Computation noise promotes cognitive resilience to adverse conditions during decision-making

Abstract

Random noise in information processing systems is widely seen as detrimental to function. But despite the large trial-to-trial variability of neural activity and behavior, humans and other animals show a remarkable adaptability to unexpected adverse events occurring during task execution. This cognitive ability, described as constitutive of general intelligence, is missing from current artificial intelligence (AI) systems which feature exact (noise-free) computations. Here we show that implementing computation noise in recurrent neural networks boosts their cognitive resilience to a variety of adverse conditions entirely unseen during training, in a way that resembles human and animal cognition. In contrast to artificial agents with exact computations, noisy agents exhibit hallmarks of Bayesian inference acquired in a 'zero-shot' fashion -without prior experience with conditions that require these computations for maximizing rewards. We further demonstrate that these cognitive benefits result from free-standing regularization of activity patterns in noisy neural networks. Together, these findings suggest that intelligence may ride on computation noise to promote near-optimal decisionmaking in adverse conditions without any engineered cognitive sophistication.

Efficient sampling and noisy decisions

“…Crucially, in our work we provide a direct link of the necessity of noise for systems that aim at optimizing decision behavior under our encoding and limited-capacity assumptions, which can be seen as algorithmic specifications of the more realistic population coding specifications mentioned above ( Nikitin et al, 2009 ). We argue that our results may provide a formal intuition for the apparent necessity of noise for improving training and learning performance in artificial neural networks ( Dapello et al, 2020 ; Findling and Wyart, 2020 ), and we speculate that an implementation of 'the right' noise distribution for a given environmental statistic could be seen as a potential mechanism to improve performance in capacity-limited agents generally speaking ( Garrett et al, 2011 ). We acknowledge that based on the results of our work, we cannot confirm whether this is the case for higher order neural circuits, however, we leave it as an interesting theoretical formulation, which could be addressed in future work.…”

Gated recurrence enables simple and accurate sequence prediction in stochastic, changing, and structured environments

“…Finally, there is substantial interindividual variability which does not exist in the optimal solution (Khaw et al, 2021;Nassar et al, 2010Nassar et al, , 2012Prat-Carrabin et al, 2021). In the future, these suboptimalities could be explored using our networks by making them suboptimal in three ways (among others): by stopping training before quasi-optimal performance is reached (Caucheteux & King, 2021;Orhan & Ma, 2017), by constraining the size of the network or its weights (with hard constraints or with regularization penalties) (Mastrogiuseppe & Ostojic, 2017;Sussillo et al, 2015), or by altering the network in a certain way, such as pruning some of the units or some of the connections (Blalock et al, 2020;Chechik et al, 1999;LeCun et al, 1990;Srivastava et al, 2014), or introducing random noise into the activity (Findling et al, 2021;Findling & Wyart, 2020;Legenstein & Maass, 2014). In this way, one could perhaps reproduce the quantitative deviations from optimality while preserving the qualitative aspects of optimality observed in the laboratory.…”

The effects of base rate neglect on sequential belief updating and real-world beliefs

“…Systematic manipulations of evidence order such as the ones we introduced here may thus be helpful in clarifying the computational mechanisms underlying delusions. Further, our results also emphasize that alterations in noisy-sampling (and other limited-capacity) inference processes should be evaluated as candidate explanations for maladaptive or pathological beliefs, particularly given increasing support for their role in adaptive behaviors [ 97 , 98 ].…”