Learning From Success or Failure? - Positivity Biases Revisited

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Behavioral Economics

Background:

• Positivity biases, a tendency to overweight positive outcomes, are observed in human learning.
• Computational models, such as Q learning, are used to understand decision-making and learning processes.
• Understanding the interplay between learning biases and behavioral strategies like exploitation and exploration is crucial.

Purpose of the Study:

• To reexamine positivity learning biases using a Q learning computational model.
• To investigate the relationship between these biases and behavioral characteristics of exploitation and exploration.
• To compare a simple asymmetric Q learning model with a time-varying learning rate model.

Main Methods:

• Utilized a Q learning computational framework to model learning processes.
• Incorporated a time-varying learning rate mechanism where rates depend on success/failure magnitudes.
• Analyzed the relationship between model parameters and behavioral measures of exploitation and exploration.

Main Results:

• Positivity biases were present in the simple asymmetric Q learning model but vanished in the time-varying model.
• In the time-varying model, learning rates adjusted to performance (success/failure), reflecting outcomes.
• These adjusted learning rates showed balanced associations with both exploitation and exploration, unlike the constant parameter model where biases linked solely to exploration.

Conclusions:

• The time-varying Q learning model offers a more intuitive explanation of learning biases and their relation to behavior than simpler models.
• Positivity biases are not inherent but may be an artifact of assuming constant learning rates.
• Individual differences exist in how participants utilize different learning models, suggesting flexible strategy selection.