TIME is reporting that Geoffrey Irving, the former chief scientist of the U.K. AI Security Institute, has warned there is roughly a 50% chance that superintelligent artificial intelligence will kill all humans. In a Weekend Essay, Irving said he is not claiming precision with that estimate but is highlighting that several of the most fundamental debates about the future and safety of AI remain unresolved. He said he believes those disagreements will not be settled until it is too late to change course, and that action must be taken despite the uncertainty.
Irving said a superintelligent AI system would need just four types of skills to kill all humans. It said the first is hacking, which would allow a system to take over and migrate between computer systems. The second is persuasion, which would let it manipulate humans into acting on its behalf. The third is the ability to conceal its thoughts from humans, which he said is increasingly common in the latest and most capable models. The fourth is planning and coordination between agents to multiply their effectiveness.
Irving said those skills are very close to the capabilities that AI companies intentionally train for. He said AI's capacities and motivations, along with human actions, all feed into the likelihood of human survival, and he argued the time to pause AI development is now.
Separately, TIME is reporting that President Donald Trump is appearing in taxpayer-funded ads airing ahead of the November midterms. It said the ads have drawn criticism from lawmakers in both parties over whether the government is crossing a legal line. Federal laws prohibit taxpayer dollars from being used for publicity or propaganda, but the government is also allowed to promote and defend its policies. TIME said the White House describes the ads as public service announcements and points to similar campaigns under previous administrations.