Vishne and Deouell Track Sustained and Transient Conscious Content in the Brain
The content you consciously hold stays stable in sensory cortex for as long as the stimulus lasts, while the prefrontal cortex fires one brief burst at stimulus onset and then stops carrying the content. Gal Vishne, Edden M. Gerber, Robert T. Knight and Leon Y. Deouell published that split in Cell Reports on July 25, 2023, from intracranial recordings in ten human patients viewing images of different durations (DOI 10.1016/j.celrep.2023.112752). The study is a direct probe of the two strongest predictions in consciousness science. Sustained experience needs a neural representation that persists with it. The sensory regions provided exactly that, in a stable subspace the authors call the experience subspace. The prefrontal cortex provided a transient onset response, an ignition, with no sustained trace, in trials that required no report.
The multi-duration paradigm separates onset from presence
Most consciousness experiments study a stimulus that flashes and disappears, so every measured response mixes onset with ongoing perception. The multiduration design breaks that confound. Ten patients with drug-resistant epilepsy, implanted with subdural electrodes for clinical reasons, viewed diverse images from four categories, including faces, objects and a watch subcategory, for 300, 900 or 1500 milliseconds. On 10 percent of trials the patient had to respond, to hold attention, and those trials were excluded from analysis, which keeps report-related signals out of the data entirely. Because the same images appeared at multiple durations, the analysis can separate what fires when an image starts from what tracks the ongoing presence of the image. The study analyzed categories and individual exemplars, with exemplar level decoding reaching down to specific objects.
Sensory regions hold a stable experience subspace
In occipital and ventral temporal regions the activation magnitude changed dramatically over the presentation, as earlier studies had found when activity drops shortly after onset. The content did not. Decoders trained on the population response at stimulus onset kept discriminating categories and even individual exemplars through the full presentation time. The representational geometry stayed stable, and the same classifier worked from first millisecond to last. The authors name this the experience subspace, an embedded subspace of the vast space of possible neural responses that maintains the distinctions between categories and exemplars while the rest of the response varies moment to moment. Their proposal is precise. To the extent sensory representations contribute to conscious experience, it is the projection of the population response onto this stable subspace that gives a percept its quality, while the variable dimensions serve other functions or reflect stochasticity. The subspace also explains a familiar puzzle. Repeating a stimulus suppresses neuronal firing everywhere, yet the percept stays identical, and the paper’s answer is that perception follows location within the subspace, not response amplitude.
The prefrontal cortex ignites at onset and does not persist
The prefrontal result answers a standing dispute. No-report studies had found minimal prefrontal activation when subjects need not report awareness, which suggested the prefrontal cortex serves reporting rather than experience. Vishne and colleagues found something more specific. In trials that required no response, the prefrontal cortex showed a burst of visual information at stimulus onset, an ignition, carrying accurate category information and even exemplar information for non-target categories, and that representation did not persist. Parietal cortex behaved the same way. There was no offset response when the stimulus ended. So the human prefrontal cortex does carry content, not just non-specific task activity, but it carries it transiently, and the content it carries tracks the arrival of a new stimulus rather than its continuing presence. The authors note the result does not rule out a sustained prefrontal role in monitoring that carries no specific content.
Both the IIT and GNWT predictions hold
The COGITATE adversarial collaboration adopted this multiduration paradigm as one of its two key tests and registered detailed predictions from each side. Integrated Information Theory predicts that activity in the posterior hot zone, occipital and ventral temporal regions, persists with a stable representation as long as the visual experience persists, and stays largely agnostic about the prefrontal cortex. Global Neuronal Workspace Theory predicts transient onset and offset representations in the prefrontal cortex without a sustained component, including without overt responses. Both predictions bore out. Posterior visual areas showed stable persistent representation tracking stimulus duration. The prefrontal cortex showed onset ignition without persistent representation. The registered workspace prediction also included an offset component, and no offset response appeared. The first Cogitate test read both theories as partially surviving, and this result sharpens that reading. The temporal structure of experience maps onto the anatomy, with sustained content in sensory cortex and discrete updating in frontoparietal regions, so the where and the when of consciousness stop being separate questions.
Comparison to The Consciousness AI
This project binds its workspace with five Kuramoto oscillators and reads their alignment as a bid boost, so the distinction this paper draws, sustained content against transient ignition, names a measurement the project has not run. The Brain Waves Console shows the agent’s five oscillators in its agent chapter and the workspace they feed, and nothing in that pipeline yet carries a content-specific sustained state that could be contrasted against an onset response. The multiduration design is the template such a measurement would need, since it separates onset signals from presence signals without a report. The phase gated perception evidence on this site shows a different temporal handle on perception, the wave phase at stimulus arrival, and the two results together describe a cortex where what is perceived and when it is perceived both depend on ongoing dynamics.
What the study settles and what stays open
The study settles three points at intracranial resolution in humans. Sustained conscious content is represented stably in sensory cortex across the full duration of a percept, decodable from onset to offset. The prefrontal cortex carries specific visual content transiently, without report, and does not sustain it. And the registered predictions of the two leading theories both fit, which is why the COGITATE program adopted the paradigm. The open list is equally specific. The patients performed a task that made stimuli task relevant even in analyzed trials, so a fully passive version remains untested at this resolution. The subspace proposal is an interpretation of stable geometry, and identifying which dimensions carry the quality of a percept is a program, not a result. And the finding that neither theory’s full picture held is the reason the adversarial collaboration continues, with phase two work underway.
Sources
- Vishne, G., Gerber, E.M., Knight, R.T., Deouell, L.Y. (2023). Distinct ventral stream and prefrontal cortex representational dynamics during sustained conscious visual perception. Cell Reports 42, 112752. DOI 10.1016/j.celrep.2023.112752
- Cogitate Consortium et al. (2025). Adversarial testing of global neuronal workspace and integrated information theories of consciousness. Nature. DOI 10.1038/s41586-025-08888-1
Researchers covered here
- Leon Y. DeouellEdmond and Lily Safra Center for Brain Sciences and Department of Psychology, The Hebrew University of JerusalemCo-leader of the COGITATE adversarial collaboration paradigm, and senior author of the sustained versus transient content study (Vishne, Gerber, Knight and Deouell, Cell Reports 2023)
- Gal VishneEdmond and Lily Safra Center for Brain Sciences, The Hebrew University of JerusalemFirst author of the sustained versus transient representational dynamics study (Vishne, Gerber, Knight and Deouell, Cell Reports 2023) and the experience subspace proposal