Correction: Using Conversational AI to Facilitate Mental Health Assessments and Improve Clinical Efficiency Within Psychotherapy Services: Real-World Observational Study.

[This corrects the article DOI: 10.2196/44358.].

Closing the accessibility gap to mental health treatment with a personalized self-referral chatbot.

Inequality in treatment access is a pressing issue in most healthcare systems across many medical disciplines. In mental healthcare, reduced treatment access for minorities is ubiquitous but remedies are sparse. Here we demonstrate that digital tools can reduce the accessibility gap by addressing several key barriers. In a multisite observational study of 129,400 patients within England's NHS services, we evaluated the impact of a personalized artificial intelligence-enabled self-referral chatbot on patient referral volume and diversity in ethnicity, gender and sexual orientation. We found that services that used this digital solution identified substantially increased referrals (15% increase versus 6% increase in control services). Critically, this increase was particularly pronounced in minorities, such as nonbinary (179% increase) and ethnic minority individuals (29% increase). Using natural language processing to analyze qualitative feedback from 42,332 individuals, we found that the chatbot's human-free nature and the patients' self-realization of their need for treatment were potential drivers for the observed improvement in the diversity of access. This provides strong evidence that digital tools may help overcome the pervasive inequality in mental healthcare.

Why and When Beliefs Change.

Why people do or do not change their beliefs has been a long-standing puzzle. Sometimes people hold onto false beliefs despite ample contradictory evidence; sometimes they change their beliefs without sufficient reason. Here, we propose that the utility of a belief is derived from the potential outcomes associated with holding it. Outcomes can be internal (e.g., positive/negative feelings) or external (e.g., material gain/loss), and only some are dependent on belief accuracy. Belief change can then be understood as an economic transaction in which the multidimensional utility of the old belief is compared against that of the new belief. Change will occur when potential outcomes alter across attributes, for example because of changing environments or when certain outcomes are made more or less salient.

Using Conversational AI to Facilitate Mental Health Assessments and Improve Clinical Efficiency Within Psychotherapy Services: Real-World Observational Study.

BACKGROUND: Most mental health care providers face the challenge of increased demand for psychotherapy in the absence of increased funding or staffing. To overcome this supply-demand imbalance, care providers must increase the efficiency of service delivery. OBJECTIVE: In this study, we examined whether artificial intelligence (AI)-enabled digital solutions can help mental health care practitioners to use their time more efficiently, and thus reduce strain on services and improve patient outcomes. METHODS: In this study, we focused on the use of an AI solution (Limbic Access) to support initial patient referral and clinical assessment within the UK's National Health Service. Data were collected from 9 Talking Therapies services across England, comprising 64,862 patients. RESULTS: We showed that the use of this AI solution improves clinical efficiency by reducing the time clinicians spend on mental health assessments. Furthermore, we found improved outcomes for patients using the AI solution in several key metrics, such as reduced wait times, reduced dropout rates, improved allocation to appropriate treatment pathways, and, most importantly, improved recovery rates. When investigating the mechanism by which the AI solution achieved these improvements, we found that the provision of clinically relevant information ahead of clinical assessment was critical for these observed effects. CONCLUSIONS: Our results emphasize the utility of using AI solutions to support the mental health workforce, further highlighting the potential of AI solutions to increase the efficiency of care delivery and improve clinical outcomes for patients.

Confirmation bias is adaptive when coupled with efficient metacognition.

Biases in the consideration of evidence can reduce the chances of consensus between people with different viewpoints. While such altered information processing typically leads to detrimental performance in laboratory tasks, the ubiquitous nature of confirmation bias makes it unlikely that selective information processing is universally harmful. Here, we suggest that confirmation bias is adaptive to the extent that agents have good metacognition, allowing them to downweight contradictory information when correct but still able to seek new information when they realize they are wrong. Using simulation-based modelling, we explore how the adaptiveness of holding a confirmation bias depends on such metacognitive insight. We find that the behavioural consequences of selective information processing are systematically affected by agents' introspective abilities. Strikingly, we find that selective information processing can even improve decision-making when compared with unbiased evidence accumulation, as long as it is accompanied by good metacognition. These results further suggest that interventions which boost people's metacognition might be efficient in alleviating the negative effects of selective information processing on issues such as political polarization. This article is part of the theme issue 'The political brain: neurocognitive and computational mechanisms'.

