Continuous time causal structure induction with prevention and generation.

Area of Science:

• Cognitive Science
• Causal Inference
• Human Learning

Background:

• Causal learning research predominantly uses atemporal data, neglecting continuous-time systems.
• Learning about preventative causal influences in dynamic systems remains under-explored.

Purpose of the Study:

• Investigate how humans use temporal information to infer generative and preventative causes.
• Examine the interaction between generative and preventative causes in causal mechanism learnability.

Main Methods:

• Participants observed causal devices with interventions and spontaneous activations.
• Human causal structure learning was analyzed within a hypothesis space combining generative and preventative relationships.

Main Results:

• Participants demonstrated strong learning capabilities, identifying most causal relationships.
• Attribution errors were observed, indicating specific learning challenges.
• A computational model approximating normative inference captured human judgment patterns.

Conclusions:

• Humans can learn complex causal structures involving preventative influences in continuous time.
• Cognitive models based on simulation and local computation can explain observed learning behaviors.