Nobody needs AI to search the Internet, court says in ruling against Google

[Bluetus]

39. That's an astute judge. But it goes beyond that.

I'm very impressed with the clarity of this ruling. Most judges don't have the world knowledge to figure these things out. However, I would like to add an element that is floating out there.

Most people don't want to say it out loud, but AI is not controllable, period, full stop. In computer science terms, AI is not deterministic, which is to say that it is impossible to know exactly how these very large models will react. And perhaps they share that with organic brains. We can't be sure a dog that looks docile will not bite under some set of circumstances.

For the past 25 years, we have talked about "algorithms." An algorithm is essentially a program: If this, then do that. Algorithms can be very complex, but they are deterministic. It is always possible to find the point in the algorithm where things go bad. There is no "program" with AI. The neural networks are "trained", and with proper training, they hopefully develop tendencies that are more good than evil. But we can see vast differences between Grok (which was largely trained on the Twitter cesspool) and Claude, for example. Grok tends toward the evil, antisocial side, where Claude may be more humane and altruistic by nature.

But the main point is that, when there is a bad result, you can't "fix the algorithm". All you can do is try to throw more training data that reduces the odds of that bad behavior.