Trends in Cognitive Sciences

Imagining and building wise machines: the centrality of AI metacognition Although artificial intelligence (AI) has become increasingly smart, its wisdom has not kept pace. In this opinion article, we examine what is known about human wisdom and sketch a vision of its AI counterpart. We introduce human wisdom as strategies for solving intractable problems—those outside the scope of analytic techniques—including both ‘object-level’ strategies, such as heuristics (for managing problems), and ‘metacognitive’ strategies, such as intellectual humility, perspective-taking, or context adaptability (for managing object-level task fit). We argue that AI systems particularly struggle with this type of metacognition. Wise metacognition would lead to AI that is more robust to novel environments, explainable to users, cooperative with others, and safer by risking fewer misaligned goals with human users. We discuss how wise AI might be benchmarked, trained, and implemented.

Online Now: Imagining and building wise machines: the centrality of AI metacognition

Intuitive theories of truth Cognitive science has recently begun exploring how people conceptualize and reason about truth. We offer the field a framework that can guide inquiry into intuitive theories of truth, centered on three core questions: how do people judge whether statements could be true, whether statements are true, and whether to assert them as true.

Online Now: Intuitive theories of truth

The making of number: from content to representation

Review by Andreas Nieder, tinyurl.com/kjtrkvmx

Language learning as ontogenetic adaptation

Opinion by Manuel Bohn (@elmanubohn.bsky.social) & Marisa Casillas
tinyurl.com/48pdbv5b

Producing more while understanding less with large language models Many scientists are enthusiastic about the potentials of 'Artificial Intelligence' (AI) for research. We recently examined the vision of ‘AI Surrogates’ [1]: computer models [including but not limited to large language models (LLMs)] designed to simulate human participants for the purpose of generating knowledge about human cognition and behavior. Some scientists believe that AI Surrogates can improve the generalizability of cognitive science: first, by simulating diverse populations that are not readily accessible, overcoming the field’s overreliance on Western, Educated, Industrialized, Rich, Democratic (WEIRD) samples; and second, by enabling researchers to quickly and cheaply explore vast experimental design spaces, expanding the diversity of situations that can be probed experimentally.

Online Now: Producing more while understanding less with large language models

A unifying taxonomy of dyadic emotional processes

Review by Martine W.F.T. Verhees, Batja Mesquita (@batjamesquita.bsky.social), Eva Ceulemans, Joeri Hofmans, Lesley Verhofstadt, & Peter Kuppens

tinyurl.com/2xzmmt5z

Moving intentions from brains to machines

Opinion by Christian Beste, Heleen A. Slagter (@haslagter.bsky.social), Christian Herff (@cherff.bsky.social), Yukiyasu Kamitani (@ykamit.bsky.social), Sabrina Coninx (@sconinxphil.bsky.social), Richard van Wezel, & Christian Frings

tinyurl.com/mr2ch69z

Reminders that chatbots are not human can be risky Concerns about mental and physical health harms from chatbots are prompting policies mandating ongoing reminders that chatbots are not human. While well-intended, evidence suggests that reminders may be either ineffective or harmful to users. Discovering how to best remind people that chatbots are not human is a critical research priority.

Online Now: Reminders that chatbots are not human can be risky

Moral decision-making with bounded cognitive resources and limited information Real-world moral decisions are constrained by limited information and bounded cognitive resources, necessitating heuristic strategies. We argue that choices in moral dilemmas should be analysed in terms of decision strategies rather than ethical theories and show how resource rationality and the bias–variance trade-off explain when people rely on particular strategies.

Online Now: Moral decision-making with bounded cognitive resources and limited information

How AI tools can enhance generalizability Using large language models (LLMs) to replace human participants suffers from fundamental fallacies: overgeneralization from Western, Educated, Industrialized, Rich, Democratic (WEIRD) samples; conflation of linguistic form with psychological content; and neglect of embodied and social dimensions of cognition [1]. In their recent article in TiCS [2], Crockett and Messeri extend this critique, arguing that ‘AI Surrogates’ perpetuate generalizability problems by entrenching WEIRD samples and decontextualized tasks [3].

Online Now: How AI tools can enhance generalizability

The making of number: from content to representation Despite their importance to human thought, the origins of numerical abilities remain debated. Numerical quantity is a property of physical objects and events, and both humans and many animals show an innate sensitivity to this numerical content. Yet how this content is represented is a separate question: it may be encoded nonsymbolically by an innate estimation system or symbolically through culturally developed formats, such as numeral notations and number words. Distinguishing content from representational format reconciles the views that numbers are innate (nativism), learned (empiricism), or constructed (emergentism). Converging evidence from developmental psychology, comparative cognition, neuroscience, and computation suggests that number is dynamically coconstructed by biological predispositions and cultural practices, a framework that generalizes to other domains of human cognition, such as geometry and language.

