OpenAI’s Existential AI Risk Team Disbanded, Leaving Future Uncertain
The Disbandment of OpenAI’s Superalignment Team
The Superalignment team at OpenAI, responsible for controlling the existential risk of a superhuman AI system, has reportedly been disbanded, according to the article. The news comes just days after the team’s founders, Ilya Sutskever and Jan Leike, left the team.
Wired reports that the Superalignment team at OpenAI, first launched in July 2023 to prevent future superhuman AI systems from becoming dishonest, is no more. The report states that the group’s work will be absorbed into other research efforts at OpenAI. Research on the risks associated with more powerful AI models will now be led by OpenAI co-founder John Schulman, according to Wired. Sutskever and Leike were some of the leading scientists at OpenAI focused on AI risks.
Leike’s Departure and Concerns
Leike’s Friday post vaguely explains why he left OpenAI. He says he has been arguing with OpenAI leadership about core values for some time, but reached a breaking point this week. Leike noted that the Superalignment team has been “sailing against the wind,” struggling to secure enough computing power for crucial research. He believes OpenAI needs to be more focused on safety, security, and alignment.
Uncertain Future
Later, an OpenAI spokesperson clarified that “superalignment will now be more deeply rooted throughout research, which will help us better achieve our Superalignment Goals.” The company says this integration began “weeks ago” and will ultimately move Superalignment team members and projects to other parts of the company.
Currently, we do not have a solution to direct or control a potentially superintelligent AI and prevent it from becoming dishonest,” said the Superalignment Team in a blog post from OpenAI when. “But humans will not be able to reliably supervise AI systems that are much smarter than us, so our current alignment techniques will not scale to superintelligence. We need new scientific and technical breakthroughs.”
