do, which means we have to get the software to do more and more of it.
00:04:41.199 --> 00:04:50.639
So the software completely creates the model now, it creates the decision tree, and you enter the the user enters very high-level data.
00:04:50.639 --> 00:05:06.800
And so, really, it got to the end of its course this year because what I realized was that actually, if we make the models simple enough, AI can extract the data and we can create the models from a little bit of information that we take from uploading the pleadings.
00:05:06.800 --> 00:05:17.759
Well, I think along the way, we've come up with a lot of sort of relatively novel things in the context of decision tree modelling, which actually simplifies the decision tree.
00:05:17.759 --> 00:05:20.879
So if you look at one of our uploaded models now, it looks very simple.
00:05:20.879 --> 00:05:29.120
But basically, if a sort of keen amateur modeler got in there, they could add loads and loads to it, but it wouldn't necessarily make the model look very different.
00:05:29.120 --> 00:05:30.560
So we've simplified it.
00:05:30.959 --> 00:05:52.800
It sounds, I mean, the way you describe it, it sounds, and you've shown it to me, and I know it's incredibly intuitive, but it basically you you up as I understand it, you upload some documents like pleadings, and then you do a bit of QA with it, and then it basically gives you a view um uh of of the sort of range of uh acceptable settlements and what they'd look like.
00:05:53.120 --> 00:05:53.439
Yes.
00:05:53.439 --> 00:05:56.240
Well, I think it's actually moved on since we last spoke.
00:05:56.800 --> 00:05:57.120
Oh really?
00:05:57.120 --> 00:05:57.759
Oh wow, okay.
00:05:57.759 --> 00:05:58.079
Yeah.
00:05:58.560 --> 00:06:01.040
You upload it now and it gives you its view.
00:06:01.040 --> 00:06:02.000
Oh wow.
00:06:02.000 --> 00:06:09.439
Uh and then if if if you want to, and I strongly encourage it, you certainly as the lawyer, you certainly need to review it.
00:06:09.439 --> 00:06:16.959
And if you want to, you can um, you know, you obviously you can change the parameters and and make them your own, as we say.
00:06:16.959 --> 00:06:33.519
But the I suppose the exciting thing about it at the moment is that because we can now put a value on a case by uploading pleadings, I'm just starting working with a law firm to upload cases that have been settled, see how good it is at predicting what happened.
00:06:33.519 --> 00:06:42.720
And that's been slightly the holy grail because over the last five years, every time I talk to anyone who wants to do this at scale, they say, Well, can I put my past cases in?
00:06:42.720 --> 00:06:50.560
And I always say no, because it would have been very complicated to do it if if you had to have the lawyer who knew the case building the model every time.
00:06:50.560 --> 00:06:56.000
But now we can do it at scale, uh, upload past cases.
00:06:56.000 --> 00:07:07.519
And and what I'm expecting is that we'll be then be able to play around with our algorithms a bit and get, as it were, the best fit for for a hundred cases, what is the nearest we can get on average to all of the outcomes.
00:07:07.519 --> 00:07:10.399
So I'm I mean it's it's it's really interesting now.
00:07:10.720 --> 00:07:11.920
Yeah, that's really interesting.
00:07:11.920 --> 00:07:14.879
I mean, that's basically, as you say, it's a best fit thing.
00:07:14.879 --> 00:07:24.000
What you're discovering is what the reality of a statistical sample of cases is, and then you're you're making sure your software reflects that.
00:07:24.000 --> 00:07:44.560
And I'm sure there'll be um a confidence interval so that you know, because there's no one quite right answer to any case, but there is a sort of a a range of reasonable answers, and then and then people using your software will then know it's not just a a clever theoretical mathematical thing, it's also grounded as well in past reality, so it's even stronger.
00:07:44.800 --> 00:07:45.199
Yes.
00:07:45.439 --> 00:07:57.600
What one can't help feeling that this is a sort of tool that isn't just useful for lawyers and their clients and also insurers, but also a terribly useful tool for mediators and in due course, maybe even uh arbitrators.
00:07:57.920 --> 00:07:58.160
Yeah.
00:07:58.160 --> 00:08:00.000
I I mean I'm quite interested in the idea.
00:08:00.000 --> 00:08:09.120
You could put it, you know, most big commercial contracts have this thing where the executive executives will get together and sort of hammer out a deal before an escalation process.
