Limits in decision making arise from limits in memory retrieval.

Area of Science:

• Cognitive Science
• Machine Learning
• Decision Making

Background:

• Human decisions involving probabilistic outcomes are often based on retrieving a limited set of relevant memories.
• This selective memory retrieval can introduce noise, potentially hindering optimal performance.

Purpose of the Study:

• To investigate the impact of memory retrieval processes on decision-making accuracy.
• To determine if idealizing training data improves human performance in probabilistic predictions.
• To compare human performance with machine learning models under similar conditions.

Main Methods:

• Comparing human prediction accuracy when trained on actual versus idealized data distributions.
• Developing machine learning classifiers that mimic human memory retrieval processes (selective, stochastic sampling).
• Analyzing the effects of immutable bottlenecks in memory retrieval on decision noise.

Main Results:

• Optimal performance is unattainable with selective memory retrieval due to inherent noise.
• Humans exhibit higher accuracy in probabilistic predictions when trained on idealized data.
• Machine learning models that utilize all training data do not benefit from data idealization.
• Modified machine classifiers simulating human retrieval patterns replicate human performance.

Conclusions:

• Human rationality has inherent limitations imposed by memory retrieval bottlenecks.
• Idealized data presentation can enhance human accuracy in classification tasks.
• Findings have implications for training professionals in fields requiring probabilistic decision-making.