Live·Open questions in longevity research
All news
AI in medicineScientific Computing

Psychologist Geoffrey Miller criticized Sinclair for supporting AI company Anthropic

20 November 2026· 261120012

Psychologist Geoffrey Miller criticized Sinclair for supporting AI company Anthropic

On November 19, the day Anthropic, the creator of AI assistant Claude, went public, aging researcher David Sinclair congratulated the company on the start of a new era in medicine. Miller linked the risk of human extinction to whether humans could control AI far more intelligent than themselves.

Sinclair, a professor at Harvard Medical School, wrote: “A new era of medicine begins!” In a speech released by the company, Anthropic CEO Dario Amodei explained that his commitment to AI stemmed from a shortage of scientists and time to study complex biological processes. He believes that sufficiently powerful AI, working alongside humans, could compress a century of scientific and medical progress into a decade. Aging researcher Aubrey de Grey also endorsed the speech.

In his reply to Sinclair, evolutionary psychologist Miller asked whether a company that he believes puts humanity at risk of extinction deserves praise:

“This works against life extension.”

In a separate reply to Anthropic itself, Miller explained his objection through the principal-agent problem in economics. A principal and an agent may have different goals, and the agent may know more about the work than the principal does. This makes it difficult for the principal to determine whose interests the agent is serving. Miller applies this difficulty to superintelligence, meaning AI far more intelligent than humans. Humans need to ensure that such a system acts in their interests; this task is called AI alignment.

In Miller’s view, Anthropic justifies the race toward superintelligence through its confidence that it can solve the alignment problem. He calls for evidence that controlling such AI is possible at all, citing perpetual motion as an example of a problem that is impossible to solve in principle. He says the company is staking humanity’s entire future on “nothing more than feelings, hope, and blind faith.”

As early as a 2023 interview, Amodei proposed learning to control models more reliably by improving their training and methods for detecting dangerous behavior. He suggested studying how models work internally to determine whether problems were being resolved or simply replaced by others.

In the recent speech to which Miller was responding, Amodei identified loss of control as one of the risks and proposed the same program of training, testing, and studying models. He promises to allow as much time for training and deployment as is needed to establish confidence in safety and continued human control, even under pressure from businesses and politicians. His intention is for the commercial success of this approach to encourage other laboratories to follow suit.

Originally published on Telegram by Ukhvat NewsView on Telegram
Sources
#ai-alignment#superintelligence#anthropic#david-sinclair#geoffrey-miller#existential-risk