How Occam's razor guides human decision-making.

Area of Science:

• Cognitive Science
• Decision-Making
• Artificial Intelligence

Background:

• Occam's razor, a principle favoring simpler explanations, is hypothesized to guide human decision-making.
• The precise mechanism and adaptive value of this principle in human cognition remain unclear.

Purpose of the Study:

• To empirically investigate whether humans exhibit a preference for simpler explanations when faced with uncertain data.
• To compare human simplicity preferences with predictions from formal model selection theories.
• To explore the persistence and adaptiveness of these preferences in humans versus artificial neural networks.

Main Methods:

• Conducted preregistered behavioral experiments to assess human choices between alternative explanations.
• Utilized formal theories of statistical model selection, incorporating integration over possible explanations.
• Compared human decision-making with the behavior of select artificial neural networks under similar conditions.

Main Results:

• Humans consistently preferred simpler explanations over more complex ones when interpreting uncertain data.
• Observed human preferences closely matched predictions from model selection theories penalizing model flexibility.
• Simplicity preferences were found to be persistent in humans but not in tested artificial neural networks, even when maladaptive.

Conclusions:

• Human decision-making appears to be guided by a fundamental preference for simplicity, consistent with Occam's razor.
• Principles of statistical model selection, such as integrating over latent causes to prevent overfitting, may underpin human cognitive biases.
• This inherent simplicity bias in humans contrasts with the behavior of certain artificial neural networks, highlighting potential differences in cognitive architectures.