Amitai Shenhav, Matthew M. Botvinick & Jonathan D. Cohen (2013), Neuron 79(2):217–240 — DOI-10.1016/j.neuron.2013.07.007
Key takeaways
- Control has three component functions. Regulation is the control signal acting on lower-level processing. Specification is deciding which signal (identity) and how strong (intensity). Monitoring is detecting the information that specification needs. The theory gives dACC monitoring and specification, and gives regulation to lateral PFC and other structures.
- The decision maximizes the expected value of control. EVC is the sum of the outcomes’ values weighted by their probabilities, with value defined recursively and future rewards discounted, minus a cost of control that rises with signal intensity. The control system picks the signal with the highest EVC.
- Control is costly. “[M]ental effort is assumed to carry intrinsic disutility.” People avoid control-demanding tasks and will forgo reward to do so. dACC’s output can be read as “a willingness-to-pay signal in the currency of cognitive control”.
- Default override is the general case. Stroop colour-naming, exploration, foraging and patient intertemporal choice all require overriding what would happen without control. The EVC model predicts dACC engagement in each, and the review cites evidence for each.
Standing in the record
Filed on 2026-10-03 at the owner’s request, for the record’s coverage of behavioural control. It is the record’s main reading of the view that control is a distinct decision process with a dedicated substrate: a specification system that weighs payoffs against effort costs and directs separate regulators.
The boundary. The paper is a theory of a brain region and of human cognitive control. Its formalism borrows from reinforcement learning and optimal control, and it notes that EVC “generalizes what is referred to as a Q-value”. It gives no instruction for machine-learning practice, so it belongs here.
That makes it the record’s clearest counterweight to THEORY-056. The theory
holds that regulation is one motive state checking another, with “no
distinct regulating system above them”. EVC posits exactly a distinct
system that decides control, though not one that regulates: specification
is kept apart from regulation. The theory is about emotion and EVC is about
cognitive control, but the paper lists “modulators of emotion” among the
parameters a control signal can set, and treats override of
emotion-driven automatic responses as within dACC’s remit. The two accounts
are not tested against each other, so no rivals is declared.
NOTE-446 sets out where they meet.
It also stands beside Ainslie’s précis (LIT-586). Both treat patience as overcoming a default preference for immediate reward. EVC treats it as costly control allocated by a separate decision. Ainslie treats it as bargaining among interests, with no organ of will, and his précis argues against “organ” models of exactly this kind.
No instruction for machine-learning practice; nothing here belongs in the anthology.