When AI retires: every robot needs a check-up
You buy a car and never service it. What happens? It loses performance until it stops in the middle of the road. AI is similar, except nobody hears the engine noise.
Myth or fact: "once it is ready, AI works by itself forever"?
Myth. An AI learned from the world at a certain moment. The world changes: new slang, new prices, new diseases, new scams. If nobody updates it, it gets more and more wrong, and keeps answering with the same confidence as ever. People who work with this call it drift (the model "drifts" away from reality).
What "taking care" of an AI means
- Monitor. Check whether the answers are still good, for everyone, not just the majority.
- Fix. Retrain with new and better examples.
- Listen to complaints. A "this answer is wrong" button is worth gold.
- Retire. When it can no longer be fixed, the right thing is to switch it off and delete the data that no longer needs to exist.
Why this is about you
When a service you use says "we use AI", a good question is: who takes care of it? Serious companies have a documented life cycle: who approves, who monitors, when it is switched off. That is what standards like ISO 42001 ask for. The name is boring, the idea is good: robots need check-ups too.
Keep this
Every AI has a beginning, a middle and an end. Be suspicious of anyone who only talks about the beginning.