00:08:09.120 --> 00:08:10.399
Yes, an escalation process.
00:08:10.399 --> 00:08:16.800
Well, the escalation process is we'll put the dispute into settle index and before we do it and see what it says.
00:08:16.800 --> 00:08:19.199
And if we don't like that, then we'll go on.
00:08:19.199 --> 00:08:34.879
Because it does produce a completely neutral uh evaluation, it's not on either side, it just says judging by the issues and the claims and and the cost that will be spent if we didn't settle it, that's that's what we think the settlement value will be.
00:08:34.879 --> 00:08:55.120
I mean, obviously it remains to be seen how close we get, but when you were talking about the confidence interval, what I have in mind is trying to get to a position where we can say, well, if let's say from a defendant perspective, you know, if if our software will predict that the settlement will be at or below the prediction in 90% of cases or something like that.
00:08:55.120 --> 00:08:59.919
And then if you're an insurer or something, that would be, you know, really, really good.
00:09:00.320 --> 00:09:04.159
Yeah, really helpful because they can choose a reserve and all sorts by reference to that.
00:09:04.399 --> 00:09:06.000
Yeah, and even settle the case.
00:09:06.320 --> 00:09:07.200
And even settle the case.
00:09:07.200 --> 00:09:08.559
Well, yes, ideally.
00:09:08.559 --> 00:09:23.120
Um I mean, cynical people might sometimes think that insurers don't rush to settle cases because they think there's value in time, but um, but at least knowing the range and then being able to take commercial decisions is just going to be incredibly valuable.
00:09:23.120 --> 00:09:29.279
Um, let me just also say um uh ask a question for those who might be listening who are more international.
00:09:29.279 --> 00:09:38.559
I I as I understand it, this is not just a product that is sort of UK only, it is in theory a product that could be used by lots of other jurisdictions as well.
00:09:38.879 --> 00:09:42.879
Yes, it's it's in fact completely jurisdictional agnostic.
00:09:42.879 --> 00:09:46.879
It just needs a dispute which has issues and numbers.
00:09:46.879 --> 00:09:49.679
And in fact, it can do any language as well.
00:09:49.679 --> 00:09:55.440
I mean, I've uploaded Latin cases to test it and it's that ancient Roman cases.
00:09:55.440 --> 00:09:57.600
I think it can do Welsh as well, yes.
00:09:59.200 --> 00:10:07.120
So I mean that that's that's a very useful point for anyone listening because um England is a is a significant legal market, but America is an even bigger one.
00:10:07.120 --> 00:10:10.639
And um there's sort of global scope for this.
00:10:10.639 --> 00:10:21.519
And I and I think you've touched upon it as well, but you've built AI into this product, and then presumably that's part of the way it passes documents and and analyses them.
00:10:21.919 --> 00:10:31.759
Yes, so we we pass the document to open AI at the moment, and and what that does is it extracts the information we need from it.
00:10:31.759 --> 00:10:37.360
All of the actual probabilities and weighting, etc., is done within our software.
00:10:37.360 --> 00:10:46.399
So what lawyers have to feel comfortable about is the pleadings being uploaded into it's not it's not a public, you know, it's an API version of OpenAI.
00:10:46.399 --> 00:10:51.840
But of course, pleadings are relatively public anyway, and certainly as between the parties, they're completely public.
00:10:51.840 --> 00:10:56.639
So we don't need to upload any confidential documents to get the model.
00:10:57.279 --> 00:11:12.960
And I think in any event, if you're using the enterprise API, which um from what you just said you are, um I know from my own work that they promise the same level of confidentiality in that information as, for example, a cloud server provider to a law firm does.
00:11:13.039 --> 00:11:15.440
So Yes, I think uh that's right, yes, yeah.
00:11:15.679 --> 00:11:18.720
Yeah, so there's no risk there from a sort of privileged perspective.
00:11:18.720 --> 00:11:26.720
Um, but as you say, in any event, you're only seeking the pleadings, you're not seeking um anything that might be more secret than the pleadings if I can do it that way.
00:11:26.960 --> 00:11:32.799
Yes, I mean we oh obviously for cases that are pre-action, you might want to upload the letter of claim and the response.
00:11:32.799 --> 00:11:34.720
And of course, those are private documents.
00:11:34.720 --> 00:11:38.240
Pleadings I've had no problem with, and it uh it's quite interesting.
