AI Hallucinations Believed To Be Stoppable By Letting AI Know It Is Okay To Be Wrong Sometimes

In today’s column, I examine the ongoing efforts to figure out why AI hallucinations occur, along with finding ways to stop them from arising. It is an immense and highly significant battle. I provide a close look at a recent suggestion about fighting off AI hallucinations that seems to have heartening tendencies.

This latest research asserts that AI is being principally devised to generate answers even when the AI computationally and mathematically doesn’t have a suitable answer in hand. As such, the AI takes a wild stab at generating an answer. Voila, AI hallucinations are provided to user prompts. If you alleviate some of that pressure, presumably the AI would be less likely to generate an AI hallucination. This is a consideration that has floated around for a while, and having robust research to support the contention is notable.

Please do realize that this is one of many working theories. If this is a keystone, the likely approach to overcome AI hallucinations would be to give AI a semblance of latitude toward not forcing answers when an answer isn’t readily available. Ease up on the AI. This can be done at training time, at run-time, and possibly even undertaken during the construction of AI. By taking off the pressure, via one means or another, presumably AI won’t dip into hallucinations to appease users.

Let’s talk about it.

This analysis of AI breakthroughs is part of my ongoing Forbes column coverage on the latest in AI, including identifying and explaining various impactful AI complexities (see the link here).

The Woe Of AI Hallucinations

You have probably heard about or encountered AI hallucinations. The news provides blaring headlines about the thorny matter. It has to do with AI going somewhat awry and generating a response that seems valid, but closer inspection would reveal that it is fake.

They Are Sneaky

When an AI produces a fake or made-up response, this is construed as an AI hallucination.