The DOI matches its Crossref record; 96 of the 96 references it cites were found in a registry or an index; 9 of the 130 sources its claims cite were checked for support.
What is consciousness, and could machines have it?
Open accessAbstract
The controversial question of whether machines may ever be conscious must be based on a careful consideration of how consciousness arises in the only physical system that undoubtedly possesses it: the human brain. We suggest that the word "consciousness" conflates two different types of information-processing computations in the brain: the selection of information for global broadcasting, thus making it flexibly available for computation and report (C1, consciousness in the first sense), and the self-monitoring of those computations, leading to a subjective sense of certainty or error (C2, consciousness in the second sense). We argue that despite their recent successes, current machines are still mostly implementing computations that reflect unconscious processing (C0) in the human brain. We review the psychological and neural science of unconscious (C0) and conscious computations (C1 and C2) and outline how they may inspire novel machine architectures.
Overview
CoChat Summary
TL;DR
The paper argues that “consciousness” conflates two computations—C1 global availability/broadcasting and C2 self-monitoring—while most current AI implements only unconscious C0 processing. It proposes that machines endowed with C1 and C2 would behave as though conscious, though philosophical questions about subjective experience remain beyond its scope.
Key findings
- Unconscious C0 processing is computationally deep: subliminal primes support invariant face/word recognition, semantic violations produce an N400 response as late as 400 ms, and unconscious mechanisms contribute to cognitive control, evidence accumulation, and reinforcement learning; nonconscious information decays quickly (~1 s), though sometimes slower than expected.
- C1 consciousness is an all-or-none, capacity-limited global workspace: conscious access to item A prevents or delays perception of item B in attentional blink/refractory-period paradigms, and conscious percepts correlate with “ignition” in high-level cortical networks.
- C2 self-monitoring includes confidence, error detection, metamemory, and reality monitoring, linked especially to prefrontal cortex; it appears early, with 1-year-old infants showing an error-related negativity and 1.5-year-olds using opt-out/help and persistence measures.
- Current machines mostly implement C0 computations; possible routes toward C1/C2 include blackboard systems, Pathnet, Bayesian confidence models, second-order networks that predict a first network’s performance, and adversarial learning for reality monitoring.
Methods
- Review/conceptual article; no new experiments or primary data are reported.
- Synthesizes cognitive neuroscience, psychology, and AI evidence, including behavioral, fMRI, single-neuron, EEG, and infant studies, to propose the C0/C1/C2 taxonomy.
- Main analysis is theoretical: it maps unconscious, globally available, and self-monitoring computations onto brain systems and discusses how machine architectures might implement C1 and C2.
Limitations
- The paper states that it does not deal with vigilance/wakefulness because of space, that philosophical questions about the experiential component (“what it is like”) lie beyond its scope, and that a purely functional definition of consciousness may leave some readers unsatisfied.
- Evident from the text: as a review, it presents no new empirical tests of its machine proposals; the C1/C2 framework is functional and does not resolve subjective experience.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
8 Valid · 7 Not valid · 65 Not checked · 42 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Remarkably, in humans, such learning appears to proceed even when the cues, reward, or motivation signals are presented below the consciousness threshold ( 17 , 18 ).”
What the sources say
- Partly supportedcites [17] 17. M. Pessiglione et al ., Neuron 59 , 561 -567 (2008). sourceExperimental — accuracy not yet measured
The Results report learning from unseen cues, supporting the cue part of the claim. But the Results describe prediction errors as based on visible outcomes, and the Experimental Procedures say monetary outcomes were displayed at the end of each trial. This paper does not show learning when rewards or motivation signals themselves were presented below the threshold of awareness. (LLM, conf=0.90)
- Partly supportedcites [18] 18. M. Pessiglione et al ., Science 316 , 904 -906 (2007). sourceExperimental — accuracy not yet measured
The paper’s Abstract (page unavailable) reports that masked monetary incentives affected hand-grip force even when subjects could not report the reward at stake, supporting unconscious motivation in humans. It does not establish that reinforcement learning proceeded below the consciousness threshold; its task tested how subliminal incentives energized behavior, not whether participants learned from them. (LLM, conf=0.87)
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Bayesian approaches to AI ( 2 , 73 ) have recognized the usefulness of learning generative models that can be jointly used for actual perception (present), prospective planning (future), and retrospective analysis (past).”
What the sources say
- Partly supportedcites [2] 2. B. M. Lake, T. D. Ullman, J. B. Tenenbaum, S. J. Gershman, Behav. Brain Sci. 2016 , 1 -101 (2016). sourceExperimental — accuracy not yet measured
Section 4.2, “Learning as rapid model building” (page unavailable), links causal generative models to perception and describes models of environmental transitions for control. Section 4.1, “Developmental start-up software” (page unavailable), says simulated planning processes can predict what an agent might do next and support inverse inference from observed actions. These passages support related uses of models, but do not establish that a learned generative model is jointly used for perception, prospective planning, and retrospective analysis. (LLM, conf=0.79)
- Not matchedcites [73] 73. J. B. Tenenbaum, C. Kemp, T. L. Griffiths, N. D. Goodman, Science 331 , 1279 -1285 (2011).
