Human-like dissociations between confidence and accuracy in convolutional neural networks

Prior research has shown that manipulating stimulus energy by changing both stimulus contrast and variability results in confidence-accuracy dissociations in humans. Specifically, even when performance is matched, higher stimulus energy leads to higher confidence. The most common explanation for this effect is the positive evidence heuristic where confidence neglects evidence that disconfirms the choice. However, an alternative explanation is the signal-and-variance-increase hypothesis, according to which these dissociations arise from low-level changes in the separation and variance of perceptual representations. Because artificial neural networks lack built-in confidence heuristics, they can serve as a test for the necessity of confidence heuristics in explaining confidence-accuracy dissociations. Therefore, we tested whether confidence-accuracy dissociations induced by stimulus energy manipulations emerge naturally in convolutional neural networks (CNNs). We found that, across three different energy manipulations, CNNs produced confidence-accuracy dissociations similar to those found in humans. This effect was present for a range of CNN architectures from shallow 4-layer networks to very deep ones, such as VGG-19 and ResNet -50 pretrained on ImageNet. Further, we traced back the reason for the confidence-accuracy dissociations in all CNNs to the same signal-and-variance increase that has been proposed for humans: higher stimulus energy increased the separation and variance of the CNNs' internal representations leading to higher confidence even for matched accuracy. These findings cast doubt on the necessity of the positive evidence heuristic to explain human confidence and establish CNNs as promising models for adjudicating between low-level, stimulus-driven and high-level, cognitive explanations of human behavior.

The causal role of the prefrontal and superior medial frontal cortices in the incidental manipulation of decision strategies

A key strategic decision one must make in virtually every task context concerns the speed accuracy trade-off (SAT). Experimentally, this ubiquitous phenomenon, whereby response speed and task accuracy are inversely related, is typically studied by explicitly instructing participants to adjust their strategy: by either focusing on speed, or on accuracy. Computational modelling has been applied to deconvolve the latent decision processes involved in the SAT, with considerable evidence suggesting that response caution (the amount of evidence needed for a decision to be reached) is a key variable in the setting of SAT strategy. Neuroimaging has implicated the prefrontal cortex, the pre-supplementary motor area (preSMA), and the striatum in the setting of response caution. In addition, brain stimulation has provided causal evidence for the involvement of the left prefrontal cortex and superior medial frontal cortex (SMFC, which includes the preSMA) in adjustments of response caution following explicit instructions, although stimulation of the two regions has dissociable effects. Here, in a double-blind and preregistered study we investigated the role of these two regions using an incidental manipulation of SAT strategy - via stimulus signal variability - which has previously been shown to influence decision confidence. We again found tDCS applied to both regions modulated response caution, and there was a dissociation: stimulating prefrontal cortex increased, and stimulating SMFC decreased, response caution. These findings provide further support for key, but dissociable, roles of these brain regions in decision strategies whether they are implemented explicitly or incidentally.

Consensus Goals in the Field of Visual Metacognition

Despite the tangible progress in psychological and cognitive sciences over the last several years, these disciplines still trail other more mature sciences in identifying the most important questions that need to be solved. Reaching such consensus could lead to greater synergy across different laboratories, faster progress, and increased focus on solving important problems rather than pursuing isolated, niche efforts. Here, 26 researchers from the field of visual metacognition reached consensus on four long-term and two medium-term common goals. We describe the process that we followed, the goals themselves, and our plans for accomplishing these goals. If this effort proves successful within the next few years, such consensus building around common goals could be adopted more widely in psychological science.

Perceptual confidence of visual stimulus features is associated with duration perception

It has been shown that the perceived duration of an object in the subsecond range is closely associated with its nontemporal perceptual properties, the mechanism under which remains unclear. Previous studies have revealed a modulatory effect of early visual feature processing on the apparent duration. Here, we further examined the relationship between perceptual confidence and subjective time by asking participants to simultaneously perform temporal and nontemporal perceptual judgments. The results revealed a significant effect on confidence levels. When participants' confidence in judging the coherent motion direction or relative dot numerosity increases, their perceived duration of the stimulus also appears longer. These results are discussed in the context of perceptual evidence accumulation and evaluation for the decision-making of perceptual properties. They suggest a profound contribution of object processing to the computation of subjective time and provide further insights into the mechanism of event timing.

