Veni Research Grant for AI Scientist Enrico Liscio

AI that adapts to what matters to you
Enrico Liscio is investigating how AI systems can recognise, during a conversation, what matters to individual users and how they can adapt their responses accordingly. By combining technical AI development with experiments involving real users, his research project contributes to the development of AI that is transparent and flexible, and operates in line with societal expectations.

AI systems such as chatbots are increasingly being used to support people in sensitive sectors such as public services and healthcare. To be trustworthy, these systems must not only avoid unsafe behaviour but also respond in a way that aligns with human values, such as safety and autonomy.

Not everyone has the same needs
AI systems learn from vast amounts of online data. As a result, they often reflect the most common opinions and ways of thinking, rather than adapting to the person they are communicating with. You have probably experienced this yourself: sometimes an AI system gives a general answer that does not quite match your situation. At best, this is slightly frustrating. In more sensitive situations, such as voting advice or healthcare, this can have much more serious consequences. For example, prior to the last Dutch elections, the Dutch Data Protection Authority noted that popular chatbots gave disproportionate attention to the largest political parties, whilst smaller parties were rarely mentioned.

The challenge: recognising human values
This raises a fundamental question: how can AI systems understand what really matters to the person they are helping, and adapt accordingly? According to Liscio, the key lies in human values. Values such as autonomy, safety, fairness or respect for tradition play an important role in the choices people make. At the same time, these values are difficult for AI to recognise.

“Values differ from person to person, change depending on the situation and are often not explicitly stated,” says Liscio. “Moreover, people are not always aware of their own values until they consciously reflect on them.”

AI that adapts during conversation
Liscio is developing AI systems that can recognise these values during a conversation and tailor their responses accordingly. In previous research, he developed, together with an interaction designer, a prototype exploring how users could reflect on and communicate their values through chatbot interfaces. Building on these insights, Liscio’s Veni project will develop new methods enabling language models to recognise users’ values from conversations and generate responses that better align with what matters to them, whilst ensuring the system remains transparent and users retain control.

Applications with a social impact
Liscio is testing its approach in two areas of application where personal values play a major role. The first application is voting advice. Political preferences are often linked to different value priorities. A better understanding of those values can help provide more transparent advice that supports people in reflecting on their own preferences.

The second application focuses on lifestyle coaching for people with diabetes. Here too, personal considerations play an important role, for example, the balance between health, independence and daily routines. An AI system that understands these differences can offer support that is better tailored to the needs of individual users.
Enrico Liscio: “I hope this research will contribute to AI systems that do not give everyone the same answer, but can adapt responsibly to different people and situations, whilst remaining transparent and aligned with what users truly value.”