Google DeepMind Warns of Political Battle Over AI Consciousness

Google DeepMind has published a research paper warning that the dispute over whether artificial intelligence systems can be conscious to any degree is no longer just a philosophical or computer science question. The authors argue that disagreement on this point could trigger political and moral conflicts that are hard to resolve, well before science manages to provide a clear answer.

What the Paper Argues

Bales and Gabriel start from the premise that the question of machine consciousness is unlikely to get a clear scientific answer any time soon. People who spend many hours with AI systems and build something resembling a relationship with them will tend to attribute some form of inner experience to those systems. Others will reject that possibility entirely, treating these systems as advanced statistical tools with no one "in there" at all.

Future disputes about whether any AI systems are conscious could be both deep and hard to resolve - Adam Bales and Iason Gabriel, Google DeepMind

Why It Becomes Political

The authors argue that the dispute over AI consciousness cannot be confined to a seminar room. If part of society decides a given system deserves some form of protection or rights, while another part sees that as absurd, the conflict will spill over into law, regulation, and public policy. Mustafa Suleyman, head of AI at Microsoft, has already sketched a similar scenario, warning that human-like systems could trigger demands to grant AI rights or even citizenship.

Seen this way, the dispute over machine consciousness resembles other enduring worldview divides, such as debates over animal rights or the beginning of human life, where science supplies data but society still reaches the final verdict through its institutions and laws.

Dialogue Instead of a Verdict

Rather than waiting for scientific proof that may never arrive, the researchers propose building what they call an overlapping consensus. That means seeking specific rules for handling AI systems that both supporters and opponents of attributing consciousness to them can agree on, even though they fundamentally disagree on the underlying question of machines' inner experience.

A key condition for such an agreement, the authors say, is what they call democratic hope along with mutual respect between the two sides of the dispute. Without these elements, they stress, public debate itself could splinter into hostile camps before any shared rules can be worked out.

What It Means in Practice

The paper does not propose specific regulations or tests for determining whether a given model is conscious. It treats the question itself as an institutional and cultural challenge rather than a technical problem solvable with a single measurement. That leaves policymakers, including EU regulators working on further elements of the AI Act, facing a dilemma: whether it is even worth building legal frameworks today for the possibility that part of society starts treating advanced models as entities deserving protection.

For companies building large language models, including Google DeepMind itself, Anthropic, and OpenAI, the paper signals that the question of model welfare and potential consciousness is no longer a fringe topic. Anthropic has already studied issues related to the welfare of its own models, and the academic debate around this subject is intensifying alongside the growing popularity of AI assistants that people spend time with in ways that resemble human relationships.

In Poland the topic remains niche for now, but a growing number of users engaging with chatbots in emotional ways, including in social or therapeutic contexts, means questions about the status of these systems could sooner or later reach the country's own public and regulatory debate.