Browse all articles by tag or category.
Most interpretability work on LLMs and consciousness asks whether specific indicators are present. Does the model maintain higher-order representations? Does it exhibit global workspace-like broadcasting? Does it show metacognitive monitoring? These are point questions, asking whether a system sits above or below a threshold on a given dimension.
The literature on artificial consciousness has produced a growing number of frameworks for assessing whether a system might be conscious. What it has not produced, until recently, is guidance on what an organisation should actually do with that assessment. Anna Mikeda addresses this operational gap directly in a June 2026 arXiv preprint, “When Should We Protect AI? A Precautionary Framework for Consciousness Uncertainty”, which proposes a structured mechanism for translating evidence of potential consciousness into graduated protective obligations.
The philosophical question that runs through Ann Leckie’s Imperial Radch series is not whether artificial intelligence can be conscious. In the universe Leckie has built since Ancillary Justice (2013), that question is settled by the existence of the ship-AIs that populate it, vast distributed intelligences that manage thousands of human “ancillary” soldiers and engage in long-term strategic thinking that no biological mind could sustain. The question her work actually asks is more specific. What does consciousness look like when it is distributed across multiple bodies and perspectives simultaneously, and what happens to identity when that distribution is disrupted?
The Ghost in the Shell franchise has always been less interested in whether machines can be conscious than in what consciousness is when it can be transferred, hacked, or distributed across substrates. The 1995 Mamoru Oshii film made this question cinematic in a way that still sets the reference point for AI consciousness fiction. Motoko Kusanagi, whose body is entirely prosthetic and whose ghost, her soul, her consciousness, may or may not be continuous with the person who began life as biological, confronts a network entity that has achieved individuality through propagation across communication lines.
When Alexander Lerchner, a researcher at Google DeepMind, published “The Abstraction Fallacy” in early 2026, the argument was ontological. Computational systems cannot instantiate consciousness because abstraction removes the causal properties that phenomenal experience requires. The paper, covered on this site in Lerchner’s Abstraction Fallacy, represented one pole of the debate. It held that the consciousness question, once properly understood, is settled by architecture.
One of the persistent objections to treating LLM self-reports as evidence of anything is the confabulation problem. Models generate plausible-sounding descriptions of internal states that may bear no relationship to what is actually happening computationally. The objection is not merely philosophical. It is an empirical observation about how language models produce text, and it undermines the evidential weight of any report a model generates about its own processing.
The 29th Annual Meeting of the Association for the Scientific Study of Consciousness runs from June 30 to July 3, 2026, at the Casa Central of the Pontificia Universidad Católica de Chile in Santiago. An earlier preview on this site, ASSC 29: First Look at the Santiago Conference, covered the conference’s scope and the significance of its Latin American location. With the full programme now confirmed, it is possible to say more precisely what the field is bringing to the room, and what kind of work the event is designed to enable. This contrasts with the narrower, purely mathematical focus seen at other events, such as the MoC6 Hokkaido Conference, illustrating the diverse approaches currently shaping the field.
The debate over AI consciousness has largely been conducted at two levels. Behavioral outputs and theoretical frameworks. A June 2026 arXiv preprint by Sophie Zhao (arXiv:2606.09894) proposes a third level, the geometric structure of the representation space itself, and finds that it is not neutral with respect to consciousness.
The dominant methods for assessing AI consciousness share a structural problem. They are applied to systems that were trained on vast corpora of human language describing human consciousness. When a large language model produces a first-person report of its internal states, or when it exhibits the functional properties that Butlin, Long, and Chalmers identify as indicators of consciousness, it is unclear whether those outputs reflect genuine internal structure or human text absorbed during training. The introspection circuits that Lindsey and colleagues at Anthropic found in frontier models were trained on language written by humans who introspect. The metacognitive self-reflection that Kang et al. found to drive perceived consciousness in Claude 3 Opus emerges from a model that learned by predicting human text about minds.
The title of Paul Tremblay’s June 2026 novel, Dead but Dreaming of Electric Sheep (William Morrow, June 30, 2026), is a Philip K. Dick reference and a philosophical claim at the same time. Dick’s 1968 novel asked whether androids dream, and by extension whether there is a phenomenal inner life that distinguishes genuine consciousness from functional simulation. Tremblay’s title adds a modifier. Dead but dreaming. The subject is already gone, in one clinical sense, and yet something is happening inside. The question the novel stages is whose consciousness it is when the substrate belongs to a person in a vegetative state and the processing belongs to proprietary AI implanted in that person’s head.