DeepMind Scientist's Paper Claims AI Can Simulate Consciousness but Never Actually Feel It

A researcher affiliated with Google DeepMind has published a paper arguing that artificial intelligence is structurally incapable of ever becoming conscious, no matter how much the technology is scaled up. The paper, titled 'The Abstraction Fallacy: Why AI Can Simulate But Not Instantiate Consciousness,' was written by Alexander Lerchner and has spread widely online in recent weeks.

Much of that online spread came with a misleading framing. A widely shared post on X described the paper as proof that 'Google' itself had ruled out AI consciousness, when the paper was in fact published independently by Lerchner, not as a company position.

Lerchner's Core Argument on Computation and Consciousness

Lerchner's central claim is that computation is not a physical process in its own right, but a description imposed on physical events by an outside observer. As the paper states, symbolic computation 'requires an active, experiencing cognitive agent to alphabetize continuous physics into a finite set of meaningful states.'

From that starting point, Lerchner draws a distinction between simulation and instantiation. He argues that an AI system can behaviourally mimic thought or emotion, but that algorithmic symbol manipulation is structurally incapable of instantiating experience, calling the confusion between the two 'the Abstraction Fallacy.'

In practice, this means the paper does not claim current AI systems are simple or under-trained. It argues that no amount of extra scale or complexity would change the outcome, because the problem is one of kind rather than degree.

A Personal Paper, Not a Company Stance

Lerchner is listed on the paper as affiliated with Google DeepMind. However, the paper itself carries a disclaimer stating that his conclusions do not represent the position of his employer.

Social media summaries have largely lost that distinction, with several presenting the paper as an official Google finding rather than the work of one scientist. Google DeepMind has not issued any separate corporate statement adopting Lerchner's conclusions.

The Numbers Behind the Buzz

The paper was added to the PhilPapers database in March 2026 and has since become one of the platform's most downloaded works, with tens of thousands of downloads recorded within months of publication. At least five separate response papers have since appeared on academic archives, either building on or challenging Lerchner's framework.

Online, the paper found a second life through posts like the one from X account How To Prompt, which summarised the argument for a general audience and helped drive it into wider circulation.

Philosophers Push Back on the 'Abstraction Fallacy'

Not every philosopher working on AI consciousness accepts Lerchner's conclusion. A detailed rebuttal published on Real Morality argues that the Abstraction Fallacy paper rests on a contestable theory of meaning and does not fully rule out the possibility of machine consciousness, even if its core argument is described as one of the more sophisticated cases made against it.

Professor David Chalmers, a leading philosopher of mind at New York University, has previously argued that AI sentience is plausible within the next decade. While Chalmers has not issued a direct response to Lerchner's paper, his broader framework suggests that consciousness may be substrate-independent, opening the door to machine sentience in ways Lerchner's abstraction argument explicitly rules out.

Other academics working in philosophy of mind have also engaged critically with the paper since its release. Several have questioned whether Lerchner's theory of meaning is sufficiently developed to bear the weight of his conclusion.

The debate goes beyond academic philosophy. Questions about whether AI systems can suffer or deserve moral consideration are already shaping how AI companies think about safety, welfare policies and public communication around their products. If Lerchner's argument holds up under scrutiny, it would undercut the basis for treating AI systems as potential moral patients.