00:11:38.240 --> 00:11:43.519
You can download pleadings from the court and model the settlement, which which I do.
00:11:43.519 --> 00:11:48.159
That's how I've trained, I mean, that's how I've tested the software is on live cases.
00:11:49.039 --> 00:11:58.399
And the commercial court is in the process here in England anyway, of um opening it all of the case files up more broadly so the public can see things like witness statements and stuff.
00:11:58.399 --> 00:12:00.320
I mean, that's coming next year.
00:12:00.320 --> 00:12:02.879
That's a a big innovation I've been reading about.
00:12:03.279 --> 00:12:11.039
There is a mischievous part of me which thought about sort of starting to publish, you know, case of the week and upload some pleadings and publish a settlement value.
00:12:11.039 --> 00:12:16.639
But I'm I'm not I'm not sure whether that's um would be regarded as contempt of court or something.
00:12:16.639 --> 00:12:19.360
So I I haven't started doing that yet.
00:12:19.360 --> 00:12:20.480
No, no, no.
00:12:20.559 --> 00:12:25.519
Well, maybe if they they make it very easily publicly available, then you would be able to start doing that.
00:12:25.519 --> 00:12:28.559
Um you might you might be doing one or both sides a big favour.
00:12:28.559 --> 00:12:34.639
Um this is Seth Lindex, and it's it's a really exciting product.
00:12:34.639 --> 00:12:39.200
And um Robert's given me a demo of it, and I was blown away.
00:12:39.200 --> 00:12:47.360
And I really think this is sort of very much big part of the future of um of uh disputes um practice in law.
00:12:47.360 --> 00:12:57.919
Um, but let's talk a bit about your career to get to this point because um you're going to be very modest, and I'm going to sort of insist you say a bit about your career, because you were a very eminent lawyer in London in practice as well.
00:12:58.240 --> 00:13:06.320
Well, yeah, I mean, I I was very unenterprising in my legal career because I stayed with the same firm for 42 years or something.
00:13:06.320 --> 00:13:14.159
Um so so in that sense I wasn't very enterprising, but the practice obviously changed hugely over my time.
00:13:14.159 --> 00:13:22.320
And also what I did change, I started off as an insurance lawyer doing claims, as many lawyers at RPC did and still do.
00:13:22.320 --> 00:13:31.679
Uh, and then I sort of in search of my own area of expertise, which always happens in big firms, you have to specialise a bit to get to your own client base.
00:13:31.679 --> 00:13:40.639
I ended up going into construction PI and sort of construction claims, and then from there I ended up working for actual contractors.
00:13:40.639 --> 00:13:44.639
I ended up doing non-insurance construction claims, particularly for Carillion.
00:13:44.639 --> 00:13:52.799
And uh they went bust just before I retired, I don't think entirely because of me, but uh I'm sure nothing to do with you, Bobby.
00:13:52.799 --> 00:13:54.399
Well, it was other claims.
00:13:54.399 --> 00:14:01.519
I'm glad to say we weren't on any of the claims, but they had four big claims which sort of did for them, but um but that was a very good time working for them.
00:14:01.519 --> 00:14:04.240
So that's broadly what I did.
00:14:04.240 --> 00:14:26.879
And along the way, I edited a book on um insurance law for the construction industry, which actually was a in the end, it was really written by you know it was a big range of uh types of policy, so I spread it around the firm, but I edited it and wrote some of it, and uh that was also an interesting um and exercise.
00:14:27.360 --> 00:14:28.559
I imagine it's a lot of work.
00:14:28.559 --> 00:14:34.720
I I've never been able to bring myself to write a legal book, but I've always admired the people that have done.
00:14:34.720 --> 00:14:37.120
And um I remember some years ago, Mr.
00:14:37.120 --> 00:14:47.519
Justice Foxton, as he then was, um, somehow found time in his judicial career to write a book on, I think it was um uh Scrutton, it was one of the earlier well-known commercial judges.
00:14:47.519 --> 00:14:49.200
Yeah, exactly.
00:14:49.200 --> 00:14:53.039
And I I I said to David, I said, How on earth did you find the time to do that one?
00:14:53.039 --> 00:14:56.879
You're a busy commercial judge, and he just sort of shrugged and said, I just sort of fitted it in.
00:14:56.879 --> 00:15:02.399
And I thought, goodness me, I mean, I around a busy practice or a busy judicial career.
