nucleation

THEORY-900: A systematic review of 116 studies of avoidance learning in healthy humans finds the evidence contradicting two-factor and other associative theories and favouring expectancy and inference-based accounts, but also finding residual avoidance after the outcome is visibly removed, which a purely inferential account does not predict

2 min read

Properties6
StatusActive
Version2
Tagslearning-and-conditioning, cognition
SourceLIT-1477 — A systematic review of human avoidance learning: Cognition, computation, and methods
Bears onQUESTION-328 — Does an implemented model-based or latent-cause model of human avoidance, compared on the same data with the model-free two-factor implementations, reproduce the phenomena Mancinelli and Bach rate A or B, including residual avoidance after signalled outcome omission, and does a model without an inferential component fail them?
CorroboratesCLAIM-446 — Some persistent anxiety may be governance-maintained rather than belief-maintained: anticipated threat grants defensive processes authority over action, attention and evidence acquisition; under the avoidance that authority selects, the hypotheses that the situation is dangerous and that it is safe predict the same observation, so repeated safety need not correct the threat estimate; relief reinforces the defense without being evidence of danger; and the regime can outlast a corrected belief, its pathology, where present, being defensive authority insensitive to corrective evidence at avoidable cost

Graph View

Source

Mancinelli & Bach, LIT-1477, read in full in NOTE-1134 from the PsyArXiv preprint: pp. 2, 9–28, and Figs 1 and 4 (pp. 50, 53).

What was actually shown

Against two-factor theory and safety-signal theory:

  • Avoidance is learned when the response cancels the US without ending the CS, in eight studies with no counterevidence (p. 12).
  • An avoidance response is selected for a new CS that predicts the same US, without training (US-based transfer, p. 14), and spreads through learned stimulus equivalence (latent transfer, p. 14).
  • Avoidance appears after Pavlovian conditioning alone, without any avoidance training (p. 16): “the anticipation of a US is sufficient to elicit an AR” (p. 17).
  • Fear responses are not necessary for avoidance to be maintained, and intentions to avoid modulate them (pp. 11, 25).

Against a purely inferential account:

  • “AR are consistently emitted despite signalled US omission, suggesting that avoidance is not fully goal-directed” (p. 22), with no evidence of habits forming through overtraining. Hence avoidance “can not be fully inferential in the first place, directly challenging expectancy theory” (p. 23).

The authors’ summary: “a substantial body of evidence supports expectancy theory and aligns with an inference-based algorithmic framework. However, this perspective is tempered by findings of Pavlovian influences in certain experiments” (p. 28).

What could have come out otherwise: avoidance could have required CS termination, as Mowrer predicted, or could have stopped once the outcome was visibly gone, as expectancy theory predicts.

What this does not say

  • That inference explains clinical avoidance. No patients, and the authors say “all our theoretical deductions are specific to the paradigms” in the literature (p. 28), which are signalled, instructed and simple.
  • That a computational inferential model has been tested. None exists for avoidance: “there is no computational counterpart of expectancy theory” (p. 6).
  • That residual avoidance is habit. No study found the overtraining interaction that would show it; a small subset of participants carries the effect.

Connections

  • CLAIM-446. The claim’s account is inferential at its core (the likelihood ratio under avoidance), and this verdict is favourable to that framing. The residual avoidance after signalled omission is the laboratory case of defence outlasting a corrected expectation, which the claim needs, though its explanation here is Pavlovian rather than governance.
  • THEORY-910. Brown et al. recast expectancy theory in temporal-difference terms; this review says the existing temporal-difference models are model-free and fail the inferential phenomena.

Backlinks