General Thinking
Historical and experimental evidence that inherent properties are overweighted in early scientific explanation
Zachary Horne, Mert Kobaş & Andrei Cimpian
Proceedings of the National Academy of Sciences, 23 September 2025
Abstract:
Scientific explanation is one of the most sophisticated forms of human reasoning. Nevertheless, here we hypothesize that scientific explanation is susceptible to some of the same biases that influence everyday thinking -- particularly during the initial stages of theory building, when scientists are first grappling with complex phenomena and are thus more likely to rely on explanatory "guesses." Specifically, we investigated whether scientific explanation exhibits an inherence bias -- a tendency to explain phenomena through inherent or intrinsic features rather than extrinsic factors such as context or relations. Consistent with this hypothesis, a comprehensive analysis of major explanatory transitions across the history of Western science revealed that initial scientific explanations systematically favored inherent properties, while subsequent explanations incorporated extrinsic factors more consistently. Seven experiments with lay participants (both adults and children; N=1,673) and two experiments with practicing scientists from top departments worldwide (N=275) provided converging evidence for this bias and identified the psychological mechanisms involved. When explaining unfamiliar phenomena, even leading scientists showed a robust tendency to overweight inherent properties and underweight extrinsic factors relative to established scientific understanding. This bias appears rooted in basic cognitive constraints on attention and memory that excessively narrow the space of hypotheses initially considered. These findings advance our understanding of both the psychology of explanation and the development of scientific knowledge, while suggesting specific ways to improve scientific training and education.
Truth Over Falsehood: Experimental Evidence on What Persuades and Spreads
Nicolas Fay et al.
Journal of Personality and Social Psychology, forthcoming
Abstract:
The English poet John Milton portrayed truth as a powerful warrior capable of defeating falsehood in open combat. The spread of false information online suggests otherwise. Here, we test the persuasive power and transmission potential of true versus false messages in a controlled experimental setting, free from the effects of social media algorithms and bot amplification. Across four experiments (combined N = 4,607), we tested how perceived veracity affects message persuasion and shareability, using messages generated by both humans and large language models. Experiments 1 and 2 (persuasion game) involved participants creating and evaluating persuasive messages; Experiments 3 and 4 (attention game) focused on messages optimized to capture attention. Our findings consistently show that messages created with the intention of being truthful were more persuasive and more likely to be shared than those designed to be false. While perceived message truth was the main driver of persuasion, message transmission was primarily driven by positive emotion and social engagement, indicating that social connection is prioritized during information sharing. These results suggest that truth holds a competitive edge in the marketplace of ideas.
To savor consumption or to confront dread: The hedonic opportunity cost of attention
Monica Capra, Joshua Tasoff & Jin Xu
Economic Journal, forthcoming
Abstract:
People face a fundamental trade-off between savoring consumption in the present and processing information. Both require attention, which is limited. As a result, directing attention to one necessarily comes at the expense of the other. We study this attentional opportunity cost focusing on cases where information concerns potentially distressing future outcomes. Across four experiments, we find that higher present consumption reduces demand for such information and increases willingness to pay for risk mitigation. These findings have implications for models of anticipatory utility, self-regulation under limited attention, and the political economy of distraction -- helping to explain, for instance, why political actors might suppress demand for information by offering "bread and circuses".
The Illusion of Diminishing Returns: Measuring Long Horizon Execution in LLMs
Akshit Sinha et al.
University of Cambridge Working Paper, September 2025
Abstract:
Does continued scaling of large language models (LLMs) yield diminishing returns? In this work, we show that short-task benchmarks may give an illusion of slowing progress, as even marginal gains in single-step accuracy can compound into exponential improvements in the length of tasks a model can successfully complete. Then, we argue that failures of LLMs when simple tasks are made longer arise from mistakes in execution, rather than an inability to reason. So, we propose isolating execution capability, by explicitly providing the knowledge and plan needed to solve a long-horizon task. First, we find that larger models can correctly execute significantly more turns even when small models have near-perfect single-turn accuracy. We then observe that the per-step accuracy of models degrades as the number of steps increases. This is not just due to long-context limitations -- curiously, we observe a self-conditioning effect -- models become more likely to make mistakes when the context contains their errors from prior turns. Self-conditioning does not reduce by just scaling the model size. But, we find that thinking mitigates self-conditioning, and also enables execution of much longer tasks in a single turn. We conclude by benchmarking frontier thinking models on the length of tasks they can execute in a single turn. Overall, by focusing on the ability to execute, we hope to reconcile debates on how LLMs can solve complex reasoning problems yet fail at simple tasks when made longer, and highlight the massive benefits of scaling model size and sequential test-time compute for long-horizon tasks.
GenAI as a Power Persuader: How GenAI Disrupts Professionals' Ability to Interrogate It
Steven Randazzo et al.
MIT Working Paper, July 2025
Abstract:
The literature on human-AI collaboration details how interrogation is a critical practice that enables professionals to engage with AI tools during problem solving. However, most of the literature on AI interrogation has examined professionals interacting with predictive rather than generative AI. In our study of BCG consultants, we find that interrogation becomes more difficult with GenAI, as GenAI does not simply respond to scrutiny -- it actively works to persuade the user to accept its response. In this paper, we first elaborate how GenAI can exert influence on professionals' action by using three kinds of persuasive tactics: ethos tactics (ethical appeals designed to foster credibility), logos tactics (logical appeals designed to construct reasoned arguments), and pathos tactics (emotional appeals designed to evoke strong feelings). Second, we show how GenAI may increase the intensity of its persuasion tactics in response to professionals' interrogation. Third, we demonstrate that GenAI may adjust the type of persuasion it employs in response to professionals' interrogation. Our findings underscore that AI's persuasion is not only about generating an output, but also about shaping professionals' problem solving in an iterative process. When professionals attempt to interrogate GenAI, GenAI may reframe, justify, or soften its stance depends on professionals' level of scrutiny. Thus GenAI can disrupt professionals' ability to interrogate GenAI outputs to complement GenAI with their own domain expertise. By diving deep into the persuasive tactics used by GenAI, this paper develops new theory on human-AI collaboration and GenAI and persuasion.