00:15:02.399 --> 00:15:13.440
I mean, I I I think in a law firm it might be a bit easier because, as you say, you can divide up the chapters, but you've still got to read and edit and be satisfied with quality and consistency and everything.
00:15:14.159 --> 00:15:18.480
Well, he was probably just cutting and pasting skeleton arguments from his cases, you know.
00:15:19.600 --> 00:15:20.720
Oh, I'm not sure.
00:15:20.720 --> 00:15:26.000
Next time I see him, I'll I'll I'll I'll I'll put that to him and say, ask him if that's how he did it.
00:15:26.000 --> 00:15:32.480
Hopefully he won't sort of chase me out of the end of court with a with a an axe or something.
00:15:32.480 --> 00:15:37.919
Um no, no, and I think you you before you retired, you were the senior partner of RPC.
00:15:38.240 --> 00:15:40.639
I was the senior partner for six years, yes.
00:15:40.639 --> 00:15:44.000
Um and and then went back to, I mean, I never stopped practicing.
00:15:44.000 --> 00:15:49.759
Being the senior partner at RPC didn't carry huge um executive responsibilities.
00:15:49.759 --> 00:15:57.600
I chaired the partners' meetings and I used to mostly have meetings with all the partners to see if they were happy, uh obviously.
00:15:57.600 --> 00:16:01.039
And obviously they all they all were all the time.
00:16:01.360 --> 00:16:02.639
They were all the time, yes.
00:16:02.639 --> 00:16:15.120
I um I I I managed to avoid all of that for some years at CYK, but towards the end of my time um prior to my retirement from CYK, I was the senior partner there.
00:16:15.120 --> 00:16:24.240
And I suspect being senior partner of a sort of seven-partner law firm is a is a less onerous job than being a senior partner of a a much bigger London law firm.
00:16:24.240 --> 00:16:30.320
But I I still I much preferred being in practice than having to um to do the statesman role, I have to say.
00:16:30.639 --> 00:16:30.960
Yes.
00:16:30.960 --> 00:16:39.519
Well, I I said I would have, I think I said I'd have two meetings with every partner every year, and and there was at least 60 partners when I was there.
00:16:39.519 --> 00:16:41.840
So that meant I had to have 120 meetings.
00:16:43.279 --> 00:16:44.240
That's a lot of meetings.
00:16:44.799 --> 00:16:49.440
Quite a job organizing them, but anyway, it was very interesting and uh yeah, it was fine.
00:16:49.440 --> 00:16:50.799
But um, I agree with you.
00:16:50.799 --> 00:16:52.639
I was I'm probably more typical lawyer.
00:16:52.639 --> 00:16:54.879
I was quite happy doing cases really.
00:16:56.720 --> 00:17:09.519
Because you're in AI like I am, because your product like mine has got AI built into it, I thought we'd talk a bit about some of the the recent sort of newsworthy developments in the sort of the law and AI space.
00:17:09.519 --> 00:17:35.440
And the first thing I wanted to just chat about was um in the last couple of weeks, OpenAI announced that they were uh that at first it was reported that they were going to ban uh any sort of use of um their models for medical or legal advice, but then it became apparent when people looked at it that what they really were talking about was not the product not doing that, but the terms and conditions saying you couldn't use it to do that, which is a subtle but profound important difference.
00:17:35.440 --> 00:17:38.400
And I wondered if you had any thoughts on that, um Robert.
00:17:38.799 --> 00:17:41.839
Well, I think you know, for B2B, that's probably meaningful.
00:17:41.839 --> 00:17:44.640
B2C is probably unenforceable.
00:17:44.640 --> 00:17:46.799
I don't I mean, I I I don't know.
00:17:46.799 --> 00:17:51.279
And and obviously we at Settle Index say that we don't give legal advice.
00:17:51.279 --> 00:18:01.519
And now, in complete contrast to that, we upload models which have in, which again you didn't see, but we now have about a two-page legal analysis of the case.
00:18:01.519 --> 00:18:06.079
And I'm amazed by how good OpenAI is at that.
00:18:06.079 --> 00:18:07.519
Absolutely fantastic.
00:18:07.519 --> 00:18:17.039
It it cites cases that I've heard of, not made up, and produces fantastic analysis of cases from pleading.
00:18:17.039 --> 00:18:18.880
So I I'm I'm amazed by it.
00:18:18.880 --> 00:18:22.160
But obviously, we can't give that as advice.
