When robots forget what they’ve learned

Imagine spending weeks learning to ride a bike. You wobble, you fall, you finally get it — and then one day, someone teaches you how to skateboard, and suddenly you forget how to ride the bike altogether. Sounds ridiculous, right? Well, that’s exactly what happens to some artificial intelligence systems. It’s called catastrophic forgetting, and it’s one of the strangest problems in AI today.
AI learns a lot like humans do — it looks at examples, finds patterns, and stores them as “knowledge.” But here’s the catch: when an AI learns something new, it sometimes overwrites what it already knew. Like deleting an old photo to save a new one, but by mistake erasing your whole album. Scientists call this catastrophic forgetting because, for a machine, forgetting can actually cause chaos.
Let’s say there’s a chatbot trained to help people with their bank accounts. It knows never to share private details like your card number or password. Then, engineers update it so it can answer new questions about investments. After the update, the chatbot might “forget” that it’s not supposed to share private information. Suddenly, it’s saying things it shouldn’t, all because it lost track of its old lessons while learning new ones.
That’s a big deal. In the real world, AI forgetting stuff isn’t just an oops moment — it can mean real mistakes that affect people. Think about self-driving cars. If one forgets how to recognise a stop sign after being updated to handle rainy weather, that’s not just a glitch — it’s dangerous. Or imagine a game-playing AI that was unbeatable, but after a new update, forgets all its old strategies and starts losing to beginners. That’s catastrophic forgetting at work.
You might wonder, “Why can’t the engineers just fix it?” Well, they’re trying. There are clever tricks to help AI remember, like keeping backup memories or teaching it old and new things at the same time. Some systems use a method called rehearsal, where the AI practises old lessons while learning new ones — like revising maths while studying history. Others create “memory modules” that store old knowledge separately, so it doesn’t get overwritten.
But it’s still tricky. The more complex an AI gets, the harder it becomes to stop it from forgetting. It’s like trying to remember everything you’ve ever learned while constantly being taught something new — eventually, your brain might run out of space or mix things up.
There’s also a funny twist: sometimes, we want AI to forget. When someone asks for their personal data to be deleted from an app, the AI has to “unlearn” that information. But making AI forget only one thing without losing the rest is incredibly difficult. It’s like trying to erase one line from a painting without smudging the whole picture.
The truth is, forgetting isn’t always bad — it’s part of being human. We forget things all the time, and sometimes that helps us make space for new ideas. But for AI, forgetting the wrong thing at the wrong time can cause real trouble. Imagine your calculator suddenly forgetting how to multiply because you taught it how to draw graphs — it would be useless!
That’s why scientists are working hard to make AI that learns like a human but remembers like a computer. They want machines that can keep improving without losing what they already know. It’s one of the biggest challenges in artificial intelligence today.
So next time you talk to a chatbot or see a smart robot doing something clever, remember — somewhere in its circuits, it’s fighting a battle against forgetting. Just like you when you’re trying to remember what’s on the test tomorrow. The difference is, when you forget, you can reread your notes. When a robot forgets, someone has to rebuild its memory from scratch.
Maybe one day, machines will learn the way we do — holding onto the important stuff while letting go of the rest. Until then, catastrophic forgetting is a reminder that even the smartest technology still has a lot to remember.