MT 43 News Articles View a Published Article

AI – Amusing and a Little Terrifying

Author: Victor Sample Vic Sample: MT43 News Treasurer

Victor Sample

The other day, I asked AI what I thought was a simple question: On many of the Food Network Baking Championship shows, the host doesn’t get served the food. Only the judges get the food. Why?

It seemed like a pretty straightforward question. The answer I got was well formulated and coherent and given with all confidence – and was absolutely wrong. Basically, the answer was that it would affect the baker's time to create the extra serving, and it would cost extra for the ingredients.

A totally reasonable and well-written answer – and totally wrong. It’s a baking show. The bakers are asked to create a dozen cookies, a pan of brownies, a whole cake or pie, etc. In other words, no extra ingredients would be required, and no extra time would be required to serve a dish to the host.

When I told the AI engine the answer was a total fabrication, the AI engine apologized and admitted that it gave an answer based on statistical probabilities without having any actual basis for the answer.

In this case, I knew immediately the AI engine hallucinated the answer, but in many cases, that is not the case. People accept the answers even when there is absolutely no basis. I find that a little frightening.

I was reading an article written by someone in the marketing business that was lamenting that professional marketers were some of the first people to adopt use of AI in their work, although AI has advanced, marketers have not advanced their usage. They still just treat what they get from AI as a rough draft and go from there.

The author said that in the early days of AI, while the output was well formulated, it was not always very accurate or usable. There were many AI hallucinations. He went on to say that with all of the new advances, only around 7% of the answers are hallucinated.

ONLY 7%!

Imagine if you are a manager of an employee, or a coworker of an employee, who knew 7% of what the employee told you was totally made up; no actual facts and false references for what was said.

When I was a manager, I would have fired an employee who totally fabricated 7% of what they told me!

In the world of AI (at least according to the author of the article I read), 7% hallucinations are perfectly acceptable (of course, it is marketing…)

I find that attitude to be more than frightening!