Government and academic assessments, including reports from the U.K. Government Office for Science and the RAND Corporation, have not ruled out the possibility that AI could lead to human extinction or societal collapse. To protect against such severe risks, we must first understand how those risks could unfold.
However, the scenarios describing existential risks from AI are scattered across journals, books, forum posts and reports, and few studies have collected them systematically. To address this, researchers at the Existential Risk Observatory, the Future of Life Institute and MIT FutureTech (affiliated with the MIT Sloan School of Management and the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL)) conducted a systematic literature review with colleagues from 15 other institutions, including the University of Queensland, Mila—Quebec AI Institute and the Oxford Martin School (University of Oxford).
After screening 3,184 documents, the researchers found 58 that developed at least one scenario in which AI could threaten humanity's long-term potential, whether through human extinction, an unrecoverable collapse of civilization or a permanently entrenched dystopia. From these, they identified 95 scenarios and grouped them, based on their causal steps, into eight "scenario families."
The findings are available as a paper and free public dataset with an interactive scenario explorer.
To read more, click here.