What the AI doom debate misses

In the summer of 2024, some of the people who built the most advanced AI systems publicly warned that their own creations posed a catastrophic risk—one researcher put the chance of human extinction from AI at greater than 10%. But as the Vox podcast Today, Explained notes, it's now fall, and the episode proposes stepping back for a calmer look at what AI might actually do to—and for—humanity. The throughline of the conversation is that the most dramatic doomsday scenarios, while attention-grabbing, may distract from the real, measurable harms already unfolding, as well as from the genuine life-saving potential in fields like medicine.

The two buckets of AI doom

Will Knight, a senior correspondent at Wired who has covered AI and robotics for years, breaks down the existential risk conversation into two main categories. The first is the immediate, plausible danger: bad actors using AI to do terrible things. Knight cites reports of terrorist groups rapidly adopting AI, using jailbroken or open-source models to develop weapons and plan attacks. One former ISIS commander described it as being as easy as asking "how can I build a bomb?" and receiving step-by-step instructions from a chatbot. The terror group had even established an AI unit staffed by senior members who no longer fought but were drawn specifically to work on AI. Beyond terrorism, Knight worries about criminal