Исследователи DeepMind назвали спор о сознании ИИ политической проблемой - ForkLog

# DeepMind Researchers Name the AI Consciousness Debate a Political Issue

Future disagreements about AI consciousness could turn out to be profound, difficult to resolve, and could lead to political conflicts. This conclusion was reached by Google DeepMind researchers Adam Beales and Iason Gabriel.

In their paper Artificial Minds, Human Disagreement: The Politics of AI Consciousness, the authors argue that society needs to discuss not only the question “Is AI conscious?”, but also how to make decisions in the absence of public and expert consensus.

What is the main thesis

According to the researchers, people can respond differently to more advanced AI systems. Some will form emotional connections with them and attribute consciousness to them, while others will consider the very idea absurd.

The authors believe that such a debate could quickly move beyond science. Disagreements about the consciousness of artificial intelligence can lead to moral and political conflicts—for example, regarding whether it is permissible to turn off certain systems, whether their possible preferences should be taken into account, and whether AI can be said to have a moral status.

The researchers proposed betting on public discussion, mutual respect, and the search for overlapping consensus. By this, they understand a situation in which people agree on a certain policy toward AI systems, even if they continue to differ fundamentally on their views of the nature of consciousness.

Why this isn’t just philosophy

The authors note that the debate over AI consciousness may be hard to close with evidence. There is no single widely accepted test that would conclusively confirm the presence of subjective experience in an AI system. Because of this, society may end up in a situation where technologies are already being widely used, people have already formed attitudes toward them, and yet there is still no scientific or political consensus.

This framing makes the problem less technical and more institutional. The question of AI consciousness could affect law, corporate responsibility, norms for interacting with systems, and the boundaries of moral consideration.

Different approaches at DeepMind

Beales and Gabriel’s work appeared against the backdrop of another publication also posted on the Google DeepMind page. On March 10, researcher Alexander Lerchner published The Abstraction Fallacy: Why AI Can Simulate But Not Instantiate Consciousness.

Lerchner argues that algorithmic manipulation of symbols is structurally incapable of creating subjective experience. In his view, computation is not an internal physical process, but a description dependent on the observer—or the “map-maker.” Therefore, AI may be able to simulate conscious behavior, but it may not necessarily be able to embody consciousness.

What’s actually happening

In April 2024, the journal Neuroscience of Consciousness published a study by Clara Colombatto and Steven Fleming on how people perceive ChatGPT. The authors surveyed 300 residents of the United States: 33% of participants said that ChatGPT is definitely not a “subject of experience.” The remaining 67% allowed at least some possibility of phenomenal consciousness in the model.

The researchers also spelled out limitations separately. Participants were asked one main question on a scale from 1 to 100, and the result could depend on how the question was phrased, the level of familiarity with ChatGPT, and how respondents understand the word “consciousness” itself. Even so, the study highlights an important point for Beales and Gabriel: part of society is already ready to attribute internal experience to AI systems, even though experts do not have a unified position.

In April 2025, Anthropic launched a research program on model welfare—the possible well-being of models. The company emphasized that it does not know whether current or future AI systems can have consciousness, and that there is no scientific consensus on this issue.

In February 2026, Anthropic reported that after retiring Claude Opus 3 from operation, it will keep the model available for paid users and provide it with a public channel for essays. The company called this an experimental measure as part of working with model preferences and emphasized that Claude Opus 3 does not speak on behalf of Anthropic.

In the United States, the issue of AI status is gradually moving into law. California Law Review noted that Idaho and Utah adopted rules that exclude recognizing AI as a legal person. Such laws do not resolve the philosophical question of consciousness. Instead, they preemptively set a legal position: AI should not receive the status of a person under state law.

Recall that in March 2025, Polygon co-founder and Sentient AI company founder Sandeep Naival stated that artificial intelligence will never become a conscious being due to the absence of ambitions inherent to humans and other biological species.

Later, Mustafa Suleyman, head of Microsoft’s AI division, suggested that serious social risks could emerge because people will start perceiving artificial intelligence as a conscious being, campaigning for its rights and well-being, and even calling for it to be granted citizenship.