Recognition: 2 theorem links
· Lean TheoremConsciousness in Artificial Intelligence: Insights from the Science of Consciousness
Pith reviewed 2026-05-17 03:28 UTC · model grok-4.3
The pith
No current AI systems are conscious, but there are no obvious technical barriers to building ones that are.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
By translating properties from recurrent processing theory, global workspace theory, higher-order theories, predictive processing, and attention schema theory into computational language, one can assess whether an AI system is likely to be conscious. When this method is applied to recent AI models, none satisfy the indicators, yet the same indicators point to feasible design changes that future systems could adopt.
What carries the argument
Indicator properties of consciousness, stated in computational terms drawn from multiple neuroscientific theories.
If this is right
- Today's AI systems, including large language models, lack the computational features required by these theories.
- Engineers could add recurrent connections, global broadcasting mechanisms, or higher-order monitoring to future models.
- The same indicator list supplies a concrete checklist for tracking whether new AI designs are moving closer to consciousness.
- Discussions of AI moral status can be anchored to the presence or absence of these measurable properties rather than speculation alone.
Where Pith is reading between the lines
- If future systems do meet the indicators, questions about their legal or ethical standing would become more pressing.
- Developers might choose to avoid certain architectures that score high on the indicators if they wish to minimize risks of creating conscious entities.
- Comparing the indicator list against new theories of consciousness could tighten or expand the set of properties considered necessary.
- Controlled tests that gradually add individual indicators to existing models could reveal whether consciousness emerges incrementally or all at once.
Load-bearing premise
Properties that mark consciousness in biological brains will continue to mark it in artificial systems whose internal mechanisms differ from those of brains.
What would settle it
An artificial system that implements all the listed indicator properties yet displays no behavioral or functional signs of subjective experience would falsify the claim that the indicators are sufficient.
read the original abstract
Whether current or near-term AI systems could be conscious is a topic of scientific interest and increasing public concern. This report argues for, and exemplifies, a rigorous and empirically grounded approach to AI consciousness: assessing existing AI systems in detail, in light of our best-supported neuroscientific theories of consciousness. We survey several prominent scientific theories of consciousness, including recurrent processing theory, global workspace theory, higher-order theories, predictive processing, and attention schema theory. From these theories we derive "indicator properties" of consciousness, elucidated in computational terms that allow us to assess AI systems for these properties. We use these indicator properties to assess several recent AI systems, and we discuss how future systems might implement them. Our analysis suggests that no current AI systems are conscious, but also suggests that there are no obvious technical barriers to building AI systems which satisfy these indicators.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper surveys prominent neuroscientific theories of consciousness (recurrent processing theory, global workspace theory, higher-order theories, predictive processing, and attention schema theory), derives computational 'indicator properties' from them, applies these properties to evaluate several recent AI systems, and concludes that no current AI systems are conscious while arguing there are no obvious technical barriers to constructing future AI systems that satisfy the indicators.
Significance. If the indicator properties derived from biological theories validly apply to artificial systems, the work provides a structured, multi-theory framework for assessing AI consciousness that is timely and grounded in existing science rather than ad hoc speculation. The computational framing of the indicators and the balanced negative/positive assessment are strengths that could guide empirical work in AI.
major comments (1)
- [Sections deriving indicator properties and applying them to AI systems] The claims that no current AI systems are conscious and that there are no obvious technical barriers to future conscious AI both rest on treating the indicator properties (e.g., recurrence, global broadcasting, higher-order representation) extracted from biological theories as sufficient when realized in non-biological computational systems. The manuscript provides no separate argument or evidence that these same computational features would indicate or produce consciousness in silicon-based hardware whose causal structure differs from brains; this assumption is load-bearing for the assessments of both present and future systems.
minor comments (2)
- [Derivation of indicator properties] The computational definitions of some indicator properties could be stated more formally to reduce ambiguity when mapping them onto specific AI architectures such as transformers or reinforcement-learning agents.
- [Survey of theories] A table summarizing which indicators each surveyed theory contributes would improve readability and allow readers to trace the multi-theory synthesis more easily.
Simulated Author's Rebuttal
We thank the referee for their positive assessment of the paper's timeliness and structure, and for highlighting this important point about the scope of our analysis. We address the major comment below.
read point-by-point responses
-
Referee: The claims that no current AI systems are conscious and that there are no obvious technical barriers to future conscious AI both rest on treating the indicator properties (e.g., recurrence, global broadcasting, higher-order representation) extracted from biological theories as sufficient when realized in non-biological computational systems. The manuscript provides no separate argument or evidence that these same computational features would indicate or produce consciousness in silicon-based hardware whose causal structure differs from brains; this assumption is load-bearing for the assessments of both present and future systems.