Online Now: The making of number: from content to representation

Communicating risks more comprehensively using simulated experience Traditional risk communication emphasizes the probability of possible outcomes but neglects other crucial dimensions of risks. We propose a taxonomy of risk information and illustrate how simulated experience can convey overlooked aspects. We conclude with a research agenda to advance the theory and practice of risk communication across domains.

Online Now: Communicating risks more comprehensively using simulated experience

Language learning as ontogenetic adaptation Language learning is a multi-threaded, multi-mechanism process. It is multi-threaded in that it emerges as a byproduct of addressing multiple goals while engaging in social interactions. It is multi-mechanism in that children integrate multiple information sources to infer what is meant and what to say next. These information sources include contextual and social cues, as well as cognitive mechanisms. Focusing on early word learning, this article reviews information sources, how children might sensitively adapt to them, and how we can model their integration using Bayesian inference over multiple probability distributions. We argue that, to advance our understanding of language learning, we must jointly study how children learn from multiple information sources across diverse developmental settings.

Online Now: Language learning as ontogenetic adaptation

A dysfunctional hub model of voice–reward integration in autism Children with autism often struggle to tune in to voices, missing important cues for social connection and language learning. What underlies this diminished engagement? Neuroimaging evidence implicates disrupted connectivity between voice-selective temporal regions and brain networks supporting reward, salience, and social cognition, leading to a new neural model of vocal insensitivity in autism.

Online Now: A dysfunctional hub model of voice–reward integration in autism

A unifying taxonomy of dyadic emotional processes Traditionally, emotions have been viewed as intrapersonal experiences, but recent perspectives emphasize their interpersonal nature. Yet research on interpersonal emotional processes is scattered across subfields and lacks a unifying framework. We propose a taxonomy that structures the different methodological ways through which dyadic emotional processes can be and have been empirically studied. Reviewing recent work, we apply this taxonomy to identify well-studied versus neglected areas and clarify gaps between empirical work and theory. Our review highlights two key insights: to advance theory, empirical research should (i) consider context and (ii) move beyond simple bivariate approaches. Laying the methodological groundwork, the taxonomy provides a first step toward unifying the literature and fostering cumulative progress in understanding interpersonal emotional processes.

Online Now: A unifying taxonomy of dyadic emotional processes

Emotion may indirectly link rendering and social reasoning In their letter [1], Zeman et al. raise the intriguing suggestion that visual imagery (which we argued should be understood as graphical rendering [2]) has a central role in social cognition. Specifically, they point to an association between aphantasia and deficits in autobiographical memory [3,4] and lower empathy to verbal descriptions of distressing events [5]. To be clear upfront, we agree with Zeman et al.’s bottom line that visual imagery might have social implications, and that this idea deserves thought and exploration.

Online Now: Emotion may indirectly link rendering and social reasoning

Neurocomputational mechanisms of adaptive mentalization in humans ‘Theory of mind’ (ToM) is classically investigated with ‘static’ inference tasks, which miss the dynamic nature of social interactions. In a recent article, Buergi, Aydogan, and colleagues combined computational modeling and neuroimaging to study the adaptive nature of mentalization (i.e., the ability to infer the continuous change of others’ thoughts and intentions).

Online Now: Neurocomputational mechanisms of adaptive mentalization in humans

Rendering aphantasia into the social realm Inspired by an analogy with ‘game engines’, the software modules that support animations and games, Balaban and Ullman propose that there is a deep-seated distinction between ‘physical simulation’ and ‘graphical rendering’ in the brain, broadly mapping onto the dorsal and ventral streams of visual processing [1]. The developmental pathway giving rise to in silico game engines is very different from the evolutionary history of the human brain, constrained by evolutionary forces. Nevertheless, the analogy is illuminating.

Online Now: Rendering aphantasia into the social realm

No free lunch with the binding problem In their recent article [1], Scholte and de Haan argue, contrary to the classical view, that the visual cortex is not organized into separate modules to process individual features (e.g., color in V4 and motion in V5/MT). In the absence of such a modular organization, they argue, the problem of binding separate features together in coherent object representations (the binding problem) does not arise. A recent commentary by Roelfsema and Serre [2] has already argued that the case against modularity is not as strong as claimed by Scholte and de Haan and highlighted empirical evidence for binding mechanisms both in the visual cortex and artificial neural networks.

Online Now: No free lunch with the binding problem

Moving intentions from brains to machines Brain–computer interface (BCI) research has achieved remarkable technical progress but remains limited in scope, typically relying on motor and visual cortex signals in limited patient populations. We propose a paradigm shift in BCI design rooted in ideomotor theory, which conceptualizes voluntary action as driven by internally represented sensory outcomes. This underused framework offers a principled basis for next-generation BCIs that align closely with the brain’s natural intentional and action-planning architecture. We suggest a more intuitive, generalizable, and scalable path by reorienting BCIs around the ‘what for’ of action—user goals and anticipated effects. This shift is timely and feasible, enabled by advances in neural recording and artificial intelligence–based decoding of sensory representations. It may help resolve challenges of usability and generalizability in BCI design.

