AI risks

The most real risks of AI for someone who uses it every day are three: that it gives you false information with total confidence, that you share data you shouldn't have, and that you end up depending on it to think. There are other bigger, much-debated risks, like jobs, disinformation or control of very powerful systems, but day to day it's those three that catch you out.

The false information has a name, hallucination: AI doesn't look for the truth, it generates the answer that seems most likely, and sometimes that's a very well-written invention. The data issue depends on each service: what you type may be stored or used according to its terms, and with client data that's already a serious matter. It's misleading to think the risk is the robot-movie kind or, the other way round, that there isn't any. And the risk grows when the AI doesn't just answer but acts for you (sends emails, touches files): a mistake no longer stays on the screen.

To see these risks up close, and not just in theory, I did a slightly absurd challenge I don't recommend repeating as is. In "A week using artificial intelligence for everything" I tell you what happened when I consulted it on decisions, texts, purchases and work for seven days: what got better, what got worse and where I stopped thinking for myself.