Between ex­cite­ment and com­pos­ure: How one’s emo­tion­al state in­flu­ences how we deal with AI re­com­mend­a­tions

Researchers from the Universities of Paderborn and Bielefeld Present a Study

When do people follow recommendations from artificial intelligence (AI) - and when do they ignore or question them? Researchers from Project A03⁽¹⁾ of the Collaborative Research Centre/TRR 318 'Constructing Explainability' at Paderborn University and Bielefeld investigated this question. They sought to determine the influence of users' emotional states on how they respond to AI recommendations. Their findings revealed that people in a moderate state of emotional arousal, in particular, appear to scrutinise AI recommendations more critically. The study's findings are presented in the paper 'User Arousal and Advice Taking', which has been accepted for the '4th World Conference on eXplainable Artificial Intelligence' (XAI 2026), taking place in Brazil from 1 to 3 July.

In the paper, the researchers - Prof. Dr. Kirsten Thommes, Dr. Olesja Lammert, and Maurice Pape - describe a U-shaped relationship between emotional arousal and the acceptance of AI recommendations: individuals with high or low levels of emotional arousal were more likely to follow the AI's recommendations. Participants with moderate arousal, by contrast, were less likely to follow the recommendations. To explain: in relationship psychology, a U-shaped relationship describes the typical trajectory of relationship satisfaction over a longer period of time spent together. It is very high at the beginning, then drops during the first few years - for example, due to everyday life and routine - and rises again after the critical phase.

Experiment under emotional pressure

For the study, the researchers conducted an experiment with 73 participants. First, the participants put themselves into different emotional states by visualising specific situations and then completing tasks. Whilst one group was given a solvable task, the other was confronted with an unsolvable one. In addition to the participants' subjective perceptions, the researchers measured their heart rates to objectively assess their emotional arousal. "Our results suggest that people in a state of moderate arousal scrutinise recommendations from AI systems more critically. We are currently investigating in further studies whether people with high or low arousal levels are, by contrast, less interested in the AI's explanations," explains Pape. He is presenting the latest findings at XAI 2026.

Potential for medical applications

These findings could prove particularly relevant in future for so-called 'arousal-sensitive' XAI systems - that is, explainable AI systems that take greater account of users' emotional state or situational context. The aim is to better understand when people critically evaluate AI recommendations and when they are more likely to accept them without question. In further studies, the researchers now intend to investigate the influence of factors such as time pressure on emotional arousal and what impact this has on the perception and acceptance of AI explanations. In the long term, the findings are also intended to be applied in medical contexts, for example in situations where people are under stress or experiencing emotional distress.

Anyone interested can contact the project leaders to find out more about the ongoing studies or to take part themselves. Further information is available on the project website.

⁽¹⁾ Project name : Co-construction of explanations between AI explainers and human recipients with and without arousal

This text was translated automatically.