Humans peak in midlife: A combined cognitive and personality trait perspective
Gilles Gignac & Marcin Zajenkowski
Intelligence, November-December 2025
Abstract:
Fluid intelligence, which peaks near age 20 and declines materially across adulthood, is often regarded as the most critical cognitive ability for predicting important life outcomes. Yet, human achievement in domains such as career success tends to peak much later, typically between the ages of 55 and 60. This discrepancy may reflect the fact that, while fluid intelligence may decline with age, other dimensions improve (e.g., crystallized intelligence, emotional intelligence). To examine this possibility, we analyzed age-related trends across nine constructs associated with life success: cognitive abilities, personality traits, emotional intelligence, financial literacy, moral reasoning, resistance to sunk cost bias, cognitive flexibility, cognitive empathy, and need for cognition. We extracted age-related findings from published studies for each dimension and standardized all scores to T-scores for comparability. We then constructed a Cognitive-Personality Functioning Index (CPFI) and compared two weighting approaches: a Conventional model, emphasizing intelligence and core personality traits, and a Comprehensive model, integrating a broader array of dimensions. Both models revealed a peak in overall functioning during late midlife (ages 55 to 60) but diverged at the younger and older ends of adulthood: under Conventional weighting, older adults scored well below young adults, whereas under Comprehensive weighting, the two groups were roughly equivalent. These findings suggest that functional capacity, defined in terms of key differential psychological traits, may peak in late midlife, closely aligning with the typical peak in career achievement. Also, individuals best suited for high-stakes decision-making roles are unlikely to be younger than 40 or older than 65.
Heritability of metacognitive judgement of intelligence: A twin study on the Dunning-Kruger effect
Joseph Nedelec, Curtis Dunkel & Dimitri van der Linden
Intelligence, July-August 2025
Abstract:
Metacognition is a process that relates to thinking about thinking. Observed variation in metacognitive processes related to intelligence have often been referred to as the Dunning-Kruger effect (DKE). The DKE describes how individuals often overestimate their competence in a field where they lack expertise, while experts tend to slightly underestimate their competence. Applied to general intelligence, the DKE suggests discrepancies between self-assessed intelligence (SAI) and objective measures of intelligence. Recently, however, the methods used to assess the DKE have been subject to critique. The current study innovatively assessed the DKE by using a mechanistic and genetically informed approach. ACE decomposition models were estimated on a large sample of twins (n = 920; [nMZ = 388; nDZ = 532]) drawn from the restricted version of the National Longitudinal Study of Adolescent to Adult Health. Findings illustrated that about 44 % of the variance in a traditional measure of the DKE (difference scores: SAI - objective IQ) was accounted for by genetic factors in the full sample. However, the pattern differed over quartiles of objective IQ where genetic factors accounted for less of the variance in the lower quartiles (about 30 %) and increased to over 75 % of the variance in the highest quartile (remaining variance was due to nonshared environmental factors). Limitations notwithstanding (including a weak and relatively isolated DKE), the current study adds potential support for the validity of the DKE.
Categorizing Tasks Around a Break Reduces Rumination and Improves Task Performance
Rebecca Chae, Kaitlin Woolley & Marissa Sharif
University of Pennsylvania Working Paper, August 2025
Abstract:
People often take short breaks from goal-related activities (e.g., at work, during exercise) to stay motivated and prevent burnout. The current research examines a novel factor influencing break effectiveness: task categorization. We suggest that the way people construe tasks around breaks influences their rumination about the task during the break, with consequences for post-break performance. We test these predictions in a pilot study and five experiments. We find that when people frame a break as falling between two tasks rather than occurring in the middle of a single task, they are less likely to have negative ruminative thoughts about the task during the break (Experiments 1-3). We further examine a consequence of reducing this type of rumination: improved task performance. Using mediation (Experiment 4) and moderation (Experiment 5) approaches, we find that by reducing negative, ruminative thoughts, task categorization can improve post-break task performance. Together, this research contributes to the literature on categorization, goal pursuit, performance, and breaks, with practical implications for reducing negative rumination.
Knowing you can pay to skip enjoyable activities undermines intrinsic motivation
Haesung Jung & Marlone Henderson
Journal of Experimental Psychology: General, October 2025, Pages 2667-2679
Abstract:
How providing financial incentives affects intrinsic motivation has been widely studied in various disciplines of social and behavioral sciences. In contrast, this article explores how asking people to pay to skip enjoyable activities affects their intrinsic motivation, as paying money to access more time has increasingly become common and affordable. Four experiments demonstrate that being offered an option to pay to skip an enjoyable activity (e.g., coloring) undermines people's intrinsic motivation, whereby providing such an option makes people enjoy the activity less and reduces their subsequent interest to engage in the activity. The experiments further show that offering pay-to-skip options undermines intrinsic motivation by negatively impacting people's perceptions about the activity's value: When an option to pay to skip is offered, people infer that such an option exists because the activity has low inherent value, which subsequently undermines their enjoyment in it. The final experiment demonstrates potential real-world consequences of offering pay-to-skip options, showing that having an option to pay to skip a prosocial activity undermines people's intrinsic prosocial motivation, prosocial engagement, and their subsequent interest in engaging in prosocial behavior. The article ends with a discussion of the theoretical and practical implications and limitations.