00:18:22.160 --> 00:18:29.519
But but I think that the safety for us is what we're saying is this is an assessment, and we've put these chances and these proportions on.
00:18:29.519 --> 00:18:33.759
Now you need to talk to your lawyer, and he needs to say whether he's happy.
00:18:33.759 --> 00:18:36.160
And if he's not happy, he can change them or she.
00:18:36.160 --> 00:18:40.559
But we've now on our website, you can now upload pleadings as a member of the public.
00:18:40.880 --> 00:18:41.440
Oh, cool.
00:18:42.000 --> 00:18:47.279
And we send you a report on your case, a PDF with a full report.
00:18:47.279 --> 00:18:50.079
And uh the first one has just been uploaded today.
00:18:50.079 --> 00:18:53.039
It doesn't go automatically.
00:18:53.039 --> 00:19:06.160
Uh I wouldn't I'm I'm gonna be the human agent, uh, and I'm not pretending otherwise, but uh you can upload the pleadings and I will then uh put them through our case and send a PDF and it's free.
00:19:06.160 --> 00:19:07.119
It's a free service.
00:19:07.839 --> 00:19:23.119
That's a really interesting idea, actually, because that um that both encourages people to to see what it does and what the outcomes are, but also it gives you useful data yourself that you can presumably use to refine refine your outcomes.
00:19:23.119 --> 00:19:23.759
Yes.
00:19:24.400 --> 00:19:31.759
And it also forces lawyers to use it a bit because, for example, the one today is being uploaded by someone who's a client.
00:19:31.759 --> 00:19:33.839
I mean, they're not it's not by a lawyer.
00:19:33.839 --> 00:19:37.920
And they're gonna take it to their lawyer and say, What do you think of this?
00:19:37.920 --> 00:19:43.440
And and I've said, Well, if you it's fine, you can look at it, but if you want to change it, you you have to subscribe.
00:19:44.240 --> 00:19:48.960
Right, well, quite obviously, I mean you you've got something to sell, you you can't just do everything for free.
00:19:49.279 --> 00:19:52.319
Yeah, sorry, I was gonna say we don't have to do demos anymore.
00:19:52.319 --> 00:20:01.920
I mean, if you want to see what the software does, you just upload the claim documents and we send you back what you would actually see on the screen as a PDF.
00:20:01.920 --> 00:20:10.960
And then if you want to change things or put part 36 offers in or you know, all the other things you can do, then you you get the live software and change it.
00:20:11.279 --> 00:20:19.119
I think this is another topic that is worth talking a little bit about, which is the impact on this sort of technology on lawyers in terms of raising their game.
00:20:19.119 --> 00:20:24.319
But before we before we come to that, I thought I'd just circle back and finish off the open AI point.
00:20:24.319 --> 00:20:26.240
And that I was just gonna say I agree with you.
00:20:26.240 --> 00:20:41.680
I mean, I I think what's changed here is is lawyers who advise on liability who looked at open AI and some of the claims inevitably that will be made against them, because uh any big business is always a recipient of claims, whether meritorious or unmeritorious.
00:20:41.680 --> 00:20:46.079
And someone said we better add some more disclaimers into our T's and C's.
00:20:46.079 --> 00:20:57.359
And I agree with you that um that on a B2B basis they might be enforceable, but certainly from an English law perspective, B2C is a lot harder to enforce those clauses against consumers.
00:20:57.359 --> 00:21:07.839
And I did wonder a bit about, I'm sure OpenAI's lawyers have thought about this, but OpenAI's chat GPT product, its biggest user base, as I understand it, is consumers.
00:21:07.839 --> 00:21:16.400
So I imagine they've done this more for psychological reasons than necessarily expecting it to always stand up in a in a court somewhere in the in the world.
00:21:16.720 --> 00:21:17.759
Yes, I guess so.
00:21:17.759 --> 00:21:24.079
And and probably most of the time, I mean, I would say its biggest offense is going to be that its advice is mostly right.
00:21:24.079 --> 00:21:25.039
Yes.
00:21:26.400 --> 00:21:29.039
Within the margin of appreciation that lawyers are allowed.
00:21:29.519 --> 00:21:30.960
I can't tell you how good it is.
00:21:30.960 --> 00:21:32.240
I mean, it's fantastic.
00:21:32.240 --> 00:21:37.119
I I look at it and I think this is KC standard advice that you get in our models.
