I think we need a serious reckoning on this. AI does not “hallucinate.” It can’t. It is not a human. What it does though is fail. Yes, fail.
When you ask it to do something and it comes back with incorrect information that it presents confidently as correct, that is not a hallucination. That is a failure of its basic function. We absolutely need to start calling it what it is.
Because to keep calling it hallucination makes it seem innocuous. Or like a cute little oopsie. But these systems want more and more access to sensitive data. Without addressing this very basic issue. They fail at least 20% of the time, often more depending on the ask. Would you accept at 20% failure rate from another service? Of course not. But yet here we are not only accepting this terrible performance but wildly integrating it into more and more and more systems to where it will soon become unavoidable.
What are we doing?
And the more this gets normalized, the less we will even question AI results. That should terrify us all. The idea that we outsource most of our thinking to where we just blindly trust a product that we know fails a lot. The hype machine for AI is intense, I will grant that. But it is not inevitable, it’s just being pushed as if it was. Whether that turns out to be true is actually up to us, not the AI overlords who desperately want to keep their shell game going for a little longer before this bubble inevitably bursts.
So I am begging us all to stop letting these companies get away with marketing a mistake engine as something less insidious – oh it just hallucinates sometimes. No biggie! It actually is quite the biggie.
And it is why I offer human powered PPC services. AI May eventually have a place, but that day is not today. Not for me, when it fails as often as it does. I take my clients’ business too seriously to use a tool that fails at a 20% rate. If I did that I would expect to be fired immediately and a refund requested.
Let’s hold the machines to that standard before it’s too late.
Leave a Reply