Skip to content

Stephen M Fleming

Department of Experimental Psychology, University College London, London, United Kingdom.

5 papers in the library · 175 citations · publishing 2024-2025

Papers

Folk psychological attributions of consciousness to large language models.

Neuroscience of consciousness January 1, 2024 Clara Colombatto, Stephen M Fleming 71 citations

A majority of a sample of 300 US residents were willing to attribute some possibility of phenomenal consciousness—subjective experiences like feelings and sensations—to large language models. These attributions were robust, predicting attributions of mental states typically linked to phenomenality, but also flexible, as they varied with individual differences such as how often participants used the technology. The findings suggest that folk intuitions about AI consciousness can diverge from expert views, with potential implications for the legal and ethical treatment of AI.

Unpacking the complexities of consciousness: Theories and reflections.

Neuroscience and biobehavioral reviews March 1, 2025 Liad Mudrik, Melanie Boly, Stanislas Dehaene et al. 68 citations

In a structured public debate at the 2022 meeting of the Association for the Scientific Study of Consciousness, proponents of five major theories—Global Neuronal Workspace Theory, Higher-Order Theories, Integrated Information Theory, Recurrent Processing Theory, and Predictive Processing—clarified their theories' core mechanisms, foundational premises, and what each theory aims to explain. The discussion revealed more controversy than agreement, particularly on the most basic questions: what consciousness is, how to identify conscious states, and what any adequate theory must account for. Addressing these foundational disagreements is essential for advancing the field and enabling meaningful comparison of competing theories.

Sensory Horizons and the Functions of Conscious Vision.

The Behavioral and brain sciences April 21, 2025 Stephen M Fleming, Matthias Michel 26 citations

Conscious vision is surprisingly slow, with unconscious integration windows lasting up to 400 milliseconds, as shown by postdictive effects. Because it is slow, conscious vision cannot guide online actions; instead, it evolved to support offline cognition, such as planning and internal simulation. This shift likely accompanied the water-to-land transition, where larger terrestrial visual horizons made model-based planning advantageous over the fast, reflexive actions typical of aquatic environments. The capacity for internal simulation created pressure for reality monitoring—distinguishing internal from external signals and solving when to stop integrating evidence to fix a model of reality. This reality monitoring function is linked to the emergence of consciousness, in line with higher-order theories. The account generates novel predictions about conscious versus unconscious vision in aquatic and terrestrial animals.

Identifying content-invariant neural signatures of perceptual vividness.

PNAS nexus February 1, 2024 Benjy Barnett, Lau M Andersen, Stephen M Fleming et al. 7 citations

Perceptual vividness—the intensity of conscious experience—varies across different experiences, but how the brain registers this variation is unclear. In other psychological domains like number or reward, magnitude is represented independently of sensory features. Reanalyzing existing magnetoencephalography and functional MRI data from two studies that quantified vividness via subjective awareness and visibility ratings, evidence emerged for content-invariant neural signatures of vividness distributed across visual, parietal, and frontal cortices. These findings suggest that neural correlates of subjective vividness may share properties with magnitude codes in other cognitive domains.

Confidence reports during perceptual decision making dissociate from changes in subjective experience.

Communications psychology May 21, 2025 Nicolás Sánchez-Fuenzalida, Simon van Gaal, Stephen M Fleming et al. 3 citations

Confidence reports during perceptual decision-making do not uniquely reflect subjective experience. Across two experiments with 204 participants and three bias manipulations, non-perceptual factors such as changes in stimulus base rates or asymmetric payoff matrices leaked into confidence judgments. This shows that confidence can be influenced by response biases unrelated to actual perception, complicating its use as a pure measure of subjective experience. The relative strength of biases in first-order choices versus confidence may help distinguish whether a manipulation truly alters perception or only affects decision strategy.