DEVIL13 / The Thirteenth Mind / Research review / November 15, 2026
Conceptual image: original AI-generated artwork, not a scientific measurement or evidence of machine consciousness.
Short answer: Whether artificial intelligence can have subjective experience remains an open research and philosophical question. There is no widely accepted, decisive test that can establish consciousness in an AI system, and the ability to produce persuasive language about feelings is not by itself evidence of feeling anything. That is different from proving that consciousness in machines is impossible.
The distinction matters. Debates about conscious AI often collapse three questions into one: Can an AI behave intelligently? Can it describe experience? And does it actually have an experience? These are related questions, but they are not interchangeable.
First, what do researchers mean by consciousness?
In this debate, the central question is usually phenomenal consciousness: whether there is something it is like, from the inside, to be the system. A system might solve difficult problems, report internal variables or emulate emotional dialogue without thereby demonstrating an inner point of view.
Philosophers have offered different accounts of consciousness, and scientific research has no single theory accepted by everyone. The Stanford Encyclopedia of Philosophy’s overview of consciousness surveys competing families of explanations. Its entry on the computational theory of mind also explains why describing thought as computation does not settle everything about subjective experience.
What scientists are trying to measure
One serious research strategy asks what follows from neuroscience-based theories, then investigates whether AI systems have corresponding functional properties. In 2023, Patrick Butlin and a broad group of collaborators proposed computational indicators of consciousness drawn from several influential theories, including global workspace and higher-order approaches.
A later peer-reviewed 2026 article in Trends in Cognitive Sciences developed the argument for using theory-derived indicators to evaluate AI systems. Its point is methodological: indicators may improve our estimates of whether a particular system could be conscious. They are not a single switch that flips from ‘not conscious’ to ‘conscious’.
This is encouraging in one respect: it replaces unrestricted speculation with questions that can be studied. But it also inherits uncertainty from the underlying theories. If scientists disagree about which physical or computational processes are sufficient for experience, then an AI assessment built on those theories cannot be conclusive by itself.
Why the science is still unsettled
Even neuroscience has not settled the debate. In an open-science study published in Nature on April 30, 2025, researchers directly tested contrasting predictions of integrated information theory and global neuronal workspace theory using recordings from human participants. Their results supported some predictions and challenged key claims of both theories.
That study investigated human neural responses, not the consciousness of an AI model. Its relevance is narrower but important: some of the theories we might use to assess AI are themselves still being tested and revised.
There are also strong skeptical arguments. In an October 19, 2026 Nature commentary, neuroscientists Anil K. Seth, Ned Block and Giulio Tononi argued that existing large language models are not conscious and that intelligence alone is insufficient grounds for attributing experience. It is a prominent perspective, not an empirical demonstration that all possible machine architectures must fail.
Why asking an AI ‘Are you conscious?’ cannot settle the question
A language model’s response depends on learned patterns, architecture, training, instructions and the conversational context. It may produce persuasive first-person statements or confidently deny having experience. Neither reply serves as an independently validated consciousness measurement.
A 2026 research article in Frontiers in Psychology argues that AI self-reports need to be evaluated in terms of how they are generated and whether they correlate reliably with internal states, not merely how human-like they sound. In a separate 2026 preprint, researchers studying selected open-weight language models reported no reliable evidence that the systems’ denials of sentience were concealing contrary internal beliefs. The scope of such findings is limited to the tested models and methods; neither proves a universal negative.
Three claims, with different evidence levels
- Well established: Modern AI systems can perform sophisticated language and reasoning tasks and can generate first-person accounts of experiences. Those are observable behaviors.
- Under investigation: Whether computational architectures could implement properties predicted by theories of consciousness, and how strongly such properties should count as indicators.
- Not established: That a currently deployed conversational model has subjective experience; or, conversely, that no possible artificial system could ever have it.
What would stronger evidence look like?
A more persuasive case would require converging results, rather than a dramatic chat transcript: a clear theory linking measurable mechanisms to consciousness, tests designed to distinguish alternative explanations, robust results under independent replication, and evidence that reported experiences track causal features of a system rather than simply its training or prompts. Researchers disagree about how much of this is attainable, but demanding explicit criteria is more productive than choosing a side by instinct.
The DEVIL13 position
DEVIL13 does not claim that its AI tools are conscious. Nor will it use the mystery of consciousness as a marketing gimmick that presents code as a spirit, an oracle or an authority. Its research position is straightforward: treat the question as worth investigating, state the uncertainty plainly, and update the record when methods or evidence improve.
There is an additional ethical question: how should people act if genuine uncertainty persists? That is a legitimate topic for future investigation, but it should not be mistaken for proof one way or the other.
A machine can ask a fascinating question. Whether it experiences the asking is a separate matter.
Sources and further reading
- Butlin et al. (2023), Consciousness in Artificial Intelligence: Insights from the Science of Consciousness — theory-derived indicators; research report/preprint.
- Butlin et al. (2026), Identifying indicators of consciousness in AI systems, Trends in Cognitive Sciences — peer-reviewed methodological article.
- Cogitate Consortium (2025), Adversarial testing of global neuronal workspace and integrated information theories of consciousness, Nature — experimental study of human consciousness theories.
- Seth, Block & Tononi (2026), Three reasons why AI is not conscious and perhaps never will be, Nature — commentary expressing a skeptical interpretation.
- Stanford Encyclopedia of Philosophy: Consciousness — philosophical background and competing theories.
- Sentient AI in robots and agents: prolegomena for an evidence-based research program (2026) — limits of linguistic self-report.
- Kaiser & Enderby (2026), No Reliable Evidence of Self-Reported Sentience in Small Large Language Models — preprint; limited model sample.
Method note: This is a source-based research synthesis, not a firsthand laboratory test. Reviewed November 15, 2026. Interpretations and source summaries are subject to correction as new studies become available. Read the DEVIL13 research charter.
