Framing the Crowd: How Task Design Shapes Collective Expectation in Crowdsourcing Pedestrian Behavior Change

Understanding how humans form expectations about pedestrian behavior change is crucial for improving predictive models and social intelligence in autonomous vehicles. This paper examines how task framing counteracts cognitive bias and shapes de-biased collective expectations. We conducted a crowdsourced study in which participants predicted pedestrian intentions from naturalistic pedestrian–vehicle interaction videos under different task framings. Results show that neutral (Exp-N) and continuity-emphasis framing (Exp-C) led the crowd to default toward expectations of behavioral continuity. In contrast, transition-emphasis framing (Exp-T) reduced this default tendency and supported greater sensitivity to behavior change when evidence for a transition was present. These findings provide empirical insight into the cognitive reliability of crowdsourced intention judgments by showing that task design functions as a framing mechanism that shapes collective reasoning and that subtle changes in annotation framing can counteract systematic bias, informing the design of annotation tasks for high-level cognitive inference in human–AI systems.