How Do People Generalize Causal Relations over Objects? A Non-parametric Bayesian Account.

Area of Science:

• Cognitive Science
• Psychology
• Artificial Intelligence

Background:

• Understanding causal generalization is key to human cognition and artificial intelligence.
• Deciding the scope of causal laws and identifying new ones efficiently is a complex cognitive task.

Purpose of the Study:

• Investigate how people determine the generality of causal relationships.
• Explore the features people use to generalize causal laws across situations.
• Develop a computational model for causal generalization.

Main Methods:

• Two experiments involving participants generalizing causal interactions between objects.
• Analyzing participants' inferences based on object interactions and question order.
• Developing a computational model combining program induction and category inference.

Main Results:

• Participants favor simpler causal laws that generalize over categories of similar objects.
• Inference order significantly influences generalization patterns (Experiment 1).
• A consistent asymmetry was observed, with agent features preferentially shaping causal categories.

Conclusions:

• Human causal generalization exhibits systematic patterns, favoring simplicity and agent-centric features.
• The developed computational model successfully explains observed generalization biases and order effects.
• The model provides a plausible mechanism for efficient real-world causal generalization.