Dogmatism manifests in lowered information search under uncertainty.

When knowledge is scarce, it is adaptive to seek further information to resolve uncertainty and obtain a more accurate worldview. Biases in such information-seeking behavior can contribute to the maintenance of inaccurate views. Here, we investigate whether predispositions for uncertainty-guided information seeking relate to individual differences in dogmatism, a phenomenon linked to entrenched beliefs in political, scientific, and religious discourse. We addressed this question in a perceptual decision-making task, allowing us to rule out motivational factors and isolate the role of uncertainty. In two independent general population samples (n = 370 and n = 364), we show that more dogmatic participants are less likely to seek out new information to refine an initial perceptual decision, leading to a reduction in overall belief accuracy despite similar initial decision performance. Trial-by-trial modeling revealed that dogmatic participants placed less reliance on internal signals of uncertainty (confidence) to guide information search, rendering them less likely to seek additional information to update beliefs derived from weak or uncertain initial evidence. Together, our results highlight a cognitive mechanism that may contribute to the formation of dogmatic worldviews.

Confidence drives a neural confirmation bias.

A prominent source of polarised and entrenched beliefs is confirmation bias, where evidence against one's position is selectively disregarded. This effect is most starkly evident when opposing parties are highly confident in their decisions. Here we combine human magnetoencephalography (MEG) with behavioural and neural modelling to identify alterations in post-decisional processing that contribute to the phenomenon of confirmation bias. We show that holding high confidence in a decision leads to a striking modulation of post-decision neural processing, such that integration of confirmatory evidence is amplified while disconfirmatory evidence processing is abolished. We conclude that confidence shapes a selective neural gating for choice-consistent information, reducing the likelihood of changes of mind on the basis of new information. A central role for confidence in shaping the fidelity of evidence accumulation indicates that metacognitive interventions may help ameliorate this pervasive cognitive bias.

Judgments of effort exerted by others are influenced by received rewards.

Estimating invested effort is a core dimension for evaluating own and others' actions, and views on the relationship between effort and rewards are deeply ingrained in various societal attitudes. Internal representations of effort, however, are inherently noisy, e.g. due to the variability of sensorimotor and visceral responses to physical exertion. The uncertainty in effort judgments is further aggravated when there is no direct access to the internal representations of exertion - such as when estimating the effort of another person. Bayesian cue integration suggests that this uncertainty can be resolved by incorporating additional cues that are predictive of effort, e.g. received rewards. We hypothesized that judgments about the effort spent on a task will be influenced by the magnitude of received rewards. Additionally, we surmised that such influence might further depend on individual beliefs regarding the relationship between hard work and prosperity, as exemplified by a conservative work ethic. To test these predictions, participants performed an effortful task interleaved with a partner and were informed about the obtained reward before rating either their own or the partner's effort. We show that higher rewards led to higher estimations of exerted effort in self-judgments, and this effect was even more pronounced for other-judgments. In both types of judgment, computational modelling revealed that reward information and sensorimotor markers of exertion were combined in a Bayes-optimal manner in order to reduce uncertainty. Remarkably, the extent to which rewards influenced effort judgments was associated with conservative world-views, indicating links between this phenomenon and general beliefs about the relationship between effort and earnings in society.

Postdecision Evidence Integration and Depressive Symptoms.

Background: Metacognition, or the ability to reflect on one's own thoughts, may be important in the development of depressive symptoms. Recent work has reported that depressive symptoms were associated with lower metacognitive bias (overall confidence) during perceptual decision making and a trend toward a positive association with metacognitive sensitivity (the ability to discriminate correct and incorrect decisions). Here, we extended this work, investigating whether confidence judgments are more malleable in individuals experiencing depressive symptoms. We hypothesized that depressive symptoms would be associated with greater adjustment of confidence in light of new evidence presented after a perceptual decision had been made. Methods: Participants (N = 416) were recruited via Amazon Mechanical Turk. Metacognitive confidence was assessed through two perceptual decision-making tasks. In both tasks, participants made a decision about which of two squares contained more dots. In the first task, participants rated their confidence immediately following the decision, whereas in the second task, participants observed new evidence (always in the same direction as initial evidence) before rating their confidence. Participants also completed questionnaires measuring depressive symptoms and self-esteem. Analysis: Metacognitive bias was calculated as overall mean confidence, whereas metacognitive sensitivity was calculated using meta-d' (a response-bias free measure of how closely confidence tracks task performance) in the first task. Postdecision evidence integration (PDEI) was defined as the change in confidence following postdecision evidence on the second task. Results: Participants with more depressive symptoms made greater confidence adjustments (i.e., greater PDEI) in light of new evidence (ß = 0.119, p = 0.045), confirming our main hypothesis. We also observed that lower overall confidence was associated with greater depressive symptoms, although this narrowly missed statistical significance (ß = -0.099, p = 0.056), and we did not find an association between metacognitive sensitivity (meta-d') and depressive symptoms. Notably, self-esteem was robustly associated with overall confidence (ß = 0.203, p < 0.001), which remained significant when controlling for depressive symptoms. Conclusions: We found that individuals with depressive symptoms were more influenced by postdecisional evidence, adjusting their confidence more in light of new evidence. Individuals with low self-esteem were less confident about their initial decisions. This study should be replicated in a clinically depressed sample.

