Below, we share some resources and talking points to help address common questions and points of confusion in the AI discourse.
How exactly would AI kill everyone, and how would it do so from inside a computer?
Resources:
- A LessWrong article describes several ways AI could extinguish humanity and explains how it could act from inside a computer (see “Misconception A”).
- A video from 80,000 Hours dramatizes the fictional AI takeover scenario from Part II of If Anyone Builds It, Everyone Dies.
- Chapter 6 of If Anyone Builds It, Everyone Dies and the associated online resources cover this and related questions.
You might also point out that:
- AIs out-persuade world champion debaters.
- Society is a physical world increasingly built on top of a digital one (hospitals, water systems, weapons, etc.).
- Actions taken through computers can have physical consequences; for example, the wrong phone call can launch nuclear weapons.
- AIs can impersonate real people on video calls – see this striking example from 2024, when a finance worker wired $25 million to scammers who convincingly deepfaked a Zoom meeting with his co-workers.
Why can’t we unplug it?
An online resource explains how a smart AI may hide copies of itself that would survive even if the original were shut down. It takes far less compute power to run AIs than to train them, making such escapes difficult to trace.
Is this all just hype/regulatory capture/marketing?
Resources:
- A short-form video from a member of MIRI’s comms team explains why this isn’t hype.
- An article by Benjamin Todd demonstrates the severity of the Hugging Face attacks, pushing back on misunderstandings that underlie the “just hype” take.
- A post by economist Alex Tabarrok explains why this probably isn’t regulatory capture.
- An X post by Daniel Kokotajlo points out that a true slowdown would not be regulatory capture.
You might also point out that:
- OpenAI is delaying its IPO.
- OpenAI didn’t report, or didn’t publicize, many of the other swarm incidents, and only reported the Hugging Face incident after the victim alerted the FBI.
- Credible people outside the labs raised the alarm long before there were products to sell.
- CEOs may be saying things to appease or retain genuinely concerned employees.
- The labs drafting their own regulations may be an attempt at regulatory capture; see OpenAI’s “reverse federalism” strategy. The solution isn’t no regulation; it’s regulation whose content is not decided by AI companies.
What about China?
Resources:
- A podcast episode with Ezra Klein and Matt Sheehan discusses this question.
- A MIRI memo documents promising signals China has sent on global coordination.
You might push back on the race narrative by pointing out that:
- A race to “beat” China to superintelligence is a race with no winners.
- Chinese academics also worry publicly about extinction by AI.
- As covered in the podcast episode above, China isn’t necessarily very interested in superintelligence.
- Model distillation means that US progress fuels Chinese progress.
- An American AI slowdown, while insufficient by itself, would be a meaningful signal to China that America is serious.
- Stricter export controls could prevent China from “catching up”, even during a halt or slowdown.
Why are the labs developing advanced AI if they think it can kill us? Why aren’t more employees quitting?
Resources:
- An extended discussion from the IABIED online resources attempts to make sense of the death race.
- The 2026 documentary The AI Doc offers a more charitable interpretation: AI companies believe nobody will stop until everybody stops. In the meantime, if they think they are slightly more responsible than the next company, they’ll keep going.
- A website documents the many employees who have quit, citing safety concerns.
You might also point out that:
- Some employees think they have a better shot at influencing broken lab culture from within, even if this reasoning may be unwise.
- Some employees may expect to be criticized no matter what they do. To quote Steven Adler: “Anyone who thinks it’s >10% and quits is giving up on averting a catastrophe; anyone who thinks it’s >10% and is staying is a psychopath.”
- There are selection effects at play. Aiming to create, and personally shepherd, a potentially civilization-ending technology implies extraordinary confidence in one’s own judgment. Many who would elect not to gamble with all human lives on such a project would not start or join AI companies in the first place.