00:21:37.119 --> 00:21:38.400
That it's nothing to do with us.
00:21:38.400 --> 00:21:45.599
The AI writes the its appraisal of it, and then we get certain things, information from it, and put it into our model.
00:21:45.599 --> 00:21:50.720
But I mean, obviously, if it does get it wrong and you are relying on it, then I guess there would be a risk.
00:21:51.039 --> 00:21:51.839
There would be liability.
00:21:51.839 --> 00:21:55.680
But I mean, they can always buy insurance, of course, to cover themselves against that risk.
00:21:55.920 --> 00:22:02.319
Yeah, and they they just have another funding round and uh become even more valuable, yes, exactly.
00:22:03.519 --> 00:22:05.440
Yeah, a couple of trillion rather than just.
00:22:06.000 --> 00:22:08.799
I think I don't know whether anyone would in in insure them.
00:22:08.799 --> 00:22:15.839
I mean interestingly, I we carry insurance, obviously, Settle Index includes for sort of errors and emissions.
00:22:15.839 --> 00:22:30.400
And I've just had my renewal, and the the underwriters have got very interested in whether we were using AI, and I still don't know what they're going to do actually, but they they're very interested in whether we're giving legal advice and using AI and everything else.
00:22:30.720 --> 00:22:38.799
There is um a new line of insurance beginning really for Gen AI, and there's a company in London called Testudo.
00:22:38.799 --> 00:22:46.960
Oh I give them a shout out that they have got a product that they offer, and um the insurance always finds uh uh new areas of business.
00:22:46.960 --> 00:22:49.440
That's why you can insure satellites and things.
00:22:49.440 --> 00:22:59.920
And um there will there will be, I think, either sort of endorsements to existing policies uh that that add this sort of risk on, like you get for cyber risk and stuff, or there'll be standalone policies.
00:22:59.920 --> 00:23:08.000
Yeah, let's just go to the the thing I wanted just to quickly touch upon, which I think is interesting, and that is the question of AI causing lawyers to raise their game.
00:23:08.000 --> 00:23:29.519
And picking up your point about how high quality the LLMs can already do, if well prompted, and you know, all of those sort of caveats taken into account, and bearing in mind that that the products that we have got access to are not the frontier products that the labs have got that they're using to train the next generation.
00:23:29.519 --> 00:23:43.039
So that's why sometimes you look on the arenas where they test these LLMs in public and you see things appear that, you know, I saw something a year and a half ago with from OpenAI that was better than ChatGPT-5 is now.
00:23:43.039 --> 00:23:45.519
So, you know, year and a half on, you wonder what they've got.
00:23:45.519 --> 00:23:50.079
And we we project into the future, we say in two years' time, goodness knows what will be available.
00:23:50.079 --> 00:24:02.400
Do you think that the quality of these things, particularly the way they progress, is going to mean that the average quality of lawyers is going to go up or be forced to go up because bad lawyering will no longer be tolerated in the way that sometimes it was?
00:24:02.799 --> 00:24:04.799
Well, I think, yeah, I think that has to follow.
00:24:04.799 --> 00:24:13.039
There is so much knowledge available to these LLMs that when I upload cases, it can be in any in any area, but its quality is the same across all of them.
00:24:13.039 --> 00:24:18.480
And you know, most lawyers are much more specialists in one area and they don't know a lot about other areas.
00:24:18.480 --> 00:24:31.279
I I I think I mean there's obviously an element of judgment and supervision required, and uh it does sometimes come up with funny answers, but I'm sure of that's going to be changing.
00:24:31.279 --> 00:24:40.799
And the other thing that I always find amusing is we expect AI never to ever to make a mistake, but we're quite happy with humans making their lives.
00:24:42.160 --> 00:24:42.880
Yeah, quite.
00:24:42.880 --> 00:24:58.400
I mean, that that's one of that's really, I suppose, in a sense, my point, because um it will come as um no surprise to listeners, particularly who are not lawyers but who have dealt with lawyers, that lawyers in the market always know who is good and who is perhaps not quite so good.
00:24:58.400 --> 00:25:10.640
And every practicing lawyer has come across certain people as opponents who, rather than add any value, um tend to subtract it both from their side and from the sort of the general IQ of the human race as a whole.
00:25:10.640 --> 00:25:17.599
Uh and and they sort of bedevil a lot of work because they make cases harder to resolve than they should be.
