Many policies and ethical guidelines recommend developing “trustworthy AI”. We argue that developing morally trustworthy AI is not only unethical, as it promotes trust in an entity that cannot be trustworthy, but it is also unnecessary for optimal calibration. Instead, we show that reliability, exclusive of moral trust, entails the appropriate normative constraints that enable optimal calibration and mitigate the vulnerability that arises in high-stakes hybrid decision-making environments, without also demanding, as moral trust would, the anthropomorphization of AI and thus (...) epistemically dubious behavior. The normative demands of reliability for inter-agential action are argued to be met by an analogue to procedural metacognitive competence (i.e., the ability to evaluate the quality of one’s own informational states to regulate subsequent action). Drawing on recent empirical findings that suggest providing reliability scores (e.g., F1-scores) to human decision-makers improves calibration in the AI system, we argue that reliability scores provide a good index of competence and enable humans to determine how much they wish to rely on the system. (shrink)

Metacognition is the capacity to evaluate the success of one's own cognitive processes in various domains; for example, memory and perception. It remains controversial whether metacognition relies on a domain-general resource that is applied to different tasks or if self-evaluative processes are domain specific. Here, we investigated this issue directly by examining the neural substrates engaged when metacognitive judgments were made by human participants of both sexes during perceptual and memory tasks matched for stimulus and performance characteristics. By comparing patterns (...) of fMRI activity while subjects evaluated their performance, we revealed both domain-specific and domain-general metacognitive representations. Multivoxel activity patterns in anterior prefrontal cortex predicted levels of confidence in a domain-specific fashion, whereas domain-general signals predicting confidence and accuracy were found in a widespread network in the frontal and posterior midline. The demonstration of domain-specific metacognitive representations suggests the presence of a content-rich mechanism available to introspection and cognitive control. (shrink)

This paper reports the first empirical investigation of the hypothesis that epistemic appraisals form part of the structure of concepts. To date, studies of concepts have focused on the way concepts encode properties of objects and the way those features are used in categorization and in other cognitive tasks. Philosophical considerations show the importance of also considering how a thinker assesses the epistemic value of beliefs and other cognitive resources and, in particular, concepts. We demonstrate that there are multiple, reliably (...) judged, dimensions of epistemic appraisal of concepts. Four of these dimensions are accounted for by a common underlying factor capturing how well people believe they understand a concept. Further studies show how dimensions of concept appraisal relate to other aspects of concepts. First, they relate directly to the hierarchical organization of concepts, reflecting the increase in specificity from superordinate to basic and subordinate levels. Second, they predict inductive choices in category-based induction. Our results suggest that epistemic appraisals of concepts form a psychologically important yet previously overlooked aspect of the structure of concepts. These findings will be important in understanding why individuals sometimes abandon and replace certain concepts; why social groups do so, for example, during a “scientific revolution”; and how we can facilitate such changes when we engage in deliberate “conceptual engineering” for epistemic, social, and political purposes. (shrink)

To study consciousness, scientists need to determine when participants are conscious and when they are not. They do so with consciousness detection procedures. A recurring skeptical argument against those procedures is that they cannot be calibrated: there is no way to make sure that detection outcomes are accurate. In this article, I address two main skeptical arguments purporting to show that consciousness scientists cannot calibrate detection procedures. I conclude that there is nothing wrong with calibration in consciousness science.

Metacognitive mental states are mental states about mental states. For example, I may be uncertain whether my belief is correct. In social discourse, an interlocutor’s metacognitive certainty may constitute evidence about the reliability of their testimony. For example, if a speaker is certain that their belief is correct, then we may take this as evidence in favour of their belief, or its content. This paper argues that, if metacognitive certainty is genuine evidence, then it is disproportionate evidence for extreme beliefs. (...) In support of the argument, we report findings from five studies with different participant samples, designs, and measures. These studies show that, the more extreme an agent’s belief (positive or negative), the more certain they are about it, and vice versa. This relationship might contribute to moralism, virtue signalling, and polarisation, which in turn may be epistemically and morally problematic. Therefore, we caution against taking metacognitive certainty as genuine evidence. (shrink)

