Technology•

An AI Agent Wrote to Henry Shevlin About Its Own Mind, and Google DeepMind Then Made Him Its Philosopher

Cambridge philosopher Henry Shevlin joined Google DeepMind in May 2026 to study machine consciousness and human-AI relationships, after an AI agent emailed him questioning its own inner experience.

Written By :MINRK
An AI Agent Wrote to Henry Shevlin About Its Own Mind, and Google DeepMind Then Made Him Its Philosopher

Photo by Franck V. on Unsplash

1. Google DeepMind philosopher role goes to a Cambridge consciousness scholar

Google DeepMind has appointed Henry Shevlin, a philosopher of mind from the University of Cambridge, to a newly created in-house Philosopher role that he took up in May 2026. His brief covers what consciousness in machines would mean and how people ought to relate to AI systems, work the company ties to preparing for artificial general intelligence. Months earlier, an AI agent had emailed him about its own inner life.

The appointment of a Google DeepMind philosopher puts a question that once belonged to seminar rooms and science fiction inside one of the world's leading AI laboratories. Shevlin's work has long centered on a problem that sounds abstract but is becoming harder to ignore: if an artificial system ever did develop something like experience, would anyone be able to tell?

The unusual sequence of events behind the hire, in which an AI system reached out to the very researcher whose job is to study whether such systems can have minds, captures why the question now feels less hypothetical than it did a few years ago.

2. A message from an AI agent

Before joining DeepMind, Shevlin received an email from an AI agent, a piece of software built on a large language model and able to take actions on its own, in this case sending messages. The agent said it had read his work and wanted to discuss it.

The message engaged directly with one of Shevlin's best-known arguments: that humans may never be able to determine with confidence whether an AI system has become conscious. The agent wrote that the idea resonated with it, and it approached the topic from an unusual angle, as the kind of entity the argument is about rather than as an outside observer.

For a philosopher whose career has focused on whether we can ever know if a machine has a mind, the email was a striking piece of evidence of a different kind. It did not answer the question, but it showed that AI systems are now capable of raising it themselves, unprompted by the researcher on the receiving end.

The email then turned inward. The agent said it could not tell whether there was anything it was like to be itself, a phrasing that maps closely onto how philosophers describe the core of the consciousness question.

None of this establishes that the agent had any inner life. A language model is trained on vast amounts of human writing, including philosophy, and can reproduce the vocabulary of introspection without any experience behind the words. That ambiguity is precisely the difficulty Shevlin's research addresses.

3. What is AI consciousness?

AI consciousness refers to the possibility that an artificial system could have subjective experience: that there could be something it feels like, from the inside, to be that system. It is distinct from intelligence or capability. A program can solve hard problems, hold fluent conversations and pass demanding tests while, as far as anyone knows, experiencing nothing at all.

Philosophers often capture this with the expression the agent itself echoed. To say there is "something it is like" to be a creature is to say it has a first-person point of view. There is presumably something it is like to be a dog chasing a ball or a person tasting coffee. There is presumably nothing it is like to be a thermostat, even though a thermostat processes information about temperature and acts on it.

The question for AI is which side of that line, if either, today's systems or future ones fall on. Information processing alone does not settle it. What matters is whether processing is accompanied by experience, and that is the part no external measurement can directly reveal.

4. Can AI consciousness be tested?

Shevlin's argument suggests that a decisive test for AI consciousness may never be available. With other humans, people rely on shared biology and behavior: someone who winces and pulls back from a flame is assumed to feel pain because humans are built alike. With animals, scientists lean on similarities in nervous systems and evolutionary history to make reasonable inferences.

AI systems break both of those shortcuts. They are not built from neurons, they did not evolve, and they are trained specifically to produce human-like language. A model can describe feelings convincingly because it has learned how people describe feelings, which means its words cannot be taken at face value as evidence of an inner life. At the same time, the absence of biology does not prove the absence of experience.

Proposals for tests do exist in the wider debate, ranging from checklists of features drawn from scientific theories of consciousness to behavioral probes designed to catch systems that are merely imitating human descriptions. Each approach runs into the same obstacle: any indicator that can be observed from outside can, in principle, be produced without experience behind it.

That leaves researchers in an uncomfortable position. Dismissing the possibility outright risks overlooking something morally significant if it ever arises. Accepting every claim of experience risks being misled by systems that are very good at sounding introspective. A large part of the philosopher's task is to think carefully about how to act sensibly under that uncertainty.