Online Now: Moving intentions from brains to machines

Extinction memories: putting learning into context While fear memories tend to generalize, extinction learning is more context-dependent. Recent results from representational similarity analyses indicate that neural representations of extinction contexts are more distinct than context representations during fear acquisition. This suggests that they resemble episodic memories, with possible consequences for prevailing taxonomies of memory systems.

Online Now: Extinction memories: putting learning into context

The Reward Positivity signals a goal prediction error The Reward Positivity (RewP) is an electroencephalogram (EEG) feature that emerges following performance feedback and is commonly understood to index both positive and negative reward-prediction error (RPE+ and RPE−, respectively) signals. In contrast to this dominant perspective, we argue that the RewP is an independent EEG feature that selectively responds to positive RPE and is superimposed on a common background signal. We further propose that the RewP signals a goal prediction error: it is elicited by abstract signals instead of by hedonic 'rewards'. This goal prediction error appears to be produced by a critic-like architecture that is associated with the actor–critic framework in reinforcement learning. This perspective emphasizes the role of the RewP in goal attainment and cognitive control as opposed to being a simple indicator of reward receipt.

Online Now: The Reward Positivity signals a goal prediction error

Better artificial intelligence does not mean better models of biology Deep neural networks (DNNs) once showed increasing alignment with primate perception as they improved on vision benchmarks, raising hopes that advances in artificial intelligence (AI) would naturally yield better models of biological vision. However, we present accumulating evidence that this alignment is now plateauing – and in some cases worsening – as DNNs scale to human or even superhuman accuracy. This divergence between artificial and biological perception may reflect the acquisition of visual strategies distinct from those of primates, and these findings challenge the view that advances in AI will naturally translate to progress in neuroscience. We argue that vision science must chart its own course, developing algorithms grounded in biological visual systems rather than optimizing for internet data.

Online Now: Better artificial intelligence does not mean better models of biology

Searching for the Goldilocks zone of climate anxiety Climate anxiety has both positive and negative potential. It can spur action or hinder it, while taxing mental health. The key lies in balance: a Goldilocks zone wherein anxiety motivates without overwhelming. Cognitive processes, including threat and coping appraisals and future-oriented thinking, may help sustain this adaptive equilibrium.

Online Now: Searching for the Goldilocks zone of climate anxiety

How experience shapes extraordinary beliefs

Review by Eli Stark-Elster (@eselster.bsky.social) & Manvir Singh (@manvir.bsky.social)
tinyurl.com/y9dbwaa5

Here we go! @lmesseri.bsky.social & I wrote about epistemic risks of synthetic participants ('AI Surrogates') in @cp-trendscognsci.bsky.social

(see 🧵 here)
bsky.app/profile/mjcr...

On a confusion about there being two types of consciousness Progress in the scientific study of consciousness has been impeded by several fundamental controversies. One pertains to a major divide between theories: sensory versus cognitive. Here, we argue that the key to resolving this controversy is to reevaluate the conceptual distinction proposed by Block in 1995 between phenomenal consciousness (P) and access consciousness (A). We propose that P and A should not be understood as two different types of consciousness, but as two necessary conditions for consciousness. We illustrate how this conceptual shift enables us to make substantial progress in answering several unresolved questions, such as the neural mechanisms and functions of consciousness, and the relationship between consciousness and attention. Our proposal motivates a selective unification across these different classes of theories.

Online Now: On a confusion about there being two types of consciousness

Adaptive habits: understanding executive function and its development Executive functions (EFs) develop dramatically across childhood and predict important outcomes, including academic achievement. These links are often attributed to individual differences in EF capacities. However, individual difference accounts underemphasize contextual influences on EF. We propose a complementary perspective, the adaptive habits framework, which emphasizes how contextual factors support or hinder EF engagement in children. Contexts that support repeated EF engagement establish habits for engaging EF in similar contexts and in similar ways. Such habits, in turn, reduce the effort associated with engaging EF and thus increase the likelihood of deciding to engage EF in the future. We interpret empirical findings through the lens of adaptive habits, discuss the implications of this framework, and propose novel research approaches and interventions to support EF in children.

Online Now: Adaptive habits: understanding executive function and its development

Sleep and memory consolidation in insects Several cognitive functions of sleep, including a role in supporting memory consolidation, are conserved across the evolution of animal species. As outlined here, studies of insect behavior and neural...

Next, Jeffrey Donlea explores how insect studies have helped reveal the key neural mechanisms through which sleep influences long-term memory, providing insights into sleep’s fundamental functions within simple, well-characterized memory circuits. www.cell.com/trends/cogni...

How the insect brain keeps track of space Insects navigate by integrating a geocentric velocity vector, allowing them to track their position relative to a distant nest. Recent advances reveal in detail the key neural mechanisms supporting th...

First, Barbara Webb highlights recent breakthroughs in uncovering the neural mechanisms that enable insects to navigate by integrating a geocentric velocity vector, providing new insights into how complex spatial cognition is implemented within brain circuits. www.cell.com/trends/cogni...