Help! AI is Hallucinating …
Wow! The ‘year of AI’ has just started and AI is already hallucinating. I’m not joking. Experts and pundits alike proclaim that AI systems (mostly LLMs) are having hallucinations — which (to them) is a very positive thing: It means the bots are becoming creative. Really? It looks more like a bunch of bugs …
No reason get fired up (or scared) just yet. It turns out the (real) creativity is (still) on the human side of the aisle. Here’s the deal: Some experts (apparently quite creative) have looked at recent occurrences of chatbots spewing out garbage and completely fictitious responses and have decided to call it ‘hallucinations’ — in other words, the bots are having a moment so to speak. The experts seem to believe this is proof of creativity, which in turn is a key ingredient in real (human) intelligence. What they seem to be completely ignoring is the fact that this ‘phenomenon’ has a name already, well known and broadly understood: Bugs — aka errors, mistakes, omissions etc.
Why would someone want to call bugs ‘hallucinations’? Wouldn’t it be fair to be honest about the fact that chatbots are software and software is buggy? Big software (LLMs etc. are the largest software systems ever created) have big bugs. Discovering them should trigger bugfixing, not creative coverups. And they should remind everyone that what comes out of an ‘AI’ (ChatBot) is generally…
