Nonlinear neural network dynamics accounts for human confidence in a sequence of perceptual decisions

Abstract

electrophysiological recordings during perceptual decision tasks in monkeys suggest that the degree of confidence in a decision is based on a simple neural signal produced by the neural decision process. Attractor neural networks provide an appropriate biophysical modeling framework, and account for the experimental results very well. However, it remains unclear whether attractor neural networks can account for confidence reports in humans. We present the results from an experiment in which participants are asked to perform an orientation discrimination task, followed by a confidence judgment. Here we show that an attractor neural network model quantitatively reproduces, for each participant, the relations between accuracy, response times and confidence. We show that the attractor neural network also accounts for confidence-specific sequential effects observed in the experiment (participants are faster on trials following high confidence trials). Remarkably, this is obtained as an inevitable outcome of the network dynamics, without any feedback specific to the previous decision (that would result in, e.g., a change in the model parameters before the onset of the next trial). Our results thus suggest that a metacognitive process such as confidence in one's decision is linked to the intrinsically nonlinear dynamics of the decision-making neural network. A general understanding of the notion of confidence is that it quantifies the degree of belief in a decision 1,2. The simplest context for studying confidence is the one of perceptual decision making. In psychology and neuroscience, the most commonly used experimental protocols are the ones of two alternative forced choices (2AFC) and stimulus discrimination tasks in which, in a sequence of trials, the participant is presented with stimuli and has to make a binary choice, associating each stimulus to one of two categories (e. g. decide if a visual stimulus is the image of a cat or a dog). Many studies have tackled the issue of confidence measurement in perceptual decision tasks, either by directly requiring participants to provide an estimation of their confidence 3-5 , or by using postdecision wagering (subjects can choose a safe option, with low reward regardless of the correct choice) 6-8. Postdecision wagering has been used in behaving animals in order to study the neural basis of confidence 9-12. In order to model the neural mechanisms underlying the decision-making process, two main routes are followed. The most frequently used considers (linear) drift-diffusion models (DDM) 13,14 or independent race models (IRM) 15-17 , in which choice-specific cells accumulate evidence in favor of one or the other alternative to which they are tuned. A more biophysical approach considers attractor neural networks 18 , with competing pools of cells, leading to a nonlinear dynamics with choice-specific attractors. Within one or the other framework, researchers have tried to relate confidence to the decision-making process, making different hypotheses on the origin of confidenc...

Adaptive neurons compute confidence in a decision network

“…Another type of models define the confidence as the absolute difference between the firing rates of neuron population selective to the decision options at decision time, where the firing rates are produced either by race model 10 or dynamic attractor model 15 . The race model based confidence explains the activation of the human ventromedial prefrontal cortex 10 and the dynamic attractor model based confidence successfully reproduce the observations in monkey experiments 6 and human confidence in a sequence of perceptual decisions 16 . However, how the neural circuit calculates the absolute difference between neuron pools, i.e., how the confidence forms during the decision, is unclear.…”

Modelling decision-making biases

“…For example, Lo and Wang (2006) used a neural network model of cortico-basal ganglia dynamics to show that adjustments in response caution (decision thresholds) could be implemented by adjusting the strength between synapses of the cortico-striatal connections. Moreover, nonlinear attractor models can explain posterror response time biases in the absence of feedback (Berlemont and Nadal, 2019) and confidence-related sequential effects observed in empirical data (Berlemont et al, 2020).…”

“…Despite it is yet unclear how biases in attention or prior probability could be implemented in attractor neural network models, a recent study that fit such models to empirical RT data (Berlemont et al, 2020) offers exciting opportunities for direct comparison with evidence accumulation models.…”

Explaining distortions in metacognition with an attractor network model of decision uncertainty

“…Here, m and n are set to the model statistic (i.e., m = RT model , and n = accuracy model ) [52]. The cost function can be calculated per difficulty level (see [80]). Here, we opted for calculating the cost using the overall accuracy and overall response times (across all difficulties).…”