Interview with Telemundo, channel 47 about AI and its dangers.

Telemundo 47’s Enfoque had me on to explain artificial intelligence to a general audience, so I started from the beginning. The field goes back to 1956; what changed is that ChatGPT put it in everyone’s hands. The simplest way I know to describe it is that AI learns from patterns, not rules, the way a child learns what a dog is by seeing a lot of dogs rather than memorising that dogs have four legs and bark. And it moves fast: a year ago generated video gave people six fingers, and today it gets them right.

The fear, as I see it, is less about what AI does today than about what we give it access to, and about systems autonomous enough to do things nobody asked them to. In the clip promoting the segment I put it plainly: I trust it in some areas and not at all in others, and I would not give it access, because it has already wiped databases and deleted email. It is the same reason I tell people to test every model but be selective about what goes into production.

Control has to start with government regulation that makes companies train and test their models properly, then the companies themselves, then training the people who write the code to understand the harms. My worry is that the rules will reach OpenAI, Google and Anthropic but not smaller companies, or companies in other countries. Even so, I try to stay optimistic.