What Underlies Political Polarization? A Manifesto for Computational Political Psychology.

Polarization is one of the biggest societal challenges of our time, yet its drivers are poorly understood. Here we propose a novel approach - computational political psychology - which uses behavioral tasks in combination with formal computational models to identify candidate cognitive processes underpinning susceptibility to polarized beliefs about political and societal issues.

Metacognitive Failure as a Feature of Those Holding Radical Beliefs.

Widening polarization about political, religious, and scientific issues threatens open societies, leading to entrenchment of beliefs, reduced mutual understanding, and a pervasive negativity surrounding the very idea of consensus [1, 2]. Such radicalization has been linked to systematic differences in the certainty with which people adhere to particular beliefs [3-6]. However, the drivers of unjustified certainty in radicals are rarely considered from the perspective of models of metacognition, and it remains unknown whether radicals show alterations in confidence bias (a tendency to publicly espouse higher confidence), metacognitive sensitivity (insight into the correctness of one's beliefs), or both [7]. Within two independent general population samples (n = 381 and n = 417), here we show that individuals holding radical beliefs (as measured by questionnaires about political attitudes) display a specific impairment in metacognitive sensitivity about low-level perceptual discrimination judgments. Specifically, more radical participants displayed less insight into the correctness of their choices and reduced updating of their confidence when presented with post-decision evidence. Our use of a simple perceptual decision task enables us to rule out effects of previous knowledge, task performance, and motivational factors underpinning differences in metacognition. Instead, our findings highlight a generic resistance to recognizing and revising incorrect beliefs as a potential driver of radicalization.

Post-decision wagering after perceptual judgments reveals bi-directional certainty readouts.

Humans and other animals constantly evaluate their decisions in order to learn and behave adaptively. Experimentally, such evaluation processes are accessed using metacognitive reports made after decisions, typically using verbally formulated confidence scales. When subjects report high confidence, it reflects a high certainty of being correct, but a low confidence might signify either low certainty about the outcome, or a high certainty of being incorrect. Hence, metacognitive reports might reflect not only different levels of decision certainty, but also two certainty directions (certainty of being correct and certainty of being incorrect). It is important to test if such bi-directional processing can be measured because, for decision-making under uncertainty, information about being incorrect is as important as information about being correct for guidance of subsequent behavior. We were able to capture implicit bi-directional certainty readouts by asking subjects to bet money on their perceptual decision accuracy using a six-grade wager scale (post-decision wagering, PDW). To isolate trial-specific aspects of metacognitive judgments, we used pre-decision wagering (wagering before the perceptual decision) to subtract, from PDW trials, influences resulting from non-trial-specific assessment of expected difficulty and psychological biases. This novel design allowed independent quantification of certainty of being correct and certainty of being incorrect, showing that subjects were able to read out certainty in a bi-directional manner. Certainty readouts about being incorrect were particularly associated with metacognitive sensitivity exceeding perceptual sensitivity (i.e. meta-d' > d'), suggesting that such enhanced metacognitive efficiency is driven by information about incorrect decisions. Readouts of certainty in both directions increased on easier trials, and both certainty directions were also associated with faster metacognitive reaction times, indicating that certainty of being incorrect was not confounded with low certainty. Finally, both readouts influenced the amount of money subjects earned through PDW, suggesting that bi-directional readouts are important for planning future actions when feedback about previous decisions is unavailable.