Reality Bending Lab › News › Artificial Emotion: Our Symposium at ESCAN 2026
Artificial Emotion: Our Symposium at ESCAN 2026
On the last day of ESCAN 2026, the meeting of the European Society for Cognitive and Affective Neuroscience hosted by Sapienza University in Rome, we ran a symposium organised together with Marco Sperduti (Università di Roma Tor Vergata): Artificial Emotion: Affective Responses and Interaction with Artificial Intelligence.
The premise is simple. Artificial intelligence increasingly produces content and agents that people experience as emotionally and socially meaningful, yet research consistently documents systematic negative biases toward AI. The symposium asked how those biases shape affective, physiological and neural responses to three kinds of thing: culturally salient stimuli such as artworks, evolutionarily salient stimuli such as faces and erotic images, and artificial agents themselves. The four talks drew on experimental psychology, affective neuroscience and social neuroscience, and between them covered self-reports, personality measures, psychophysiology, electroencephalography and functional neuroimaging.
The talks
- The effect of AI-beliefs and personality on the appraisal of evolutionary salient stimuli — Ana Neves & Dominique Makowski (University of Sussex). Four studies in which real erotic images and neutral faces were randomly labelled as photographs or as AI-generated. AI-labelled images were seen as less real and drew weaker, less positive emotional responses, with beauty, attractiveness and trustworthiness ratings dropping for AI-labelled faces. Beliefs about AI and personality traits moderated the effects, pointing to a metacognitive component in how authenticity shapes emotion.
- Cognitive and neural mechanisms underlying the bias toward AI in aesthetic experience — Giulia Torromino (Sapienza Università di Roma). Visual and literary works with manipulated authorship, measured with questionnaires, electrodermal activity, heart rate variability and EEG. The bias against AI turned out to be dynamic rather than intrinsic to the content: it appeared only when an AI-attributed artwork followed a human-attributed one, reversed direction with readers' level of engagement, and disappeared once a human hand was recognised.
- From disclosure to cooperation: mapping the boundaries of affective engagement with social robots — Emily S. Cross (ETH Zurich). In longitudinal conversations, lower mood, loneliness, stress and introversion all predicted greater self-disclosure to a social robot. In prisoner's dilemma games, however, robots' displays of anger or sadness did not steer cooperation, and recognising the robot's emotion accurately reduced it. When and for whom robots provide meaningful affective engagement remains the open question.
- Social engagement with artificial agents: multimodal physiological and neural signatures of natural interaction — Thierry Chaminade (Institut de Neurosciences de la Timone, Aix-Marseille Université). Controlled human–robot and human–human interactions recorded with cardiorespiratory measures, eye tracking, speech and fMRI. Emotional engagement with artificial agents arose from structured interaction rather than surface anthropomorphism, and physiological coupling occurred even when the agent expressed no emotion at all.
The session closed with a panel bringing the four perspectives together. A recurring thread: the label "AI" changes what people feel before any property of the stimulus does, and the effect depends on who is looking, in what order, and how engaged they are. Many thanks to our speakers for a genuinely stimulating afternoon, and to the ESCAN organisers for the setting. Rome in June did not hurt either.
