How causal information affects decisions.

Area of Science:

• Decision-making
• Machine Learning
• Causal Inference

Background:

• Causality is crucial for prediction and intervention, with machine learning enabling causal model discovery from data.
• The utility of machine-learned causal models for human decision-making remains largely unexamined.
• Existing psychological studies on causal models lack the complexity of machine learning outputs and real-world contextual interpretation.

Purpose of the Study:

• To investigate whether machine-learned causal information improves human decision-making in everyday contexts.
• To explore how prior domain experience influences the impact of causal information on decisions.
• To assess the interpretability and utility of complex causal models for laypersons.

Main Methods:

• Conducted experiments on Amazon Mechanical Turk involving decisions on diet, health, and personal finance.
• Compared decision-making with and without causal information across different domains.
• Varied participant domain experience to analyze its interaction with causal information.

Main Results:

• Causal information sometimes led to worse decisions compared to no information.
• Individuals without prior domain experience benefited from causal information, while those with experience performed worse.
• Causal information decreased confidence in experienced individuals and increased it in inexperienced ones, potentially due to domain familiarity.

Conclusions:

• Machine-learned causal models do not automatically enhance everyday decision-making.
• Further research is required to bridge the gap between causal inference capabilities and practical decision support.
• Domain familiarity, rather than model comprehension, may significantly influence the utility of causal information.