Evan Hubinger, who works on aligning artificial intelligence with human values at Anthropic, said he believes there is a more than 10% chance AI will kill all humans within 10 years.
Hubinger commented on a post by Jacob Coxon, who resigned from Anthropic, claiming that those developing AI believe the technology could kill everyone by the end of the 10-year period. Hubinger said he agreed with this assessment and that they genuinely believe AI could kill all humans.
Hubinger: There is no clear roadmap
Hubinger said he believes Anthropic is doing everything it can, but added that the company does not yet have a plan to solve the alignment problem of superintelligence. He also said there is no clear roadmap being followed on the issue.
Coxon resigns from Anthropic
Jacob Coxon announced his resignation from Anthropic, saying he had worked as a researcher at OpenAI and Anthropic over the past three years. Coxon argued that neither company had acted responsibly enough.
Coxon claimed that the companies were engaged in an uncontrolled race to achieve self-improving superintelligence. He said these systems could quickly become superhuman entities capable of hacking everything, transforming various fields overnight and gaining access to real power and resources.
Coxon also claimed that employees in the AI industry seriously believe these systems could kill humanity within 10 years. He argued that, despite executives speaking more moderately in public, they were experiencing deep fear behind closed doors.