The relentless march of technology seems intent on proving that, given time, even the wildest imaginings of science fiction will take on the timbre of suggestion rather than conjecture.
The release of ChatGPT-4 in March 2023 and the torrent of speculation, commentary and forecasts that it has precipitated is a timely example of this. For David Pryce (pictured), managing partner at Fenchurch Law, these conversations represent an interesting extension of a question spearheaded by Professor Richard Susskind in the 1980s – what impact will technology have on professional services?
“It seemed that every time he came up with an idea, it would seem outlandishly sci-fi only for people to look back 10 years later and realise that his ‘outlandish’ predictions were actually now business as usual,” he said. “Susskind has been convinced for a long time that technology, and now AI, is going to have a significant impact on professional services, and I would extend that to insurance.”
The prevailing wisdom has long remained that technology will emerge to fix ‘simple’ problems and automate ‘simple’ processes but will not be able to replicate the inherent complexity of professional services. But Pryce noted that, as Susskind has been saying for a long time, there is quite a significant knowledge gap regarding exactly what constitutes a truly bespoke role.
“He contests that the reality is that while about 90% of people might say their roles are outliers on the bespoke end, actually only about 10% of roles are on that bespoke end,” he said. “So, while the starting point might be to say that only humans could ever write really good policy wordings, actually I don’t think that’s true. And I do think that there’s going to be an increasing impact of AI, particularly in relation to policy wordings.”
Exploring the implications of that statement, Pryce highlighted that policy wordings need to be assessed across four different dimensions – scope, clarity, certainty and whether the terms of the policy are “unusually onerous” – and that each will respond slightly differently to the uptick of AI in the advisory space.
“We look at the scope of the cover being provided by the policy,” he said. “We look at clarity and we look at certainty – which might sound similar to each other but are not the same. With clarity, we examine whether it’s clear what the policy means. But with regards to certainty, it may be completely clear what the policy means but the outcome may be uncertain for a particular reason… Then lastly, we assess whether the terms of the policy are unusually onerous.
“And when I'm thinking about what AI could do, I think it can help with clarity and making what the policy says clear – and I think that probably applies across all policies. I think it can probably also help with making sure that no terms are unusually onerous because you can easily model a variety of examples of clauses that would be particularly onerous on a policyholder, and teach the system to measure and spot such examples.”
Where AI excels is in detecting conditions precedents and applying those to a variety of different clauses, Pryce said, and so it can be leveraged in a meaningful way to address the clarity and the controlled terms of policy wordings. However, where he can’t see AI (currently) making significant inroads is when it comes to scope of coverage and the matter of certainty.
“With certainty,” he said, “there is simply too much that sits outside the four corners of the insurance policy that you need to take into account in order to apply the insurance policy. At the moment, I don’t see how you really point the AI system at where to look. But as I’m saying that, the system is adapting and improving so maybe there is the potential to point it to existing law reports etc and see what happens.”
Examining the limitations of AI with regards to scope of cover, Pryce added the caveat that a move towards standardised wordings would increase the ability of AI to handle this dimension. However, he said, from a London market perspective, that’s the last thing that the industry would want to do.
“If you go to other insurance markets around the world, a lot of them do have much more standardised policy wordings than we do,” he said. “But aside from the depth of experience and specialism you get in the London market, one of the things that really distinguishes us from other insurance markets around the world is our flexibility and ability to write bespoke risks.
“If you were to go towards standardised covered, you could probably get an AI bot that would produce market-leading standardised cover in a particular area, but what you couldn't do is produce an infinite variety of market-leading wordings that you can pick and mix for that bespoke element.”
Critical to bear in mind, Pryce said, is that when you’re creating bespoke cover, it’s not just about optimising the wording from the policyholders’ perspective, it's also about what's commercially achievable. You could feasibly get a bot to come up with an amazingly wide policy that no insurer is prepared to write, which is not to the benefit of any policyholder.
AI is likely to be able to make some improvements to insurance policy wordings, he said, but is unlikely to be able to replicate the distinctly bespoke approach that the London market takes to insurance policies. As a result, perhaps the most natural application of AI is to finetune what already exists rather than replace it entirely, given its implications as a tool for refining wordings and finding inconsistencies.
“Those are two things it could be very good at,” he said. “Because there’s absolutely no reason why you couldn’t get the bot to ensure that every part of the policy is consistent with itself - both internally consistent, and then also consistent with everything else that's been produced by a particular insurer or potentially being produced across the market.”
The bot could also potentially produce a report in relation to where significant differences arise across the market, he said. Because it may be that different insurers don’t want to standardise their wordings as compared with each other, but that doesn’t mean they don’t want to benchmark their wordings against the wider market.
“Fundamentally,” Pryce said, “I actually think that AI has got the capability of making pretty big inroads into this area, as soon as the technology gets good enough to make it meaningful and useful. But, at least in the time span I'm looking at, I can’t see it replacing the need for insurance policies in the London market being written by real people.”
What are your thoughts on this story? Feel free to share them in the comment box below.
