When Should Robots Speak?

Abstract

This study investigates whether proactive robot interventions undermine autonomy compared to reactive interventions during collaborative ideation tasks. A Pepper robot was implemented as a conversational agent capable of generating spoken solution suggestions using a locally hosted large language model, speech recognition, and text-to-speech functionality. The study used a between-subjects design with two conditions: a proactive condition, in which the robot autonomously intervenes based on conversational cues such as silence or signs of struggle, and a reactive condition, in which it only responds when explicitly addressed. Participants completed a collaborative task focused on improving campus life, after which perceived autonomy was measured using an adapted autonomy subscale of the Basic Psychological Need Satisfaction and Frustration Scale. Behavioral data, including robot intervention frequency, timing, and number of ideas generated, were also collected. Results showed no statistically significant differences between conditions, suggesting that proactive interventions do not substantially reduce participants’ sense of autonomy. Groups in the proactive condition generated an average of 7.70 ideas compared to 5.80 in the reactive condition, but this difference was not significant  (p = .067). These findings suggest that unsolicited robot participation may not inherently undermine perceived autonomy in short brainstorming tasks, while highlighting the importance of intervention design.