Reinforcement learning
Do reward cues push us to approach rather than just to act?
Open access · cc by · source: Europe PMC
Cues that predicted money made people more likely to approach but less likely to withdraw, so Pavlovian cues bias specific kinds of action rather than simply energising behaviour.
Study at a glance
- Design
- Human experiment — Within-subject go/nogo task in approach and withdrawal blocks, with instrumental training, Pavlovian conditioning and a transfer test; choices fit by RL models compared with group-level Bayesian model selection.
- N
- N=46 · 46 healthy adults analysed after exclusions from 54 recruited; 24 did the 'throwaway' and 22 the 'release' withdrawal version.
- Population
- Healthy adult volunteers from the Berlin area
- Outcome
- Probability of go versus nogo responses under Pavlovian background cues; fitted RL model parameters (learning rate, go biases, reward and punishment sensitivity, Pavlovian weights)
Structured fields used in claim comparison tables when every cited study has a complete layer.
Key findings
Reward-predicting backgrounds increased go responses when go meant approach but reduced them when go meant withdrawal, and loss-predicting backgrounds did the opposite; the approach effect appeared in 45/46 people. People had a fixed bias against active withdrawal, but the model with separate learning parameters for approach and withdrawal was rejected, so learning itself did not differ. Rewards moved learning much more than punishments, whose average effect was indistinguishable from zero. The best model predicted choices with an overall predictive probability of 0.7544.
Methodology
Adults learned by trial and error whether to act (go) or hold back (nogo) for mushroom pictures, in one block where acting meant collecting the mushroom and one where acting meant getting rid of it. They were separately taught that fractal images with tones predicted monetary gains or losses. In a final test those images appeared behind the mushrooms without feedback. The authors fitted a family of reinforcement-learning models to every choice and picked the most parsimonious one by group-level Bayesian model comparison.
Limitations
The task cannot tell whether the effect comes from the verbal label of the action ('collect' versus 'throw away') or from the physical movement toward or away from the stimulus, and the authors did not measure participants' insight. The money at stake was small and the stimuli artificial, so it is unclear how far the result generalises to real-world approach and avoidance. Neural and genetic mechanisms (dopamine, serotonin) are only discussed as predictions; nothing biological was measured, and questionnaire measures of anxiety and depression showed no link to the model parameters.
How this study connects
Role on claims
Each row is a claim on a concept or method page where this paper supports, challenges, or qualifies the statement. Roles are hand-checked — not a model guess.
Valence interacts with approach/withdrawal, not just with value.
Reward and punishment are not mirror images: reward-predicting cues increased 'go' responses for approach but reduced them for withdrawal (approach effect in 45 of 46 people), and rewards moved learning much more than punishments, whose average effect was indistinguishable from zero.
Evidence for the claim as stated.
Whether punishment carries a real learning signal: in the approach/withdrawal task punishment's average learning effect was indistinguishable from zero, while adults in the adolescence study used punishment-context information that adolescents lacked.
Evidence for the claim as stated.
Open questions
Tensions this paper is part of
From concept pages' “where studies disagree.” Disagreement means the same question; scope means different assays, populations, or outcomes.
Whether punishment carries a real learning signal: in the approach/withdrawal task punishment's average learning effect was indistinguishable from zero, while adults in the adolescence study used punishment-context information that adolescents lacked.
Related papers in this topic
Same topic cluster — not a recommendation engine.