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“In nonhuman primates, neurons in the prefrontal cortex distinguish between normal visual perception and active maintenance of the same visual content in memory ( 64 ).”
What the sources say
- Supportedcites [64] 64. D. Mendoza-Halliday, J. C. Martinez-Trujillo, Nat. Commun. 10.1038/ncomms15471 (2017). source
The Abstract reports recordings from monkey lateral prefrontal cortex while the same visual feature—a stimulus’s motion direction—was either visible or held in memory. It states that population activity could distinguish whether the direction was perceived or memorized. Page numbers are not provided in our copy. (LLM, conf=0.92)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“By sharing their confidence signals, two persons achieve a better performance in collective decision-making than that of either person alone ( 69 ).”
What the sources say
- Partly supportedcites [69] 69. B. Bahrami et al ., Science 329 , 1081 -1085 (2010). sourceExperimental — accuracy not yet measured
The Main Text reports that communicating confidence improved a dyad’s performance beyond that of its better member. But it limits this benefit to observers with similar sensitivity; when sensitivities differed markedly, the dyad could perform worse than its better member. The citing sentence omits that condition. (LLM, conf=0.91)
plainstates a causen = 2- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“Compared with related words (for example, animal-dog), semantic violations (for example, furniture-dog) generate a brain response as late as 400 ms after stimulus onset in temporal-lobe language networks, even if one of the two words cannot be consciously detected ( 9 , 10 ).”
What the sources say
- Not matchedcites [9] 9. S. J. Luck, E. K. Vogel, K. L. Shapiro, Nature 383 , 616 -618 (1996).
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites [10] 10. S. van Gaal et al ., Philos. Trans. R. Soc. Lond. B Biol. Sci. 369 , 20130212 (2014).
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Not matched
69 more claims
Create a free account to see all 122 claims and to verify the 1 reference still open.
- It may state two separate claims as one. 29
- Probably not a substantial claim. 22
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“In the attentional blink, a pair of stimuli are shown in rapid succession. If the first stimulus is attended to, and if the second stimulus is shown very quickly after the initial one, the second stimulus is not consciously experienced. But if the subject was not directing their attention at the first stimulus, the second stimulus is experienced consciously (Raymond et al. 1992; Dehaene et al. 201”
From A Case for AI Consciousness: Language Agents and Global Workspace Theory (2024) · cites this paper as Dehaene et al. 2017
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“The temporal stability implied by hierarchical depth (cf. Seth, 2009; Dehaene et al., 2017) should therefore be distinguished from temporal thickness in the sense implied by active inference as planning.”
From 'Seeing the Dark': Grounding Phenomenal Transparency and Opacity in Precision Estimation for Active Inference (2018) · cites this paper as Dehaene et al., 2017
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“For example, a key property of the global workspace is that it can sustain representations over time, coordinating activity in modules to support complex tasks [ 22,38,53].”
From Identifying indicators of consciousness in AI systems (2026) · cites this paper as [38]
The cited source says what the claim attributes to it.
In the citing paper
“Indeed, many of the perceptual and cognitive processes in the human brain occur without consciousness (Dehaene et al. 2017).”
From The Measurement Problem of Consciousness (2020) · cites this paper as Dehaene et al. 2017
The cited source says what the claim attributes to it.
In the citing paper
“There is also debate about the presence or possibility of consciousness in nonhuman animals [16 -21], neural organoids [22 -24], AI systems [25 -29], and xenobots [30,31].”
From Tests for consciousness in humans and beyond (2024) · cites this paper as [25]
6 more claims about this paper
Create a free account to see every claim other papers make about this one.
Citations
Cites
James, W. (1913) · Henry Holt and Company eBooks · doi:10.1037/11059-000 · Cited by 25,226
Generative adversarial networks
Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., & Bengio, Y. (2020) · Communications of the ACM · doi:10.1145/3422622 · Cited by 18,384
Mastering the game of Go with deep neural networks and tree search
Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., van den Driessche, G., Schrittwieser, J., Antonoglou, I., Panneershelvam, V., Lanctot, M., et al. (2016) · Nature · doi:10.1038/nature16961 · Cited by 15,612
The principles of psychology, Vol I.
