Harvard psychologist Steven Pinker has publicly addressed concerns regarding existential risks posed by artificial intelligence. In an open letter published on Quillette, Pinker declined a debate invitation from blogger Scott Alexander, characterizing such discussions as a spectator sport rather than a productive path forward.
Pinker argues that current doomsday scenarios, such as the paperclip maximizer theory, incorrectly equate intelligence with a drive for domination. He suggests these models rely on the flawed assumption that AI systems will fixate on a single objective at the expense of all else.
While Pinker acknowledges that he previously underestimated the capabilities of large language models, he remains critical of the reckless pace at which companies release these technologies. He notes the irony that the same community warning of existential threats is often driving this rapid deployment.
The psychologist identifies bioterrorism, cyberattacks, and autonomous AI agents as the genuine risks that require attention. He advocates for sober safety engineering, which he defines as a framework built on independent oversight, clear liability, and robust human control.
Scott Alexander, who initiated the debate request, maintains a different perspective. In a June blog post, he estimated a 20 percent probability that AI could lead to the extinction of humanity, calling for increased safety research and international agreements to regulate development.