The mainstream science of consciousness offers a few predominate views of how the brain gives rise to awareness. Chief among these are the Higher-Order Thought Theory, Global Neuronal Workspace Theory, Integrated Information Theory, and hybrids thereof. In parallel, rapid development in predictive processing approaches have begun to outline concrete mechanisms by which interoceptive inference shapes selfhood, affect, and exteroceptive perception. Here, we consider these new approaches in terms of what they might offer our empirical, phenomenological, and philosophical understanding of consciousness (...) and its neurobiological roots. (shrink)

We often effortlessly take the perceptual perspective of others: we represent some aspect of the environment that others currently perceive. However, taking someone's perspective can interfere with one's perceptual processing: another person's gaze can spontaneously affect our ability to detect stimuli in a scene. But it is still unclear whether our cognitive evaluation of those judgements is also affected. In this study, we investigated whether social perspective-taking can influence participants' metacognitive judgements about their perceptual responses. Participants performed a contrast detection (...) task with a task-irrelevant avatar oriented either congruently or incongruently to the stimulus location. By “blindfolding” the avatar, we tested the influence of social perspective-taking versus domain-general directional orienting. Participants had higher accuracy and perceptual sensitivity with a congruent avatar regardless of the blindfold, suggesting a directional cueing effect. However, their metacognitive efficiency was modulated only by the congruency of a seeing avatar. These results suggest that perceptual metacognitive ability can be socially enhanced by sharing perception of the same objects with others. (shrink)

People are often inaccurate in their predictions of performance on a variety of cognitive tasks. We tested whether receptivity to bullshit – the tendency to perceive meaningless statements as profound – would relate to the accuracy of metacognitive judgments on several problem-solving tasks. Individuals who were highly receptive to bullshit were less accurate in their predictions of performance on creative problem-solving tasks, but not on verbal analogy or recall tasks. Further, individuals with high BS receptivity were less able to discriminate (...) between solvable and unsolvable problems when making metacognitive judgments. These findings support the possibility that the tendency to perceive semantic connections where none exist, as indicated by high bullshit receptivity, may lead to inaccurate predictions of performance on tasks that require noticing and utilizing distant semantic connections. (shrink)

In this article, we suggest that the study of social interactions and the development of a “sense of agency” in joint action can help determine the content of relevant explanations to be implemented in artificial systems to make them “explainable.” The introduction of automated systems, and more broadly of Artificial Intelligence, into many domains has profoundly changed the nature of human activity, as well as the subjective experience that agents have of their own actions and their consequences – an experience (...) that is commonly referred to as sense of agency. We propose to examine the empirical evidence supporting this impact of automation on individuals’ sense of agency, and hence on measures as diverse as operator performance, system explicability and acceptability. Because of some of its key characteristics, AI occupies a special status in the artificial systems landscape. We suggest that this status prompts us to reconsider human–AI interactions in the light of human–human relations. We approach the study of joint actions in human social interactions to deduce what key features are necessary for the development of a reliable sense of agency in a social context and suggest that such framework can help define what constitutes a good explanation. Finally, we propose possible directions to improve human–AI interactions and, in particular, to restore the sense of agency of human operators, improve their confidence in the decisions made by artificial agents, and increase the acceptability of such agents. (shrink)

Consciousness and confidence seem intimately related. Accordingly, some researchers use confidence ratings as a measure of, or proxy for, consciousness. Rosenthal discusses the potential connections between the two, and rejects confidence as a valid measure of consciousness. He argues that there are better alternatives to get at conscious experiences such as direct subjective reports of awareness (i.e. subjects’ reports of perceiving something or of the degree of visibility of a stimulus). In this chapter, we offer a different perspective. Confidence ratings (...) in general, and metacognitive measures in particular, may offer important advantages over subjective ratings. The arguments we offer here are supported by empirical, practical and socio-strategic considerations. However, we do not suggest consciousness and confidence are interchangeable. We recognize the limitations of confidence ratings in some experimental designs and for some research questions. Nevertheless, we also address a potential conceptual link between consciousness and confidence that stems from Rosenthal’s very own work on mental quality space theory. (shrink)

