Social inferences from physical evidence via bayesian event reconstruction

Area of Science:

• Cognitive Psychology
• Social Cognition
• Computational Modeling

Background:

• Humans excel at social inference from observed behavior.
• Social inferences can also be made from physical evidence left by unseen agents.
• This capacity may stem from mental event reconstruction.

Purpose of the Study:

• To investigate how people make social inferences from physical evidence.
• To propose and test a computational model of mental event reconstruction for social inference.
• To quantitatively assess the model's accuracy against human judgments.

Main Methods:

• Developed a Bayesian computational model for action understanding.
• Tested the model's predictions against human inferences in controlled experiments.
• Utilized physical evidence (e.g., cookie crumbs) to infer agent behavior and presence.

Main Results:

• Human inferences about agent origin and goals align with model predictions.
• Explicit reconstruction of agent actions predicted entry point and goal inferences.
• Distinguishing between one or two agents was possible based on evidence patterns.

Conclusions:

• Mental event reconstruction is a key mechanism for social inference from physical traces.
• The proposed computational model accurately captures human social cognitive abilities.
• This research illuminates how social information is extracted from the physical environment.