Spontaneous biases enhance generalization in the neonate brain

Area of Science:

• Cognitive Science
• Developmental Psychology
• Artificial Intelligence

Background:

• Inductive generalization is crucial for adaptation in novel environments for both biological and artificial intelligence.
• Spontaneous generalization in inexperienced animals suggests the role of innate predispositions (evolutionary biases or priors) in learning from sparse data without reinforcement.

Purpose of the Study:

• To investigate whether newborn chicks possess evolutionary predispositions that facilitate generalization from limited sensory input.
• To determine if these predispositions are independent of experience, reinforcement, or supervision.

Main Methods:

• Neonate chicks were exposed to an artificial social partner of a specific color.
• Generalization patterns were subsequently assessed across different color ranges (red-yellow and blue-green).
• Observed generalization was modeled using a Bayesian approach incorporating predispositions as initial preferences.

Main Results:

• Generalization patterns were inconsistent with an unbiased learning model, indicating the presence of biases.
• Observed biases included asymmetrical generalization gradients, preference for unfamiliar stimuli, varied learning speeds, and faster learning for evolutionarily infrequent colors.
• The data supported a Bayesian model where predispositions are updated through experience.

Conclusions:

• Newborn chicks demonstrate an evolutionary preparedness for generalization, driven by innate biases.
• These biases are independent of direct experience, reinforcement, or external supervision.
• Both biological and artificial intelligence can leverage biases tuned to infrequent stimuli, like specific colors, to solve the problem of induction.