Behavior analysis and revaluation.

Donahoe et al. (2000) built a computer brain model. They wanted to see if 'revaluation' could happen without linking responses to rewards.

Revaluation is when the worth of a reward changes, so the behavior changes too. The team used fake neurons instead of the usual 'response-outcome' story.

The model showed the same revaluation patterns seen in real animals. No response-outcome wires were needed; just neuron layers adjusting weights did the job.

The paper says we can dump the old 'A leads to B' map and still explain why behavior shifts when reinforcers gain or lose value.

Greer et al. (2019) offered the Resurgence-as-Choice model. They claim resurgence is a choice driven by comparing response values. W et al. quietly challenge them by showing value can shift without any choice process.

Ritchey et al. (2023) tested resurgence in humans by shrinking alternative-reinforcer size. Their data fit RaC2 but with errors. The neural-network view says those errors disappear once you stop forcing a choice rule into the code.

Cudré-Mauroux (2010) doubts whether conditioned reinforcers strengthen behavior at all. W et al. agree that value can change without 'strengthening' links; both papers push analysts to stop treating reinforcers as simple behavior fuel.

If value changes live inside neural networks, not in response-outcome cables, you can stop hunting for the 'magic contingency.' Focus on history, context, and relative value instead. Next time a client's problem behavior resurges, check what recently re-valued the reinforcers, not just whether the response still 'pays off.'