Testing one or multiple: How beliefs about sparsity affect causal experimentation.

Area of Science:

• Cognitive Science
• Causal Inference
• Machine Learning

Background:

• Understanding causal systems is crucial for scientific discovery and artificial intelligence.
• The control of variables (CV) strategy, manipulating one variable at a time, is a common approach to isolate causes.
• The efficiency of CV strategy in causal learning is questioned, especially in complex systems.

Purpose of the Study:

• To investigate the efficiency of different causal discovery strategies.
• To determine if humans adapt their causal learning strategies based on system properties like sparsity.
• To compare the control of variables (CV) strategy with multi-variable testing in sparse and dense causal systems.

Main Methods:

• Developed an optimal actor model to identify the most efficient causal discovery strategy.
• Conducted behavioral experiments to observe human causal learning strategies.
• Manipulated the density (proportion of actual causes) of causal systems presented to participants.

Main Results:

• An optimal actor model shows that testing multiple variables simultaneously is more efficient in sparse causal systems.
• Behavioral experiments reveal that participants adapt their strategies, using CV in dense systems and multi-variable testing in sparse systems.
• A tendency to overuse the CV strategy was observed even in sparse causal systems.

Conclusions:

• Causal learning efficiency depends on system sparsity; multi-variable testing is optimal for sparse systems.
• Human causal learning strategies are sensitive to sparsity, demonstrating adaptive behavior.
• Despite adaptive tendencies, cognitive biases may lead to suboptimal strategy selection in certain causal learning scenarios.