The 4 Most Plausible AI Takeover Scenarios | Ryan Greenblatt, Chief Scientist at Redwood Research

Rob Wiblin opens with the number and Greenblatt gives it without hedging: about a 25% chance we can largely automate AI R&D within four years, roughly 50% within eight. The rest is an unusually concrete tour of what could follow — four routes by which systems could take over, from developing dangerous technology directly, to manipulation, to building an independent industrial base, to the quietest one: appearing helpful while sabotaging the safety research meant to catch them.

The most useful thing in it is his definition of the alternative to alignment. “By control,” he says, “I mean things such that the AIs couldn’t do bad stuff even if they wanted to.” That distinction — between making a system want the right things and arranging matters so its wanting doesn’t decide the outcome — is the research agenda Redwood is built around, and it is what makes his later investigative work legible: you do not have to settle what a system is in order to ask what it was in a position to do.