What Is the Red Queen Race in AI?

Kevin Roose calls the AI race a Red Queen race, a term from evolutionary biology for a survival-of-the-fittest contest among rivals. No matter how fast a lab goes, someone else is always going faster, so the labs race just to stay in the same place. He left less reassured than at his starting point that the technology can be developed safely.

He went in thinking there were grownups in the room and they seemed trustworthy. He liked that AI leaders seemed to take safety and the harms of their products very seriously, at least next to social media, which he grew up covering. Mark Zuckerberg was not going around during the early days of Facebook saying the product could kill millions of people. Facebook was just promoting the benefits. AI leaders said from the get-go that they were fearful of the technology and thought it could have enormous harms. Part of him trusted that instinct, or at least he was glad they were thinking about the downsides so early. Then the pattern locked in: OpenAI releases a better model, so Anthropic has to release a better model, so Google has to release a better model. The labs never quite get where they are going.

The early talk of harm did not slow the contest. Fear of the technology and a race to hold position now sit side by side, and the second one is what left him less reassured.