Why are AI chatbots sometimes confidently wrong or 'hallucinate'?

AI chatbots are sometimes confidently wrong because they think they know something, but really just made a guess.

Imagine you're playing a game where you have to name all the toys in a big box, but you can only see one toy at a time. You remember what each toy looks like, but if you mix them up, you might say "teddy bear" when it's actually a robot. That’s kind of like how AI chatbots work.

AI chatbots learn by reading lots and lots of text, like books, articles, or even kids' stories. They try to figure out what words go together and what makes sense. But sometimes they get confused, especially if the question is tricky or has a lot of parts. It’s like trying to solve a puzzle with all the pieces mixed up.

AI chatbots don’t just say “I think”, they usually say “I know.” That’s because they use a kind of guessing game where they pick the answer that seems most likely, even if it's not 100% right. It’s like when you guess a word in a game and feel really sure, but then someone says, “Wait, that wasn’t the word!”

So sometimes AI chatbots are confidently wrong because they’re just making their best guess, based on what they've learned before.