Help Privacy Policy Disclaimer
  Advanced SearchBrowse





Effects of Threat Imminence on Learning Under Uncertainty


Dayan,  P       
Department of Computational Neuroscience, Max Planck Institute for Biological Cybernetics, Max Planck Society;

Fulltext (restricted access)
There are currently no full texts shared for your IP range.
Fulltext (public)
There are no public fulltexts stored in PuRe
Supplementary Material (public)
There is no public supplementary material available

Satti, M., Cichy, R., Schuck, M., Dayan, P., & Bruckner, R. (2022). Effects of Threat Imminence on Learning Under Uncertainty. Poster presented at 51st Annual Meeting of the Society for Neuroscience (Neuroscience 2022), San Diego, CA, USA.

Cite as: https://hdl.handle.net/21.11116/0000-000B-35CF-1
Survival in non-stationary aversive environments is critically dependent on adapting to their inherent uncertainties. Uncertainties appear in different forms: irreducible variability (expected uncertainty) and systematic but unpredictable environmental changes (unexpected uncertainty). An optimal learner dynamically regulates learning according to both forms by weighing prediction errors using a varying learning rate (LR; which should be higher under unexpected uncertainty). An inability to adjust the LR flexibly can lead to learning biases, one source of which is internalizing disorders like anxiety disorder and depression. An important distinction is made in theoretical models in the anxiety literature (the Threat Imminence Continuum) between anxiety evoked by distal threats and fear evoked by immediate proximal ones. Anxiety and fear might be associated with different degrees of learning. In particular, we hypothesize that the often intentional vagaries of the behaviour of proximal threats eliciting fear demand a high LR, even when other aspects of the environment are stable. This idea has yet to permeate the learning literature, and so we explore adaptive learning under varying levels of threat and uncertainty. We designed a new online game-based behavioral experiment in which human participants save themselves by placing a flaming torch in the path of an attacking predator, but get more points the longer they wait before moving the torch. To succeed, they must therefore learn the arrival location of the predator under expected and unexpected uncertainty. To manipulate threat imminence, three predators attack at different times, corresponding to distal (anxiety), medium, and proximal threat (fear). Data were analyzed using regression and a normative computational model. We first investigated whether participants responded differently to different predators and found that the alacrity with which they moved the torch to protect themselves was modulated by the type of threat encountered. Participants consistently reacted faster to proximal threats; this increased their chances of survival, suggesting that they adaptively respond to threats (Mann-Whitney U = 0.0, p = 0.0002, two-sided). In line with the hypothesis sketched above, we found an effect of threat type on the LR, with elevated LRs after encountering proximal threats (fear) compared to distal threats (anxiety) (U = 26, p = 0.037). To conclude, fear induced by immediate threats might raise the subjective unpredictability of threat. Therefore, learning biases might be particularly prominent for proximal compared to more distal threats.