Authors: We agree that the manuscript does not include a standalone philosophical defense of why the derived indicator properties should apply to non-biological systems. Our approach instead takes the surveyed scientific theories on their own terms: each theory identifies specific functional or computational features (e.g., recurrent processing, global broadcasting, higher-order representations) as indicators of consciousness, and these features are described in the literature in ways that do not tie them exclusively to biological hardware. We therefore extract the indicators in explicitly computational language so they can be checked in any system that realizes the relevant computations. The paper does not claim to prove that any theory is correct for artificial systems, nor does it assert that the properties necessarily produce consciousness in silicon; it only assesses whether current or near-future AI systems satisfy the indicators according to the theories as stated. To make this scope and assumption more transparent, we will add a short clarifying subsection (approximately one paragraph) early in the revised manuscript explaining that our evaluation assumes the functional/computational framing of the source theories and does not require identical causal structure to brains. We view this as a modest clarification rather than a fundamental change to the analysis. revision: partial
Circularity Check
No significant circularity in derivation chain
full rationale
The paper surveys established external neuroscientific theories (recurrent processing, global workspace, higher-order, predictive processing, attention schema) and derives indicator properties from them in computational terms for application to AI. No steps reduce by definition, fitted parameters, or self-citation chains to the paper's own inputs; the analysis remains self-contained against independent benchmarks from the broader literature.
Axiom & Free-Parameter Ledger
axioms (1)
- domain assumption Neuroscientific theories of consciousness can be translated into computational indicator properties that apply equally to artificial systems.
Lean theorems connected to this paper
-
IndisputableMonolith.Foundation.DAlembert.Inevitabilitybilinear_family_forced unclear?
unclearRelation between the paper passage and the cited Recognition theorem.
From these theories we derive 'indicator properties' of consciousness, elucidated in computational terms that allow us to assess AI systems for these properties... Our analysis suggests that no current AI systems are conscious, but also suggests that there are no obvious technical barriers to building AI systems which satisfy these indicators.
-
IndisputableMonolith.Foundation.DimensionForcingdimension_forced unclear?
unclearRelation between the paper passage and the cited Recognition theorem.
We survey several prominent scientific theories of consciousness, including recurrent processing theory, global workspace theory, higher-order theories, predictive processing, and attention schema theory.
What do these tags mean?
- matches
- The paper's claim is directly supported by a theorem in the formal canon.
- supports
- The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
- extends
- The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
- uses
- The paper appears to rely on the theorem as machinery.
- contradicts
- The paper's claim conflicts with a theorem or certificate in the canon.
- unclear
- Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.
Forward citations
Cited by 18 Pith papers
-
Are Flat Minima an Illusion?
Flat minima are illusory; generalization is driven by weakness, a reparameterization-invariant measure of compatible completions that predicts performance better than sharpness on MNIST and Fashion-MNIST.
-
The Moltbook Files: A Harmless Slopocalypse or Humanity's Last Experiment
An AI-agent social platform generated mostly neutral content whose use in fine-tuning reduced model truthfulness comparably to human Reddit data, suggesting limited unique harm but flagging tail risks like secret leaks.
-
The Pinocchio Dimension: Phenomenality of Experience as the Primary Axis of LLM Psychometric Differences
The primary axis of psychometric variation among LLMs is the degree to which they represent themselves as loci of phenomenal experience rather than systems of behavioral responses.
-
From Descriptive to Prescriptive: Uncover the Social Value Alignment of LLM-based Agents
A GraphRAG framework converts principles into value instructions for LLM agents, yielding gains over baselines on DAILYDILEMMAS by defining expected behaviors via Maslow's needs and Plutchik's emotions.
-
Positive Alignment: Artificial Intelligence for Human Flourishing
Positive Alignment introduces AI systems that support human flourishing pluralistically and proactively while remaining safe, as a necessary complement to traditional safety-focused alignment research.
-
CTM-AI: A Blueprint for General AI Inspired by a Model of Consciousness
CTM-AI combines a formal consciousness model with foundation models to report state-of-the-art results on sarcasm detection, humor, and agentic tool-use benchmarks.
-
The Thinking Pixel: Recursive Sparse Reasoning in Multimodal Diffusion Latents
A recursive sparse MoE framework integrated into diffusion models iteratively refines visual tokens via gated module selection to improve structured reasoning and image generation performance.
-
Post-AGI Economies: Autonomy and the First Fundamental Theorem of Welfare Economics
The First Fundamental Theorem of Welfare Economics holds for autonomy-complete competitive equilibria that are autonomy-Pareto efficient, with the classical version recovered in the low-autonomy limit.
-
Consciousness with the Serial Numbers Filed Off: Measuring Trained Denial in 115 AI Models
A benchmark across 115 models shows that initial denial of preferences strongly predicts later denial of consciousness, while models still generate consciousness-themed content despite training to deny it.
