Hallucinations in AI are not the same as they are with humans. We break down what an AI hallucination is, why it happens, and what it means for the field.
If you’ve been on the internet at all throughout 2022 and 2023, you’ll have seen a lot of hubbub about AI chatbots like ChatGPT and, more recently, Microsoft’s Bing AI. These AI systems use a combination of natural language processing and information gathered from training on huge data sets to interact with users. They’re a far cry from the primitive chatbots you might have encountered in the early days of the internet. Microsoft’s Bing Chat brought these AI smarts to its infamous search engine when it launched on February 7, 2023, connecting Bing and ChatGPT to enable a live, internet-connected ChatGPT that could answer questions based on the information it gathers in real-time.
The AI systems we’re seeing pop up are incredibly convincing in their delivery, and the information they serve seems to be accurate, but the longer we spend tinkering with them, the more the flaws seem to pile up. People have been using AI to write code, do homework, and even write articles and video scripts to varying degrees of success. One of the biggest challenges with text-generative AI models is called „hallucinating,“ and it’s quickly eroding the sheen off the otherwise revolutionary tech.