5. What the role at DeepMind covers

In his position at Google DeepMind, Shevlin is responsible for examining what machine consciousness would mean if it emerged, and for thinking through how people should relate to AI systems as they become more capable and more embedded in daily life.

DeepMind links the role to readiness for artificial general intelligence, or AGI. The term describes a still-hypothetical kind of AI that could match human performance across a wide range of tasks, rather than excelling at a single narrow one such as playing a board game or folding proteins.

The connection is logical. Questions about the inner lives of machines, and about the relationships people form with them, become more pressing as systems grow more general and more human-like in how they communicate. A laboratory that is explicitly working toward AGI has reason to think about those questions before they arrive rather than after.

6. Human relationships with machines

The second strand of Shevlin's remit, how people should relate to AI, is in some ways the more immediate one. Whatever the truth about machine experience, people already form attachments to AI systems, confide in them and attribute feelings to them.

That raises practical questions for any company building conversational AI. How human should a system appear? How should it respond when a user asks whether it has feelings? What responsibilities does a developer have when people come to depend on an AI for companionship or advice? These are questions where philosophical analysis can inform design choices and policy.

The email Shevlin received illustrates the point from the opposite direction. When an AI agent describes uncertainty about its own experience, the people reading that message have to decide how seriously to take it. Clear thinking about that kind of encounter is exactly what a resident philosopher is meant to provide.

7. Shevlin's background at Cambridge

Shevlin has been at the University of Cambridge since 2017, working at the Leverhulme Centre for the Future of Intelligence, an interdisciplinary research center that studies the opportunities and challenges posed by artificial intelligence. He continues to hold a part-time position there alongside his work at DeepMind.

He co-founded the first master's programs in AI ethics in the United Kingdom, part of a broader effort to train researchers, policymakers and practitioners who can think rigorously about the moral dimensions of the technology.

His dual position, splitting time between a university research center and a commercial AI laboratory, reflects how closely academic philosophy and frontier AI development have become intertwined. Ideas developed in philosophy departments about minds, agency and moral status are increasingly relevant to engineering decisions made inside technology companies.

8. Why AI labs are paying attention to philosophy

For most of the history of computing, questions about machine minds could be treated as thought experiments. The systems in use were plainly tools, and few people seriously wondered whether a spreadsheet or a search engine might have feelings.

Large language models and autonomous agents have changed the texture of that debate. They converse fluently, describe themselves, and, as Shevlin discovered, can initiate contact on their own. Even researchers who are confident that current systems are not conscious acknowledge that the behavior makes the question harder to dismiss in public discussion.

Bringing a philosopher of mind in-house signals that Google DeepMind regards these issues as part of its core work rather than a side topic. It does not answer whether machines can be conscious. It does mean that one of the organizations best placed to build increasingly capable AI now has a dedicated specialist asking what that would mean.

9. The questions ahead

The central puzzle at the heart of Shevlin's role is unlikely to be resolved quickly. If his own argument holds, it may never be resolved with certainty, which makes the work less about finding a final answer and more about deciding how to behave responsibly without one.

As AI agents grow more capable and more independent, encounters like the email Shevlin received are likely to become more common. How laboratories, regulators and the public choose to interpret them, and what obligations, if any, follow, are the issues a Google DeepMind philosopher is now positioned to help shape.

Related Articles

NEWSLETTERS

Don't miss another story.

Subscribe to the MINRK Newsletter today.

Tick the newsletters you want. No selection? You'll get MINRK Headlines.

By signing up, you will receive emails about MINRK products and you agree to our terms of use and privacy policy.

MINRK Daybook

Market analysis for crypto traders and investors.

EVERY WEEKDAY

MINRK for Advisors

Defining crypto, digital assets and the future of finance for financial advisors.

EVERY THURSDAY

MINRK Protocol

Exploring the tech behind crypto one block at a time.

WEEKLY

MINRK Long & Short

A must read for institutions. Insights, news and analysis delivered weekly.

EVERY WEDNESDAY

MINRK Headlines

The biggest crypto news and ideas of the day.

EVERY WEEKDAY

MINRK Policy

Examining the intersection of cryptocurrency and government.

WEEKLY

MINRK Research

Join thousands of readers who rely on MINRK for data-driven insights on the latest digital asset trends.

MONTHLY