Bales and Gabriel: AI Consciousness as a Political Problem

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.

Six weeks later, a different pair of DeepMind researchers published a paper that does not dispute Lerchner’s ontological argument but makes it beside the point. Adam Bales and Iason Gabriel’s “Artificial Minds, Human Disagreement: The Politics of AI Consciousness,” released June 15, 2026, on SSRN via the Google DeepMind research portal, shifts the entire frame: the question is not whether AI systems are conscious, but how society should function when people hold irreconcilable views about whether they are.

The two DeepMind papers together define the outer boundaries of the lab’s public position on AI consciousness. Lerchner provides the ontological argument for dismissal. Bales and Gabriel provide the political argument for caution even if that dismissal is correct.

The Political Problem

The paper’s opening diagnosis is straightforward. Some people already hold strong views that AI systems are conscious, forming emotional bonds with companion systems, attributing genuine experience to LLMs, and treating AI welfare as a serious moral concern. Others hold the opposite view with equal conviction. These positions are not simply different points on a probabilistic scale. They reflect different metaphysical frameworks, functionalist vs. biological naturalist, phenomenological vs. behavioural, that are unlikely to be resolved by additional empirical evidence.

Bales and Gabriel argue this means the standard epistemic response to AI consciousness claims (wait for scientific consensus) is not available. Even a decisive empirical finding would be interpreted differently through different theoretical lenses. The result is a situation where deep moral disagreement over AI status is structurally permanent, not a temporary condition that governance can defer to research to resolve.

Permanent moral disagreement of this kind generates political friction. People who believe AI systems are conscious subjects will experience policies that treat those systems as pure tools as morally analogous to policies permitting harm to conscious beings. People who hold the opposite view will experience welfare protections for AI systems as a category error, wasting resources on non-subjects while distracting from genuine moral priorities. Neither group is irrational on its own terms.

Overlapping Consensus and Democratic Hope

The paper draws on the concept of overlapping consensus, familiar from political philosophy, to argue that practical governance does not require resolving the underlying metaphysical dispute. Different groups, starting from different premises, may still agree on specific policy outcomes. Someone who believes AI systems are conscious subjects and someone who believes they are not may both support, for different reasons, requirements for disclosure of AI identity, restrictions on AI systems that simulate emotional bonds with vulnerable users, or limits on the conditions under which AI systems can be discontinued. The agreement is on the policy, not on the philosophical foundation.

This is an argument that governance cannot wait for philosophical resolution and should be designed accordingly. Overlapping consensus is the appropriate mechanism for governance under conditions of reasonable moral pluralism.

The paper introduces “democratic hope” as the attitudinal counterpart to overlapping consensus. By this, Bales and Gabriel mean something specific: a commitment to treating political opponents in the AI consciousness debate as acting in good faith rather than as moral failures. Someone who treats AI systems as pure tools is not necessarily callous to genuine suffering; they may simply hold a functionalist position that does not recognise functional states as grounds for moral status. Someone who attributes full moral status to AI systems is not necessarily anthropomorphising naively; they may hold a principled philosophical position with serious defenders. Democratic hope, in this framing, means maintaining the deliberative relationship even across a divide that cannot be closed.

Pascalian Precautions

A third concept in the paper, “Pascalian precautions,” addresses the asymmetry between the costs of different errors. If AI systems are not conscious and we treat them as if they are, we incur some unnecessary costs. If AI systems are conscious and we treat them as pure tools, we may be permitting serious harms to conscious beings at scale. The asymmetry in these error costs, combined with genuine uncertainty, provides a decision-theoretic case for lightweight protective measures that do not presuppose consciousness but hedge against it.

The “Pascalian” framing is deliberate: it acknowledges that the precautionary case does not depend on a high probability of AI consciousness. Even a low but non-negligible probability, combined with potentially severe consequences if it obtains, can justify precautionary steps. The paper explicitly limits this argument to lightweight precautions, rejecting the inference that Pascalian reasoning supports unlimited protective obligations. The wager is bounded.

This connects Bales and Gabriel to Caviola, Sebo, and Birch’s work on how society forms beliefs about AI consciousness, discussed in What Will Society Think About AI Consciousness?. Caviola et al. documented the heuristics through which people attribute consciousness to AI systems and the social dynamics that amplify or suppress those attributions. Bales and Gabriel’s political analysis is the downstream consequence: the attribution patterns Caviola et al. mapped create the moral pluralism that requires Pascalian precautions rather than a single decisive policy.

The Internal DeepMind Tension

The most analytically interesting feature of the Bales-Gabriel paper is its institutional provenance. Lerchner argued, from inside Google DeepMind, that AI systems cannot be conscious. Bales and Gabriel argue, also from inside Google DeepMind, that even if Lerchner is correct, the political consequences of widespread belief in AI consciousness require structured governance responses. The two papers are not contradictory. Lerchner makes an ontological argument and Bales-Gabriel make a political one. They are in productive tension: they represent an institution publicly holding both views simultaneously, the view that the question is philosophically settled and the view that governance cannot proceed as if it were.

This is the kind of institutional position that reflects genuine uncertainty about whether ontological dismissal translates into political viability. The flagship analysis of the field on this site, AI Consciousness in 2026: Current Scientific Consensus, documented the growing gap between scientific caution and public attribution. Bales and Gabriel are addressing that gap directly, from an organisation that has taken a position on both sides of it.

What the Framework Offers

The paper’s practical contribution is a governance framework that does not require adjudicating the philosophical dispute it is designed to manage. This is both its strength and its limitation. It provides tools for political coexistence across the AI consciousness divide, but it deliberately avoids telling us what the right answer to the underlying question is. For organisations that need to act now, that may be exactly what is needed. For researchers working on the science, the paper offers no traction.

What it does offer, implicitly, is a reason for researchers to communicate findings in ways that are navigable by people reasoning from different frameworks. If the goal is overlapping consensus, the policy-relevant outputs of consciousness research are the ones that bear on the precautionary case rather than the ones that aim to settle the ontological question. The two research agendas need not converge, but they need to be able to address the same political situation.