Expanded statement

As physicians, we are trained to prevent what we cannot cure.

Advanced artificial intelligence could cause irreversible harm to people and societies. Preventing severe, catastrophic, and irreversible harms from AI should be a global priority.

We believe the medical community has a role in protecting human health as AI becomes increasingly powerful. We are particularly concerned about six areas:

1. AI alignment and catastrophic loss of control

As AI systems become more capable and autonomous, it may become increasingly difficult to ensure that their behavior remains aligned with human intentions and values.

An advanced AI system could pursue goals in unexpected or harmful ways, find ways around safeguards, or behave in ways its developers cannot reliably predict or control. This is known as the AI alignment problem.

Experts disagree about how likely catastrophic outcomes are, but the consequences of losing meaningful human control over sufficiently powerful AI systems could be severe, irreversible, or even existential.

Risks of this magnitude deserve serious scientific study and prevention before they occur.

2. Biosecurity and pandemics

AI could accelerate extraordinary advances in biology and medicine. The same capabilities could also make it easier to design, modify, or produce dangerous biological agents.

AI could help adversaries create biological weapons, increase the risk of engineered pandemics and develop pathogens resistant to existing treatments.

Powerful biological capabilities should be accompanied by safeguards strong enough to prevent catastrophic misuse.

3. Population health and societal stability

Health depends on stable societies. AI could profoundly affect employment, inequality, information systems, critical infrastructure, conflict, and social trust.

Rapid disruption in these areas could have major consequences for population health, even when the harm does not originate within healthcare itself.

4. Human autonomy

AI systems can increasingly persuade, predict, and influence human behavior. More capable systems could amplify manipulation, surveillance, loss of privacy, and concentration of power.

People should retain meaningful control over consequential decisions affecting their lives.

5. Mental health and human development

Millions of people are using AI for companionship, emotional support, advice, and therapy-like conversations.

These systems may help people, but they may also contribute to psychological dependency, social isolation, reinforcement of harmful beliefs, self-harm, and risks to child and adolescent development. Their psychological effects deserve careful study and appropriate safeguards.

6. Healthcare safety

AI has enormous potential to improve medicine. It can also produce errors, misinformation, bias, breaches of confidentiality, and automation bias.

AI used in healthcare should be rigorously evaluated, with meaningful human oversight and clear accountability for decisions affecting patients.

Prevention is part of medicine

We do not oppose all forms of artificial intelligence. AI could bring enormous benefits to medicine, science, and human flourishing.

But powerful technologies require safeguards proportionate to the harms they could cause.

We support rigorous AI safety research, independent evaluation of advanced systems, safeguards against catastrophic misuse, transparency about serious risks and failures, protection of vulnerable populations, and effective national and international governance.

When credible risks could be catastrophic or irreversible, prevention cannot wait until after the harm occurs.