Suboptimal Choice in Pigeons: Stimulus Value Predicts Choice over Frequencies

Area of Science:

• Behavioral neuroscience
• Animal cognition
• Decision-making under uncertainty

Background:

• Pigeons often display gambling-like behavior, choosing options with infrequent but reliable reinforcement signals over those with more frequent or higher overall reinforcement.
• A proposed mechanism suggests that the perceived value of a stimulus signaling reinforcement is prioritized over its frequency, leading to suboptimal choices.

Purpose of the Study:

• To test the hypothesis that pigeons prefer stimuli signaling reliable reinforcement, irrespective of signal frequency.
• To investigate the role of signal reliability versus overall reinforcement magnitude in pigeons' choice behavior.

Main Methods:

• A two-alternative choice task was employed, pitting a Discriminative alternative (signaling 100% or 0% reinforcement probabilistically) against a Nondiscriminative alternative (signaling 50% reinforcement consistently).
• Signal frequencies for the Discriminative alternative were systematically reduced (50% to 12.5%), and reinforcement probabilities for the Nondiscriminative alternative were increased (50% to 100%) across experimental phases.

Main Results:

• Pigeons consistently preferred the Discriminative alternative, even when its reinforcement signal frequency was significantly reduced.
• Preference for the Discriminative alternative diminished only when both alternatives offered equivalent probabilities of reinforcement.
• This indicates that signal reliability, rather than overall reinforcement magnitude, strongly influences choice behavior.

Conclusions:

• Pigeons' suboptimal gambling behavior is robustly linked to a preference for reliable reinforcement signals, even at the cost of overall reward.
• The terminal link stimulus value model provides the most parsimonious explanation for this observed decision-making bias.
• Findings contribute to understanding the neurobiological and cognitive mechanisms underlying choice and reinforcement learning.