As an AI researcher, you work with LLMs. What projects are you working on at the moment?
Hendrickx: “We are currently working on two different chat bots. One chat bot is designed to talk to people who want to stop smoking, and the other talks to people who would like to learn more about safe sex. The objective is not for people to only use the chat bots once, as that obviously will not persuade smokers to quit smoking. Rather, it is about the long term. We want to research how a chat bot can interact with a patient in the long term and in a pleasant manner.”
What exactly do you mean by “a pleasant manner”?
“For the patient, they must feel like they are talking to a healthcare provider, rather than an AI chat bot. The underlying LLM must therefore be able to generate answers about sensitive topics, like health and sex. In both projects, we are integrating a specific style called motivational interviewing. This is a technique also applied by healthcare providers. It is a conversational technique intended to change behaviours. To convince people to quit smoking, for instance, based on reasons that are important to them. So, the autonomy lies with the patient.”
What results has this research field recently produced?
“The big question, of course, is whether chat bots are able to communicate like professional healthcare providers. The answer – as recently published research shows – is yes. It is important to realise that this is always intended to support doctors, not to replace them. In our models, we are adding an extra to the existing script. So that we remain in control.”
How does this affect you as a researcher?
“I am impressed by the quality of what a language model can generate. At the same time, it is also somehow frustrating how fast an LLM can develop. I've researched automatic summarisation for years. To be able to summarise a text well, you must understand the entire text well and identify the core message. As yet, current language models do not always do this correctly, but it is still a lot better than a few years ago.”
On Wednesday 5 June, AI researchers will get together at the Radboud University campus to share knowledge on this topic. What are you looking forward to?
“I am very curious about two fields that I do not know much about myself: chemistry and physics. How do they apply LLMs in their research?” Hendrickx laughs: “In language and speech – my field – it is very logical. That is the great thing about AI at Radboud University: it is an interdisciplinary theme that you can learn a lot about from your fellow researchers.”
Sign up
The Radboud AI Meetup on ‘Breaking Boundaries with Large Language Models’ is a follow-up to the very successful Radboud AI event that was held in September last year. The event will be held in English.
When? Wednesday 5 June from 2.30 – 5.00 pm.
Where? The Salon at Huize Heyendael.
Would you also like to join?