Reward associations do not explain transitive inference performance in monkeys.

Area of Science:

• Cognitive Neuroscience
• Animal Behavior
• Reinforcement Learning

Background:

• Traditional models assume nonhuman animals maximize reward value.
• Model-free reinforcement learning struggles with transitive inference tasks.
• Reward maximization alone doesn't explain all animal decision-making.

Purpose of the Study:

• Investigate if monkeys use abstract representations beyond reward value.
• Determine if monkeys can override reward maximization for rule-based choices.
• Explore the role of implicit list order in animal decision-making.

Main Methods:

• Manipulated reward values to conflict with implicit list order.
• Trained monkeys on a transitive inference task with discordant incentives.
• Analyzed choice behavior to infer underlying decision rules.

Main Results:

• Monkeys accurately made rule-based choices despite conflicting reward signals.
• Model-free reinforcement learning failed to predict high accuracy under these conditions.
• Performance suggests reliance on an abstract, model-based representation of list order.

Conclusions:

• Monkeys' transitive inference is not solely driven by expected reward value.
• Abstract representations of order guide decision-making, overriding immediate reward incentives.
• A method was developed to infer these abstract representations from observed data.