Harvard psychologist Steven Pinker recently argued that existential fears of artificial intelligence wiping out humanity are overblown. Responding to a public debate challenge from influential tech blogger Scott Alexander, Pinker declined the event and instead published an open letter on Quillette detailing his perspective on AI risk. He asserted that catastrophic doomsday scenarios, such as the famous paperclip maximizer thought experiment, incorrectly confuse general intelligence with a biological drive to dominate and incorrectly assume AI systems would fixate maniacally on a single goal.
Although Pinker previously expressed optimism about artificial intelligence, he now admits that he underestimated the rapid capability gains of large language models and the recklessness with which AI companies commercialize their products. Despite dismissing science fiction extinction scenarios, Pinker emphasized that he does take tangible near-term dangers seriously, including bioterrorism, malicious cyberattacks, and unchecked autonomous AI agents. He contends that speculative doomsday rhetoric does nothing to mitigate these actual threats.
To address legitimate technological risks, Pinker advocates for sober safety engineering rooted in independent oversight, strict legal liability, and robust human control mechanisms. Meanwhile, Alexander maintains a much higher risk assessment, placing the probability of an AI-induced human extinction at 20 percent while calling for enhanced safety research and international treaties to slow development. This debate highlights ongoing philosophical and strategic disagreements within the technical community over how to properly prioritize and manage artificial intelligence safety.
Why it matters: Shifting the AI safety debate away from specu