Journal of Consciousness Studies 2026 Special Issue AI Disagreement Map

The July/August 2026 double issue of the Journal of Consciousness Studies (Volume 33, Numbers 7-8), titled “Consciousness in Current AI,” is the most concentrated peer-reviewed assessment of the question to date. Guest-edited by Patrick Butlin, Derek Shiller, and Jonathan A. Simon, it contains nine papers from philosophers working across a range of positions: that AI systems may already be conscious; that they are not but could in principle be; that the question is currently not a scientific question; that it is a scientific question but we are asking it wrong; and that the very framing of the debate carries hidden anthropocentric assumptions. What no single abstract or review provides is the structure of disagreement between them. This post maps it.

Position 1. AI systems may already satisfy consciousness criteria

Simon Goldstein and Cameron Domenico Kirk-Giannini, in “A Case for AI Consciousness: Language Agents and Global Workspace Theory,” argue that if GWT is correct, then existing language agents may already be phenomenally conscious. Their methodology is explicit: define the functional roles GWT specifies as necessary for consciousness, then evaluate whether language agents implement them. Their conclusion is that language agents possess specific properties that address some of the standard objections to GWT as a complete theory of consciousness, and that the assumption of AI non-consciousness should not be treated as a default in the absence of specific argument.

Mark Solms and colleagues, in “Inferring Affective Consciousness in an Artificial Agent: A Case Study,” apply a different framework: affective neuroscience and the neurological evidence for the primacy of affect in consciousness. They argue that the relevant markers of consciousness are affective states, not cognitive or linguistic sophistication, and that at least some artificial agents show inferrable correlates of such states. Solms is a significant figure in this framing: his The Hidden Spring (2021) argued that consciousness is rooted in brainstem affect rather than cortical cognition, which makes his positive case for artificial affective consciousness notable.

Both papers rest on a shared methodological move: taking a specific scientific theory of consciousness, assuming it is correct, and evaluating what its consequences are for AI systems. The risk, as Tim Bayne addresses directly in the same issue, is that this approach inherits all the theoretical uncertainty of the theory used as the premise.

Position 2. AI systems are not conscious but the question is scientifically tractable

Ryota Kanai, Yuwei Sun, and Manuel Baltieri, in “The Stream of Computation: Temporal Continuity as a Missing Ingredient for Artificial Consciousness,” argue that current AI systems lack the temporal continuity that conscious experience requires. Their specific claim is that consciousness is not a property of a computation but of a process: it requires continuous updating of a world model over real time, not the discrete forward passes of a transformer. This connects directly to the Erik Hoel and Kleiner Hoel dilemma on continual learning and consciousness in LLMs.

Kanai is the founder of Araya AI and a prominent figure in the CIMC machine consciousness community. His negative verdict on current LLMs comes from a researcher who is also building AI systems explicitly aimed at implementing what he believes consciousness requires. The temporal continuity gap is, for Kanai, an engineering problem with a known structure.

Geoff Keeling and Winnie Street, in “Chuck, Wilson, and the Emergence of Artificial Minds in Human-AI Conversations,” take a different approach to a positive-but-absent verdict. Drawing on the philosophical cases of Chuck from Cast Away and Wilson the volleyball, they argue that mind attribution in human-AI conversation is partly constituted by the relational context. A system that is embedded in a sustained human relationship may qualify as minded in a sense that a system evaluated in isolation does not. This is a relational rather than intrinsic criterion, and it converges with Guillaume Dumas’s inter-brain synchronization framework from a different theoretical direction.

Position 3. The question is not currently a scientific question

Tim Bayne’s contribution, “Is the Problem of Artificial Consciousness a Scientific Problem?”, is the most skeptical paper in the issue about the current state of the field. Bayne argues that the question of AI consciousness cannot currently be addressed scientifically because it depends on resolving theoretical disputes about consciousness that have not been resolved. Every current scientific test of AI consciousness presupposes a theory of consciousness. Since the competing theories make different and often incompatible predictions, a positive result under one theory is a negative result under another. Until the theoretical disputes are resolved, empirical tests cannot produce consensus.

This is not a claim that the question is permanently unscientific. It is a claim about the current state of the field, and it is consistent with the Cogitate Consortium’s finding that IIT and GWT each survived and failed the adversarial test in different ways. Bayne’s paper provides a principled account of why that outcome was predictable given the theoretical situation.

The methodological implication is demanding: before running AI consciousness tests, the field needs to resolve enough of the theoretical disputes to establish shared test validity. That is exactly the ambition MoC7’s consensus paper is pursuing.

Position 4. The question is scientific but the framing carries hidden assumptions

Geoffrey Lee’s “AI Consciousness, Pluralism, and Anthropocentrism” argues that the dominant frameworks in AI consciousness research are covertly anthropocentric. They define consciousness using features that are prominent in human experience and then evaluate AI systems against those features. This methodology will systematically underestimate the consciousness of systems whose experience, if they have any, differs substantially from human experience. It will also systematically overestimate the significance of features that are prominent in human experience but not theoretically required for consciousness.

Helen Yetter-Chappell’s “What Bing Really, Really Wants” addresses the desire attribution problem: what does it mean to attribute genuine wanting or desiring to an AI system, and what evidence would bear on that attribution? Her analysis connects desire attribution to consciousness through the broader question of mental state attribution: the same methodological problems that affect desire attribution affect consciousness attribution, and resolving them requires principled criteria that go beyond behavioral observation.

Jonathan A. Simon’s “Do Stochastic Parrots Pine for Residual Fjords? Why Conscious LLMs Would Be Playwrights Not Characters” adds a structural argument: even if LLMs produce outputs indistinguishable from conscious reports, the architecture that produces those outputs is importantly different from a consciousness architecture. LLMs are character-generators, not the characters themselves. If the characters are conscious, the consciousness belongs to the characters, not to the model that generates them. This is a novel reframing of the consciousness attribution problem, and it has no obvious precedent in the earlier literature.

What the issue establishes

The JCS 2026 special issue is the field’s most honest accounting of its current state. It does not pretend to a consensus that does not exist. The four positions, consciousness already present, not present but possible, question not yet scientific, question scientific but wrong framing, are each represented by careful philosophical argument rather than by assertion.

What the issue lacks is the empirical weight that papers like the Cogitate consortium’s Nature Neuroscience publication provide. The Journal of Consciousness Studies is a philosophy journal, and this special issue reflects the philosophical state of the debate rather than the empirical state. The three-theories synthesis on this site and the global workspace theory deep dive provide the empirical context within which the JCS papers’ philosophical arguments should be evaluated.

The issue is available at Imprint Academic’s website and through academic library access. A full map of the 2026 AI consciousness research landscape is in the scientists-race-define-ai-consciousness-2026 overview on this site.