Sander L. Koole
Clinical psychology
VU profile for Sander L. Koole (opens in a new tab)What makes an interaction feel like a connection? We explore how AI companions can attune to people—and what that means for bonding and emotion regulation.
A research project at Vrije Universiteit Amsterdam (opens in a new tab)
When people connect, they do more than exchange words. They respond to one another’s rhythms, share emotions, and adjust to each other over time.
Bonding Bots asks how these processes translate to interactions with artificial agents. Bringing psychology and AI together, we investigate the mechanisms of human–AI bonding and the possibility of supporting emotion regulation through interaction.
Our work spans computational models, conversational systems, and human studies. The goal is to understand how connection develops, and to test its possibilities and limits.
Meet the people behind the projectOur guiding question
How can psychologically informed AI support meaningful human–AI relationships?
From theories of human connection to models we can inspect and interactions we can study.
How do people experience closeness with an AI agent? We study emotional sharing and balancing—how an interaction partner can influence the way someone feels.
Human studies help us examine perceived connection, emotional experience, and the factors that shape a relationship with an artificial agent.
PSYCHOLOGICAL FOUNDATIONSWe use temporal-causal networks to represent changing emotional states and their influences. We also investigate predictive coding, in which a system updates its expectations in response to incoming information.
These approaches let us examine adaptation at different timescales and make assumptions about emotional dynamics explicit.
COMPUTATIONAL MODELINGOur conversational research brings language models together with models of emotion. We examine cues in words and speech—including pitch, rhythm, and energy—and how an agent might adapt during an interaction.
A planned study compares an LLM-only agent with agents supported by a temporal-causal network or a predictive-coding model.
SYSTEMS & HUMAN EVALUATIONResearch in progress. Computational simulations help us test mechanisms. They do not, by themselves, establish benefits for people. Human studies examine that next step.
Selected work by members of the project team. Follow a paper to its DOI or VU research record.
More on the VU Research Portal (opens in a new tab)Researchers in psychology and artificial intelligence, working together at Vrije Universiteit Amsterdam and with external collaborators.
Clinical psychology
VU profile for Sander L. Koole (opens in a new tab)Clinical psychology & digital interventions
VU profile for Tara Donker (opens in a new tab)Social artificial intelligence
VU profile for Koen V. Hindriks (opens in a new tab)Emotion AI & computational modeling
VU profile for Edgar Eler (opens in a new tab)Emotion regulation & human–AI interaction
VU profile for Ethel Pruss (opens in a new tab)Curious about interacting with an AI companion for research? Get in touch to ask about upcoming Bonding Bots studies.
An expression of interest is not study enrollment.
A planned online study comparing conversational agents with different approaches to emotion modeling.
The planned conversation study involves voice interaction with an AI agent and questionnaires about your experience. Exact tasks, duration, eligibility, any compensation, and study dates will be provided in the study-specific participant information before enrollment.
No. Contacting the team only starts an enquiry. Joining a study is a separate decision, following the study-specific information and informed-consent process.
This website does not collect study responses or voice recordings. The interest button helps you compose an email to the research team; nothing is sent until you send that email. Study-specific information will explain what data is collected, how it is used, and whom to contact with questions.
Bonding Bots is a research project. This website does not provide therapy, medical advice, or access to an evaluated treatment. We are investigating the possibilities and limitations of human–AI interaction.