James, W. (1890) · doi:10.1037/10538-000 · Cited by 4,514
(1988) · Cited by 3,656
A neural system for error detection and compensation
Gehring, W. J., Goss, B., Coles, M. G. H., Meyer, D. E., & Donchin, E. (1993) · Psychological Science · doi:10.1111/j.1467-9280.1993.tb00586.x · Cited by 2,862
Building machines that learn and think like people
Lake, B. M., Ullman, T. D., Tenenbaum, J. B., & Gershman, S. J. (2017) · Behavioral and Brain Sciences · doi:10.1017/s0140525x16001837 · Cited by 2,550
Building Machines That Learn and Think Like People
Lake, B. M., Ullman, T. D., Tenenbaum, J. B., & Gershman, S. J. (2016) · arXiv · doi:10.48550/arxiv.1604.00289 · Cited by 2,550
Towards a cognitive neuroscience of consciousness: basic evidence and a workspace framework
Dehaene, S. (2001) · Cognition · doi:10.1016/s0010-0277(00)00123-2 · Cited by 2,498
Self-citationThe cognitive neuropsychology of schizophrenia
(1993) · Choice Reviews Online · doi:10.5860/choice.31-0601 · Cited by 1,901
Research works printed but not linked: 2
- J. Dunlosky J. Metcalfe Metacognition (Sage Publications 2008).
- C. Fernando D. Banarse C. Blundell Y. Zwols D. Ha A. A. Rusu A. Pritzel D. Wierstra arXiv:170108734 [cs.NE] (2017).
Cited by
Conscious Processing and the Global Neuronal Workspace Hypothesis
Mashour, G. A., Roelfsema, P., Changeux, J.-P., & Dehaene, S. (2020) · Neuron · doi:10.1016/j.neuron.2020.01.026 · Cited by 1,164
“In the future, it will be fascinating to see whether some of the present ideas can be made precise enough to be implemented in an actual computational device (Dehaene et al., 2017).”
BackgroundSelf-citationSeth, A. K., & Bayne, T. (2022) · Nature Reviews Neuroscience · doi:10.1038/s41583-022-00587-4 · Cited by 880
Bengio, Y., Lecun, Y., & Hinton, G. (2021) · Communications of the ACM · doi:10.1145/3448250 · Cited by 661
“The nature of system 2 processing and cognitive neuroscience theories for them 5,30 suggests several such inductive biases and architectures, 11,45 which may be exploited to de-sign novel deep learning systems.”
BackgroundUnderstanding the Higher-Order Approach to Consciousness
Brown, R., Lau, H., & LeDoux, J. E. (2019) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2019.06.009 · Cited by 573
“In a recent review, global broadcasting and higher-order monitoring were cast as two orthogonal dimensions of processing [73].”
BackgroundSelf-citationThe Border Between Seeing and Thinking
Block, N. (2023) · doi:10.1093/oso/9780197622223.001.0001 · Cited by 367
Metadata
- Authors
- Dehaene, S., Lau, H., & Kouider, S.
- Affiliations
- Université Paris-Sud (FR)
- University of California, Los Angeles (US)
- Centre National de la Recherche Scientifique (FR)
- Corresponding author
- Dehaene, Stanislas
- Venue
- Science
- Published
- 2017-10-27
- Volume
- 358
- Issue
- 6362
- Pages
- 486-492
- ISSN
- 0036-8075 (ISSN-L)
- 1095-9203
- OpenAlex
- W2765831433
- Cited by
- 855
- References
- 96
- Citations by year
- 2026: 92 · 2025: 76 · 2024: 153 · 2023: 124 · 2022: 60 · 2021: 82 · 2020: 96 · 2019: 81 · 2018: 85 · 2017: 6
- FWCI
- 11.56
- Type
- journal article
- Language
- en
- Open access
- green
- escholarship.org
- Locations
- Science (publishedVersion)
- PubMed (publishedVersion)
- eScholarship (California Digital Library) (submittedVersion)
- HAL (submittedVersion)
- Funders
- National Institute of Neurological Disorders and Stroke (award348207, R01NS088628)
- Air Force Office of Scientific Research (award348208, FA9550-15-1-0110)
- Canadian Institute for Advanced Research (award348203)
- European Research Council (award348204, NeuroConsc, award348205, Metaware)
- Agence Nationale de la Recherche (award348206, ANR-10-LABX-0087 and ANR-10-IDEX-0001-02)
- Institut national de la santé et de la recherche médicale (award348200)
- Commissariat à l’Énergie Atomique et aux Énergies Alternatives (award348201)
- Collège de France (award348202)
- Topics
- Social Sciences › Arts and Humanities › Philosophy › Philosophy and Theoretical Science
- Physical Sciences › Computer Science › Artificial Intelligence › Psychiatry, Mental Health, Neuroscience
- Life Sciences › Neuroscience › Cognitive Neuroscience › Embodied and Extended Cognition
- Keywords
- consciousness, artificial consciousness, unconscious processes, self-monitoring, subjective certainty, brain information processing, computer architecture, neural basis of consciousness
- MeSH terms
- Machine Learning, Brain, Computer Simulation, Consciousness, Humans, Models, Neurological, Robotics
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: What is consciousness, and could machines have it? https://app.cochat.ai/papers/10.1126/science.aan8871
- The paper
- Dehaene, S., Lau, H., & Kouider, S. (2017). What is consciousness, and could machines have it? Science. https://doi.org/10.1126/science.aan8871
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.