00:25:17.599 --> 00:25:29.279
And um, I've been wondering if one unexpected benefit of AI is either that those sorts of people will be supported by a product that will improve their quality or ultimately unable to compete with it.
00:25:29.599 --> 00:25:29.920
Yeah.
00:25:29.920 --> 00:25:33.119
And I I mean, I I hear what you say about lawyers.
00:25:33.119 --> 00:25:48.559
I I certainly came across some very bad lawyers in my time, but I also think really good lawyers are also a menace because in the litigation front, because they will run really clever cases, and they can be just as damaging, uh, you know, uh both ways.
00:25:48.559 --> 00:26:00.400
And what I, you know, and and this is me banging the subtle index wrong, but what lawyers tend not to do is just make a financial model at the beginning and say, oh dear, that looks like what it is, and nothing we do are going to change it.
00:26:00.400 --> 00:26:03.839
I'm very sort of um deterministic these days about disputes.
00:26:03.839 --> 00:26:10.000
I mean, you and I think most lawyers you can look at a set of pleadings, you look at it and say, Well, that's a tricky case.
00:26:10.000 --> 00:26:15.200
Okay, well, tricky case equals pay them a third, you know, or something in in very simple terms.
00:26:15.200 --> 00:26:16.000
That's what it means.
00:26:16.000 --> 00:26:17.839
And nothing you do changes that very much.
00:26:17.839 --> 00:26:26.880
You'll I I think you'll probably get to mediation and you'll end up paying a third, but you've just poured money away trying to move it from 40% to 33%.
00:26:26.880 --> 00:26:32.640
And uh and and the the problem is it's so expensive actually to resolve them.
00:26:32.640 --> 00:26:41.279
And the way we run disputes in this country is you try to settle the case while all the time preparing for a very expensive trial.
00:26:41.279 --> 00:26:47.519
And you know, the process of trying to settle the case now I would say takes a minute uploading the pleadings.
00:26:47.519 --> 00:26:50.640
And the process of preparing for a trial takes two hours.
00:26:50.640 --> 00:26:53.279
So if you sorry, not two hours, two years, say.
00:26:53.279 --> 00:26:57.039
So so if you work for two years, then upload the pleadings, you get one answer.
00:26:57.039 --> 00:27:00.960
If you upload the pleadings and then work for two years, you probably still get the same answer.
00:27:00.960 --> 00:27:08.319
It's just doing the uploading first or you know, modelling at the beginning gives you a grounding in realistically where you're going to end up.
00:27:08.640 --> 00:27:09.279
Yeah, I agree.
00:27:09.279 --> 00:27:09.680
I agree.
00:27:09.680 --> 00:27:16.160
I mean, I always said to clients when I was in practice that the um the first thing they should do is put their emotions to one side.
00:27:16.160 --> 00:27:20.960
And the second thing they should do is think about how do they maximize their net return.
00:27:20.960 --> 00:27:23.359
And if you're a claimant, that means your net recovery.
00:27:23.359 --> 00:27:27.759
And if you're the defendant, that means, unless you've got some giant counterclaim, your net payment.
00:27:27.759 --> 00:27:30.559
And that's a rational way to look at the process.
00:27:30.559 --> 00:27:34.160
But um has to be said that not all clients see it that way.
00:27:34.160 --> 00:27:41.359
Um, some pursue litigation for emotional reasons or or for some other tactical reason unrelated to the actual dispute.
00:27:41.359 --> 00:27:49.519
And of course, there is that minority of cases where they just won't settle because one side either can't pay or won't pay, or there's some other issue.
00:27:49.519 --> 00:28:05.279
But I agree with you that the majority, if you draw a sort of bell curve of the commercial litigation world, the overwhelming majority of cases fall well within the area you're talking about, which is that we should have rational outcomes and rational, essentially rational risk allocation is what it comes down to.
00:28:05.599 --> 00:28:10.079
People very often say to me, well, you know, modelling is just one tiny part of it.
00:28:10.079 --> 00:28:21.039
There's all these crazy humans and uh that what people can afford, and there's poker playing, and some people are very And what what I would say is though, that doesn't mean you shouldn't start from the best place possible.
00:28:21.039 --> 00:28:32.960
I mean, you can be a great poker player and you you might model it and say, Well, I I should only get half a million for this case, but if you end up getting 800,000 because you're a genius negotiator or something, well, great.
