Multiple weak biases support adaptive choices without prior experience: a self-supervised strategy.

Area of Science:

• Behavioral Ecology
• Developmental Psychology
• Comparative Cognition
• Artificial Intelligence

Background:

• Newborn animals' survival depends on adaptive decision-making without prior experience.
• Recent evidence suggests early preferences are weak, transient, and variable, challenging traditional ethological focus on strong innate biases.

Purpose of the Study:

• Introduce a novel model explaining how multiple weak biases support adaptive choices in naive animals.
• Propose a self-supervised strategy for decision-making using internal predispositions and environmental structure.

Main Methods:

• Review empirical evidence on early animal preferences.
• Develop a computational model of self-supervised decision-making.
• Analyze the role of weak, multiple, and co-occurring biases in guiding behavior.

Main Results:

• Weak predispositions targeting rare features reduce false positives.
• Multiple weak biases enhance decision accuracy by integrating co-occurring cues.
• The co-occurrence of multiple rare cues enables effective choices in complex, misleading environments.

Conclusions:

• A simple self-supervised strategy using multiple weak biases provides a new framework for understanding early decision-making.
• This model has broad implications for understanding behavior across various scientific disciplines and AI.