Human perceptual and metacognitive decision-making rely on distinct brain networks

Perceptual decisions depend on the ability to exploit available sensory information in order to select the most adaptive option from a set of alternatives. Such decisions depend on the perceptual sensitivity of the organism, which is generally accompanied by a corresponding level of certainty about the choice made. Here, by use of corticocortical paired associative transcranial magnetic stimulation protocol (ccPAS) aimed at inducing plastic changes, we shaped perceptual sensitivity and metacognitive ability in a motion discrimination task depending on the targeted network, demonstrating their functional dissociation. Neurostimulation aimed at boosting V5/MT+-to-V1/V2 back-projections enhanced motion sensitivity without impacting metacognition, whereas boosting IPS/LIP-to-V1/V2 back-projections increased metacognitive efficiency without impacting motion sensitivity. This double-dissociation provides causal evidence of distinct networks for perceptual sensitivity and metacognitive ability in humans.

A robust confidence-accuracy dissociation via criterion attraction

Many studies have shown that confidence and accuracy can be dissociated in a variety of tasks. However, most of these dissociations involve small effect sizes, occur only in a subset of participants, and include a reaction time (RT) confound. Here, I develop a new method for inducing confidence-accuracy dissociations that overcomes these limitations. The method uses an external noise manipulation and relies on the phenomenon of criterion attraction where criteria for different tasks become attracted to each other. Subjects judged the identity of stimuli generated with either low or high external noise. The results showed that the two conditions were matched on accuracy and RT but produced a large difference in confidence (effect appeared for 25 of 26 participants, effect size: Cohen's d = 1.9). Computational modeling confirmed that these results are consistent with a mechanism of criterion attraction. These findings establish a new method for creating conditions with large differences in confidence without differences in accuracy or RT. Unlike many previous studies, however, the current method does not lead to differences in subjective experience and instead produces robust confidence-accuracy dissociations by exploiting limitations in post-perceptual, cognitive processes.

Sources of Metacognitive Inefficiency

Confidence judgments are typically less informative about one's accuracy than they could be; a phenomenon we call metacognitive inefficiency. We review the existence of different sources of metacognitive inefficiency and classify them into four categories based on whether the corruption is due to: (i) systematic or nonsystematic influences, and (ii) the input to or the computation of the metacognitive system. Critically, the existence of different sources of metacognitive inefficiency provides an alternative explanation for behavioral findings typically interpreted as evidence for domain-specific (and against domain-general) metacognitive systems. We argue that, contrary to the dominant assumption in the field, metacognitive failures are not monolithic and suggest that understanding the sources of metacognitive inefficiency should be a primary goal of the science of metacognition.

Among the Two Kinds of Metacognitive Evaluation, Only One Is Predictive of Illusory Object Perception

The relationship between expectation-induced hallucination proneness and self-confidence in performance was studied in a visual perception task. Participants were prompted either to recognize briefly shown faces as male or female or to rate the subjective vividness of a square surrounding the face. Importantly, in a few critical trials, the square was absent. Upon completion, participants rated their performance in the face recognition task; they were also asked whether they were sure that their estimation was correct. Out of 35 participants, 33 "hallucinated" on at least one trial, rating the square as visible when it was actually absent. Negative correlation between hallucination proneness and self-confidence in performance (metacognitive rating) was found: The more hallucinations a participant experienced, the less confident he/she was in his/her performance in the face recognition task. Most subjects underestimated their performance; higher ratings were also more accurate. Thus, higher hallucination proneness was associated with more inaccurate ratings of one's own perception. However, confidence in self-ratings as measured by the second follow-up question was unrelated to both, hallucination proneness and self-confidence in performance, supporting the view that there is no unitary mechanism of metacognitive evaluations and extending this view to the domain of visual hallucinatory perception.