To give you a sense of how quickly artificial intelligence (AI) is progressing, ChatGPT – the large language model AI chatbot – has not only passed the US bar examination, it has done so with marks high enough to put it into the top 90th percentile of candidates. According to researchers from the University of Illinois, ChatGPT’s fourth version aced its exams by ‘a significant margin’. That’s pretty impressive for an AI tool launched only last November.
For underwriters, the speed at which AI is progressing raises some fascinating challenges. Those challenges that formed the basis of Tannie Khupe’s hub session at the AIRMIC conference in Manchester last week. Speaking to an audience of predominantly corporate risk managers, Tannie, our deputy head of client management, set out to provide a whistle-stop tour of the latest developments in AI while asking some searching questions about its potential and the risks it poses.
Unprecedent growth
Today, global spend on AI is around $1 trillion per year while the number of jobs in the sector has grown by 75%. In parallel, the number of devices connected to the internet has climbed to around 15 billion and is continuing to rise. Amid this apparently unstoppable growth, the question for underwriters is increasingly ‘What represents a good AI risk?’ This is important because one survey found that over half of IT professionals believe that ChatGPT itself could be the catalyst of a cyber-attack this year. As a society, we’re moving into uncharted territory with AI-associated risk profiles changing all the time.
Businesses, Tannie said, are embarking on programmes to evaluate AI technologies and consider how they can be utilised in the workplace. For insurers, for example, there’s considerable potential to harness AI for customer service via chatbots and other tools but the technology does present other more fundamental options. Using AI to review claims to detect fraud is high on many insurers’ lists. Perhaps more profoundly, AI could play a significant role in underwriting decision-making. The technology’s ability to sift huge volumes of data looking for correlations and patterns make it an ideal tool for this. The challenge is whether AI is used to assist underwriters in making decisions or the AI makes decisions itself. There is unlimited data available. The skill is in how you interpret that data.
Emerging risks
In terms of the risks created by AI, the use of deep fake technologies poses clear threats. Already the legitimate AI tool Fliki can convert text into videos using convincing avatars. It’s not too great a leap of imagination to see how this type of technology could be applied for criminal purposes such as faking celebrities or politicians.
Algorithms – the mathematical instructions that enable AI tools to function and learn – also have potential flaws. Algorithms ‘feed’ on data; if that data is incorrect or biased, then the results of the algorithm will also be incorrect or biased. This is particularly important when dealing with factors such as ethnicity, gender and health. Already there have been controversies such as the AI system trialled in American courtrooms that gave advice to judges on sentencing but recommended longer terms for people from ethnic minorities. In professional lines insurance, how an algorithm might analyse the risk of a person educated at a prestigious university versus a person educated at a less well-known university is also a sensitive point.
New ideas, new applications
Sompo International, Tannie explained, is working closely with Sompo’s Digital Labs teams in Tel Aviv, Tokyo and Silicon Valley to see how AI can be utilised by the business. He ended his presentation by suggesting some of the ways in which AI could help risk managers and insurers that lie not too far into the future. The technology, he suggested, could be deployed to monitor a construction site and – for example – check that all employees were wearing hard hats through the use of computer vision. Similarly, AI/computer vision could be used in a large shopping mall to identify slip-and-trip-type risks.
From Tannie’s presentation, it’s clear that AI will have a radical impact on not only the insurance industry but our lives in the broadest possible sense. The challenge that Sompo International is addressing is how to harness its power while identifying and managing the risks it creates.