Hallucination
An AI system producing an unsupported or incorrect output that may appear convincing.
When AI Just Makes Stuff Up
You ask an AI for directions. It sends you to a restaurant that does not exist.
Congrats. You have met a hallucination.
What it is
A hallucination is when a model says something false like it is absolutely, positively, hand-on-heart true. Not "I’m guessing." Not "I’m not sure." More like your mate giving confident IKEA assembly advice while holding the shelf upside down.
This happens because AI does not "know" things the way you do. It predicts the next likely word, then the next one, then the next one. Usually, that works great. Sometimes, it turns into a very convincing nonsense machine.
So you get things like:
- fake facts
- invented sources
- wrong summaries
- made-up legal cases
- a recipe that somehow includes 14 cloves of garlic and emotional damage
Why it’s tricky
The dangerous bit is not that it gets things wrong. Humans do that before coffee every day.
The dangerous bit is the confidence.
A hallucinating model often sounds smooth, polished, and weirdly sure of itself. Like a bad driver who missed the exit but says, "No no, this is a shortcut."
That makes it easy to trust, especially when the answer looks detailed. More words do not equal more truth. Sometimes they just mean the blender was on longer.
Why "arms" makes it worse
If the model is just writing text, a hallucination is annoying.
If it is attached to arms, wheels, tools, or anything that can move in the real world, now the stakes jump. A fake answer in a chatbot is one thing. A robot confidently grabbing the wrong bottle in a lab is another. That is not a typo. That is a problem.
Why you should care
Hallucination is the core reminder that AI can sound smart without being right. And if you forget that, you are letting the world’s most confident intern drive the forklift.