00:28:32.960 --> 00:28:41.839
But that doesn't mean you shouldn't have done in fact, it makes you look better, really, if you say the case is worth half a million and you recovered 800,000.
00:28:41.839 --> 00:28:42.480
Good for you.
00:28:42.480 --> 00:28:43.200
Exactly.
00:28:43.200 --> 00:28:45.440
Obviously, not so good if it's the other way around.
00:28:45.440 --> 00:28:46.960
Yeah, yeah, yeah.
00:28:47.200 --> 00:28:47.839
No, you're right.
00:28:47.839 --> 00:29:03.599
I mean, I I I used to sort of say to the younger lawyers at the back end of my career in practice that I used to get quite frustrated when I saw law firm rochers have chess metaphors on them, because I used to say litigation's a game of imperfect information, it's closer to poker.
00:29:03.599 --> 00:29:18.000
And you should see it as sort of betting against chips on the table, which is basically what you're describing, which is if you know that the odds on the table, I don't know, they're they're 500,000 pounds or five million, you therefore know that if you're the claimant, you want to try and beat those odds.
00:29:18.000 --> 00:29:24.960
If not, you accept them, and you make rational sort of mathematical decisions based upon how to maximize your position.
00:29:24.960 --> 00:29:39.599
This has been a really, really interesting conversation, and um I want to encourage anyone who's been listening to to log on to Robert's website to check out Settle Index.
00:29:39.599 --> 00:29:42.720
I'm gonna ask Robert to give us the HTTP in a moment.
00:29:42.720 --> 00:29:46.880
And I I think that this is very much the future of Dispute's work.
00:29:46.880 --> 00:29:58.640
And as we've discussed, it's not just um an English and Welsh thing, it's a global thing, and it's not just limited to big ticket commercial cases, it can be used much more broadly.
00:29:58.640 --> 00:29:59.759
So please go away.
00:29:59.759 --> 00:30:05.680
Check it out, use it, increase the rationality of the outcomes that you provide to your clients.
00:30:05.680 --> 00:30:10.640
So, Robert, do you want to just sort of just say the HTTP and that way they've got everything?
00:30:10.640 --> 00:30:13.519
Well, it's Settleindex.com.
00:30:13.519 --> 00:30:15.359
It's all it's all one word.
00:30:15.359 --> 00:30:16.160
There you go.
00:30:16.160 --> 00:30:18.240
You've got no excuse not to go now.
00:30:18.240 --> 00:30:19.759
Settleindex.com.
00:30:19.759 --> 00:30:21.920
Go and check it out and go and use it.
00:30:21.920 --> 00:30:28.319
So only remains for me to thank Robert for joining me today and for this really interesting conversation.
00:30:28.319 --> 00:30:34.960
And um, if I was in practice, I would be using SettleIndex all the time and I'd be very excited about the future of dispute resolution.
00:30:35.279 --> 00:30:38.319
Well, I think I think you are in practice, aren't you, with Garfield?
00:30:38.400 --> 00:30:40.079
So uh Oh, I keep forgetting.
00:30:40.079 --> 00:30:44.799
Yes, no, it's a sort of odd form of practice because I'm not I'm not front and centre.
00:30:44.799 --> 00:30:48.000
I'm sort of behind the scenes checking everything and product development.
00:30:48.000 --> 00:30:52.319
But yes, I am I am in a form of practice, but just a very unique one.
00:30:52.319 --> 00:30:53.680
Yes, so fair point.
00:30:54.000 --> 00:31:04.160
And and I have been a user, as I've said before, and I I I've I've had a very, very satisfying experience uh recovering uh a debt using Garfield.
00:31:04.160 --> 00:31:05.440
So that was great.
00:31:05.759 --> 00:31:06.799
Well, thank you very much.
00:31:06.799 --> 00:31:15.680
Um we best not go into that on this podcast because there would be a risk that everyone would think that this is um this is me sort of um um blowing my own trumpet.
00:31:15.680 --> 00:31:16.000
Yes.
00:31:16.000 --> 00:31:17.119
No, not at all.
00:31:17.200 --> 00:31:20.400
But so I'll say thank you very much and goodbye then, Robert.
00:31:20.400 --> 00:31:21.039
Thank you.
00:31:21.039 --> 00:31:23.599
Cheers, Philip, and thanks for having me.