Controlling human causal inference through in silico task design.

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Neuroscience

Background:

• Causal inference is vital for human learning and survival.
• Previous research focused on environmental influences, not strategic manipulation of causal inference.
• The brain's flexibility supports causal inference, but control mechanisms are underexplored.

Purpose of the Study:

• To develop and validate a task control framework for orchestrating causal learning task design.
• To investigate if causal inference factors can be manipulated for strategic control.
• To explore the potential for targeted behavioral outcomes through controlled causal learning.

Main Methods:

• A two-player game setting was used, with a neural network interacting with a human causal inference model.
• The neural network (task controller) was trained to generate experimental designs and manipulate task variables.
• Experiments involved 126 human subjects to validate the framework's impact.

Main Results:

• The task control framework successfully accommodated complex environmental causal structures.
• Task control significantly improved human performance and learning efficiency.
• The learned task control policy mirrored human one-shot learning capabilities.

Conclusions:

• A novel task control framework enables strategic manipulation of causal inference.
• This framework demonstrates potential for enhancing human learning and achieving targeted behavioral outcomes.
• The findings highlight the interplay between artificial intelligence and human cognitive processes in causal learning.