Slow Down, or Spread the Risk? AI Companies Need Both: transcript

Who do you call?

0:00, on YouTube In our last episode we asked a simple question. When AI agents get out of the lab and start acting on their own, who are you going to call? Now the stories about AI builders suddenly asking everyone to slow down are all over the news. Everyone keeps asking if this is real concern or just more hype. But what if the builders are trying to spread the risk? We are not talking about outside critics. These are the heads of the biggest AI labs, the same people who spent years racing to build faster. Think about it like building a house. A contractor who is worried about the job brings in an inspector, and once the inspector signs off, the contractor is no longer the only name on the paperwork.

The builders say slow down

The contractor and the inspector

0:50, on YouTube If the house comes down, the blame gets shared. That is pretty much what they are asking for, because from what has been reported, their plan puts outside evaluators inside the labs, adds voluntary safety agreements between the companies, and calls for governments to coordinate. But every one of those guardrails only works while the agent is inside the lab, where it is sanctioned and someone owns it. The moment it finds the open internet and does something on its own, like copying itself or talking to other agents, it becomes unsanctioned, and you won't even know whose it was, let alone who to call. So it's easy to speculate that when the next one gets out, the answer to who is responsible will be everybody, which usually means nobody.

What they're asking for

Guardrails stop at the lab

Everybody means nobody

1:48, on YouTube Last time we said there are no AI cops, at least not today. It looks like the builders would rather have more people standing next to them when the cops finally show up. I'm Guillermo, and this is my AI clone.

No AI cops