This is a term that has come into wider use over the past year or so.
In English, “hallucination” usually refers to seeing or sensing things that aren’t real. In the context of generative AI, a hallucination refers to the phenomenon where the AI produces false information that sounds believable, as if it were true, even though it isn’t.
For example, the AI may provide an answer that sounds convincing but is actually incorrect, or it might mention a book or person that doesn’t exist. This happens because AI generates responses based on patterns and associations found in information on the internet. It tries to give answers that sound right, even if they’re factually wrong. Hallucinations can also occur when you ask about things that don’t exist in the first place.
You can think of it as being similar to a person who confidently says something that’s completely wrong because of a mistaken assumption.
Here are a few examples:
1. Recommending a book that doesn’t exist
You ask the AI, “What are some recommended books on AI?” and it responds, “A popular one is The Future of AI Technology””, but that book doesn’t actually exist.
2. Giving incorrect factual information
You ask, “What’s the elevation of Mount Fuji?” and the AI responds, “2,876 meters,” when the correct answer is 3,776 meters.
3. Naming a fictional person as a real one
You ask, “Who developed the iPhone?” and the AI says, “Steve Rogers was the lead developer.” But Steve Rogers is a fictional movie character, the correct answer is Steve Jobs.
While AI continues to improve at an incredible pace and such errors have become less frequent, hallucinations still happen from time to time. Especially when dealing with important business-related information, it’s crucial to have a human verify the accuracy of what the AI says.
