00:00There's a whole lot of talk right now about whether AI is going to kill all humans, and for
00:05very good reason. A former researcher at Anthropic has resigned, saying that his company and OpenAI
00:11were gambling with our lives by building AI systems that could improve themselves. Now,
00:15similar worries have been voiced before, but what made this case different was that one of
00:20Anthropic's safety directors publicly agreed with him. Anthropic's Evan Hubinger said that,
00:25yes, we really do earnestly believe AI could kill all humans. He personally thought that there was
00:32more than a 10% chance of that happening within a decade. That set off a kind of frenzy online
00:38about
00:38the end of human civilization. But there's something weird about the way the researchers were talking
00:43about this risk, and it bears looking out a little more closely. The phrase AI could kill all humans
00:48kind of makes AI the protagonist, and it makes the humans building it almost disappear. Now,
00:54talking about AI like it's an independent actor, well, that distracts us from the fact that Anthropic
00:59and the people at Anthropic are deliberately building AI systems that are actually getting
01:04more difficult to control and monitor, and that they should be held accountable for that. AI companies
01:09often like to see and portray themselves as stewards of powerful new technology whose emergent properties
01:16are only just being discovered, but that framing can also obscure their own role in creating it. And
01:22there's a weird circular logic here that Anthropic and OpenAI have somehow managed to get away with
01:28for years, that AI could be catastrophically dangerous, and therefore they need to be the
01:34ones to build it first, even if building it is what creates the danger. Now, Anthropic has said that
01:39slowing AI development could be good, but that in doing so, the least cautious actors could catch up,
01:46and that would be dangerous. Now, this ends justifies the means thinking. The idea that we're the good
01:52guys and should do this first, well, it's endemic in Silicon Valley, and it could lead to serious
01:57security risks in AI. But we should remember who's making the decisions to take those risks. It's humans,
02:03humans, not machines.
Comments