The new law allowing the sale of insurance over the Internet in Quebec, which comes into force in 2019, has raised a red flag for David Beauchemin, a doctoral student in computer science at the Faculty of Science and Engineering. Drawing on his knowledge of actuarial science, acquired during his bachelor’s degree, he saw a risk of misinformation for customers.

“How can we inform someone who doesn’t go through a human? There’s already a lot of misinformation out there about insurance, so there’s a risk of choosing the wrong product,” says David Beauchemin.

This observation was the motivation behind the choice of his doctoral project. The young researcher wants to develop a program, based on artificial intelligence (AI), that could explain an insurance contract as a human would. “Our aim is to have a language tool that provides quality information to customers, and does so in an approachable vocabulary,” explains the PhD student, who is affiliated with the Intelligence and Data Institute.

While still in the early stages of his project, David Beauchemin was faced with a major change in the field: the arrival of ChatGPT. “It was a big surprise. We had to reframe the project to see how we could put the program to good use,” he reports.

The doctoral student points out, however, that ChatGPT has no idea whatsoever about the world of insurance in Quebec. “When it doesn’t know something, it will imagine false information. It’s possible to inject it with knowledge, but it’s not certain that it will answer the right thing every time.”

In his project, David Beauchemin is working on a language model. “The program models the way humans write. It’s like a dictation with holes in it, where it has to predict the next word. If we have ‘le chat boit du…’, it’s more likely that the next word will be ‘lait’ and not ‘whisky’,” he illustrates with a smile.

Read more on ULaval Nouvelles