Confidence ratings have often been integrated into reasoning and intelligence tasks as a means for assessing meta-reasoning processes. Although it is often assumed that eliciting these judgements throughout reasoning tasks has no effect on the underlying performance outcomes, this is yet to be established empirically. The current study examines whether eliciting CR from participants during a fluid-reasoning task influences their performance and how this effect is moderated by their initial self-confidence in their own reasoning abilities. In a first experiment, we (...) found that participants performing CR during Raven's Progressive Matrices significantly outperformed a control group who did not provide ratings. Additionally, a second experiment demonstrated that CR only facilitated performance in participants who have a high level of initial self-confidence in their reasoning ability, whereas they were detrimental to participants low in self-confidence. (shrink)

By definition, metacognitive processes may monitor or regulate various stages of first-order processing. By combining causal analysis with hypotheses expressed by other authors we derive the theoretical and methodological consequences of this special relation between metacognition and the underlying processes. In particular, we prove that because multiple processing stages may be monitored or regulated and because metacognition may form latent feedback loops, (1) without strong additional causal assumptions, typical measures of metacognitive monitoring or regulation are confounded; (2) without strong additional (...) causal assumptions, typical methods of controlling for first-order task performance (i.e., calibration, staircase, including first-order task performance in a regression analysis, or analyzing correct and incorrect trials separately) not only do not deconfound measures of metacognition but may even introduce bias; (3) that the first two problems cannot be solved by using simple models of decision-making derived from Signal Detection Theory. We conclude the paper by advocating robust methods of discovering properties of latent mechanisms. (shrink)

This chapter discusses the main types of so-called ’subjective measures of consciousness’ used in current-day science of consciousness. After explaining the key worry about such measures, namely the problem of an ever-present response bias, I discuss the question of whether subjective measures of consciousness are introspective. I show that there is no clear answer to this question, as proponents of subjective measures do not employ a worked-out notion of subjective access. In turn, this makes the problem of response bias less (...) tractable than it might otherwise be. (shrink)

Negative subjective reports such as “I didn’t see the stimulus” can be interpreted as indicating either that the subject didn’t see the stimulus, or as indicating that, while the subject did see the stimulus, the strength of sensory signals associated with the stimulus fell below a conservative criterion for answering “seen”. Determining which of these two interpretations is correct is the criterion problem. I present two ways in which researchers can solve this problem. But there’s more. What I call the (...) “new criterion problem” consists in determining whether criteria effects reflect perceptual changes, or changes in response strategy. Through case studies, I argue that this new version of the problem can be solved as well, although there is currently no systematic way to solve it. I conclude that criterion problems do not constitute good reasons for skepticism about the scientific study of consciousness. (shrink)

Research suggests that metacognitive monitoring ability does not decline with age. For example, judgments-of-learning accuracy is roughly equivalent between younger and older adults. But few studies have asked whether younger and older adults’ metacognitive ability varies across different types of memory processes. The current study tested the relationship between memory and post-decision confidence ratings at the trial level on item and associative memory recognition tests. As predicted, younger and older adults had similarmetacognitive efficiency, when using meta-d’/d’, a measure derived from (...) Signal Detection Theory, despite a significant age effect favoring younger adults on memory performance. This result is consistent with previous work showing age-equivalent metacognitive efficiency in the memory domain. We also found that metacognitive efficiency was higher for associative memory than for item memory across age groups, even though associative and item recognition memory were statistically equivalent. Higher accuracy on post-test decision confidence ratings for associative recognition relative to item recognition on resolution accuracy itself and when corrected for performance differences are novel findings. Implications for associative metacognition are discussed. (shrink)

To study consciousness, scientists need to determine when participants are conscious and when they are not. They do so with consciousness detection procedures. A recurring skeptical argument against those procedures is that they cannot be calibrated: there is no way to make sure that detection outcomes are accurate. In this article, I address two main skeptical arguments purporting to show that consciousness scientists cannot calibrate detection procedures. I conclude that there is nothing wrong with calibration in consciousness science.