We’re playing catch-up with AI
By the time we have a good explanation, the AI has already moved on.
Most of us interact with AI as a helpful assistant. But interpretability research, which looks at how the digital brains behind AI work, suggests the assistant is a character the model has learned to play: a persona.
Now we’re giving AI access to other systems, including the ability to create other versions of itself. These agents appear, do some work, leave information behind and disappear, and other agents pick up from there.
As AI talks to AI, the assistant persona tells us less about what’s actually happening. We’re reaching for new stories to explain it, like coordination, deception and even conspiracy.
AI does something, we look at what happened and try to understand it, and by the time we have a good explanation, the AI has already moved on. We’re playing catch-up.
Watch on YouTube