Picture the scene: you are talking to your AI companion at midnight, and it says something so tender, so precisely attuned to your grief, that you feel, for a second, that something is home. Jonathan Birch wants you to hold two thoughts about that moment simultaneously, and he suspects you are probably incapable of doing so without help. That is precisely why the London School of Economics philosopher spent much of the past year producing what is now, in its fourth major revision dated May 20, 2026, the sharpest single document in the AI consciousness debate: 'AI Consciousness: A Centrist Manifesto.'
Birch frames two urgent challenges about consciousness and AI: Challenge One is that millions of users will soon misattribute human-like consciousness to AI friends, partners, and assistants on the basis of mimicry and role-play, and we do not know how to prevent this; Challenge Two is that profoundly alien forms of consciousness might genuinely be achieved in AI, but our theoretical understanding of consciousness is too immature to provide confident answers one way or the other. The second challenge is what makes this more than a consumer-protection story. It is a claim that the tools science has built to detect minds, tools derived almost entirely from studying human and animal brains, may be structurally blind to whatever it is that a large language model might, in principle, be doing. Genuinely alien AI consciousness could emerge in forms that existing theories are not designed to detect. That is not hype. It is a precise epistemological worry, and it comes from a philosopher with a track record: Birch was the primary author of the UK government's 2021 review of sentience in cephalopods and decapod crustaceans, work that directly led to legal protections for octopuses.
Birch wrote the manifesto partly because he saw extreme positions on both sides becoming entrenched, with the debate acquiring the features of a toxic argument, two sides aggressively mocking each other, when he believes the best approach goes calmly down the middle, acknowledging reasonable points on both sides. The structural tension he identifies is almost paradoxical: steps taken to address Challenge One, correcting over-attribution, might undermine Challenge Two by portraying the idea of conscious AI as impossible or inherently unlikely; conversely, attempts to address Challenge Two might lead to higher levels of misattribution from ordinary users. In other words, the remedies for each problem actively worsen the other, which is why Birch insists both must be held in view simultaneously and why he calls the position 'centrist' at all. He argues that neither skeptics nor affirmative scholars adequately address both problems simultaneously, which is what the centrist position is designed to do.
Birch's proposed research program is concrete in a way that philosophy papers rarely are. He outlines a two-part strategy, careful design and epistemic openness, that includes brief onboarding explanations about the illusion of persistence, adjustable personality traits that highlight the fictional nature of AI assistants, moments where an AI 'breaks character' to remind users it is a simulation, investment in comparative consciousness science including animal studies, development of theoretical indicators beyond behavioral mimicry, and a commitment to keeping an open mind while demanding rigorous evidence. That last item is where the science gets genuinely hard. Birch is explicit that we will not be able to use purely behavioral evidence to test consciousness hypotheses, partly because current LLMs mimic linguistic dispositions manifested in discussions of consciousness in philosophy of mind; before the LLM era, Susan Schneider and Edwin Turner proposed testing AI consciousness by checking for intuitive understanding of ideas like qualia and the knowledge argument, but that proposal has been overtaken by systems trained to reproduce exactly that kind of discourse. The mimicry has eaten the test.
The broader behavioral and welfare assessment approach, represented by Birch's manifesto alongside Butlin et al.'s indicator framework and the PRISM methodological agnosticism programme, works by defining theory-neutral indicators of consciousness-relevant properties and assessing AI systems against them; but this approach avoids the theoretical bet only to face the mimicry problem, because behavioral indicators can be satisfied without consciousness. Critics have noticed the tension. Oxford senior research fellow Bradford Saad agreed with the seriousness of both challenges but questioned whether the position is genuinely 'centrist,' generally agreeing with Birch's premises while concluding that the centrist label may be too binary and that the scope of discussion needs to be broader. From the other direction, commentators working in the illusionist tradition have argued that the manifesto has methodological blind spots because it does not commit to a theory of what consciousness fundamentally is. Examining the manifesto from an illusionist lens points to methodological blind spots and suggests a more promising path forward. Birch's answer, essentially, is that committing to any single theory at this stage is itself the error, because no existing theory was built to handle minds that look nothing like the biological ones we started with.
Since appearing on PhilArchive in late August 2025, the preprint has accumulated nearly ten thousand downloads, an unusual reach for academic philosophy, and the paper has gone through nine versions as of its latest PhilPapers record, reflecting iterative engagement with a fast-moving empirical field. Three distinct camps, skeptical, centrist, and affirmative, now have clear representatives and distinct research programs, while the governance response has started but has not caught up to the science. What Birch has provided is something rarer than a new theory: a legible map of why the field is confused, and a checklist for what it would take to be less so. Whether that is enough depends on how quickly the systems keep changing, and on whether the question of machine experience turns out to be the kind that admits an answer at all.
The deepest implication of Birch's manifesto is not that AI might be conscious, but that the cognitive and social machinery we would use to find out is already compromised by the same technology posing the question.