I saw this going viral: “An Open Letter to Scott Alexander,” by Steven Pinker
He writes:
I stand by my claim that most of the doomer scenarios are preposterous (beginning with the AI paperclip maximiser that consumes our bodies as raw materials). But as rationalists we should agree that attacking an idea should not be confused with attacking a person.
I broadly agree with Pinker, even if he does not adequately address the latest literature. He’s still focused on the paperclip maximizer hypothetical, but AI risk has progressed a lot since then. But I agree the “doomers” or “safetyists” have not articulated their case well. They have not adequately explained the steps or conditions that have to transpire for AI “break containment”, and have also dismissed the obvious objection, instead invoking hypotheticals or dodging the question.
The argument goes, “A sufficiently advanced intelligence cannot be predicted and will always be ahead of humans, anticipating and adapting accordingly.” But they cannot explain what it takes for AI to even get to that level. AI becomes anthropomorphized as a movie villain, without any explanation of why or how this is allowed to happen.
Astrophysicists know that the Sun will become a red giant in about 5 billion years, based on the underlying physics. This is not mere speculation. They can explain the series of events that will cause this, and despite having not yet witnessed the Sun dying, physicists still make a convincing case.
By comparison, “doomers” fail to articulate how AI “breaks free,” with anything remotely approaching that level of rigor. It’s just hand-waving and “trust us, bro.” Of course, AI is not astrophysics, but I do expect them to address questions such as, “How does AI transition from merely existing on servers as chatbots or LLMs to something that actually interfaces with tangible reality? Or are we talking about an entirely different technology?” And why would emergency shutdown measures fail?
Also, there is goalpost moving, such as forecasting catastrophic job losses and other disruption due to AI, in contrast to the end of humanity. These are obviously different levels of severity, but I have seen the same people argue both positions, effectively casting a wider net so they can say, “We told you so.”
But with the US the unemployment rate at historic lows, at just 4.1% as of August 2026, if anything the opposite has happened. Inflation is a bigger concern now than AI leading to deflation. If AI leads to productivity gains and a stronger economy, then we could expect an uptick in long-term inflation (e.g. 10-20 year treasury bonds yields), in anticipation of this. That is what is happening now, as bond yields rise to levels last seen in the mid 2000s, although there are other reasons for this, such as rising oil prices.
The best way to adjudicate this debate is to have both sides submit articles explaining their positions for independent peer review by a panel of experts. The panel could be questioned about potential biases, similar to jury selection, to help ensure fairness. Each side would have opportunities for rebuttals, which would also be reviewed, followed by closing statements. The entire exchange would be in print and reviewed, with a “winner” chosen. Because everything would be written, there would be no opportunity for audience interference, as commonly seen during live debates.