Hallucination
A hallucination is when an AI model convincingly states something that isn't true. Instead of saying it doesn't know, the model guesses an answer that sounds plausible.
Hallucinating comes from how a language model works. It predicts likely text, and sometimes the most likely answer isn't the right one. There's no bad intent behind it, but it is a known risk you have to manage, especially when customers see the answers.
You limit it by having the model draw on a controlled knowledge base, by having it hand off to a person when it's unsure, and by having a person confirm important actions.
In practice.
In the review handling for Inge de Bruijn Zwembad, a person approves the reply the AI suggests, precisely to keep a hallucinated answer from going out.
Frequently asked questions.
What is an AI hallucination?
It's when an AI model convincingly states something that isn't true. It guesses an answer that sounds plausible instead of saying it doesn't know. That comes from how language models work.
How do you prevent hallucinations in a business AI?
By having the model draw on a controlled knowledge base (RAG), having it hand off to a person when it's unsure, and having a person confirm important actions before they go out.
Related terms.
Want to see how this works in practice? Have a look at AI assistant and AI receptionist.
Curious what this could do for you?
Walk me through one process. In half an hour I will show you where this kind of work gets stuck in your business, and what it would save to hand it off.