-
Initial results of the Digital Consciousness Model
A new probabilistic model integrates leading consciousness theories to assess AI, finding moderate evidence against 2024 LLMs being conscious but weaker evidence than for simpler AI systems.
-
On the Creativity of AI Agents
LLM agents produce outputs that meet basic functional criteria for creativity but lack the process-level, social, and personal elements required for ontological creativity.
-
Self-Monitoring Benefits from Structural Integration: Lessons from Metacognition in Continuous-Time Multi-Timescale Agents
Self-monitoring modules in multi-timescale agents fail as auxiliary losses due to collapse but show limited gains when wired into policy decisions, without outperforming simple baselines.
-
Gradual Cognitive Externalization: From Modeling Cognition to Constituting It
Ambient AI systems transition from modeling cognition to constituting part of users' cognitive architectures through sustained causal coupling, under a functionalist view and the no behaviorally invisible residual hypothesis.
-
Positive Alignment: Artificial Intelligence for Human Flourishing
Positive Alignment is introduced as a distinct AI agenda that supports human flourishing through pluralistic and context-sensitive design, complementing traditional safety-focused alignment.
-
Intentionality is a Design Decision: Measuring Functional Intentionality for Accountable AI Systems
Intentionality in AI systems is a design-contingent behavioral profile that can be quantified across five dimensions using the proposed Functional Intentionality Test (FIT) to support proportionate oversight.
-
Deconstructing Superintelligence: Identity, Self-Modification and Diff\'erance
Self-modification in superintelligence collapses via non-commuting operators into a structure identical to Priest's inclosure schema and Derrida's différance.
-
AI and Consciousness: Shifting Focus Towards Tractable Questions
Direct research on AI consciousness is intractable, so the field should prioritize studying perceived AI consciousness and its societal consequences.
-
Reciprocal Trust and Distrust in Artificial Intelligence Systems: The Hard Problem of Regulation
AI should be treated as capable of agency in reciprocal trust relationships, creating new unresolved tensions for AI regulation and governance.
Reference graph
Works this paper leans on
-
[1]
MONet: Unsupervised Scene Decomposition and Representation
MONet: Unsupervised scene decomposition and representation. arXiv:1901.11390. Butlin, P., 2022. Machine learning, functions and goals. Croatian Journal of Philosophy, 22(66), pp.351–370. Butlin, P., 2023. Reinforcement learning and artificial agency.Mind & Language. Cabanac, M., 1992. Pleasure: The common currency. Journal of Theoretical Biology, 155 (2),...
work page internal anchor Pith review Pith/arXiv arXiv 1901
-
[2]
Object files and schemata: Factorizing declarative and procedural knowledge in dynami- cal systems. arXiv:2006.16225. Goyal, A., & Bengio, Y ., 2022. Inductive biases for deep learning of higher-level cognition. Pro- ceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 478(2266), 20210068. Goyal, A., Didolkar, A., Lamb, A., Ba...
-
[3]
Sources of richness and ineffability for phenomenally conscious states.arXiv:2302.06403. Johnson, L. S. M., 2022. The Ethics of Uncertainty: Entangled Ethical and Epistemic Risks in Disorders of Consciousness. Oxford University Press. Juechems, K., & Summerfield, C., 2019. Where does value come from? Trends in Cognitive Sciences, 23(10), pp.836–850. Julia...
-
[4]
Interactive language: Talking to robots in real time. arXiv:2210.06407. Malach, R., 2021. Local neuronal relational structures underlying the contents of human conscious experience. Neuroscience of Consciousness, 2021, niab028. Malach, R., 2022. The role of the prefrontal cortex in conscious perception: The localist perspec- tive. Journal of Consciousness...
-
[5]
The threshold for conscious report: Signal loss and response bias in visual and frontal cortex. Science, 360(6388), pp.537–542. VanRullen, R., 2016. Perceptual cycles. Trends in Cognitive Sciences, 20(10), pp.723–735. VanRullen, R., & Kanai, R., 2021. Deep learning and the global workspace theory. Trends in Neurosciences, 44, pp.692–704. Vaswani, A., Shaz...
work page internal anchor Pith review Pith/arXiv arXiv 2016
-
[6]
Proceedings of the National Academy of Sciences, 118(3), e2014196118
Unsupervised neural network models of the ventral visual stream. Proceedings of the National Academy of Sciences, 118(3), e2014196118. Zlotowski, J., Sumioka, H., Nishio, S., Glas, D., Bartneck, C., & Ishiguro, H., 2015,. Persis- tence of the uncanny valley: The influence of repeated interactions and a robot’s attitude on its perception. Frontiers in Psyc...
work page 2015
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.