AI - trust but verify

Or should we even trust?

Very early in my own AI learning journey, I explored some capabilities. Only very recently have I learned to use more than just fancy Google.

In that capacity I asked a broad range of questions to see what kind of responses I'd get. When it was announced that the AI had been given access to Internet knowledge up to a few years prior I thought of a question that should have a very clear and correct answer, and about something I was not intimately familiar.

"Who won the 1958 World Series"

It responded saying that the NY Yankees defeated the Milwaukee Braves in seven games to win their 18th championship in franchise history, and Whitley Ford was named MVP.

I had no idea, but was certain that this was a well documented topic, and not so recent to be after the contemporary cutoff of the time.

Next thing I knew I was in a full blown argument over the MVP, which I readily found to be Bob Turley.

I think I'd get a different result today, but that was my first peek behind the curtain of AI hallucination, and in this case, it's own defence of it's conviction.

I've found AI to be pretty trustworthy on process, procedures and trends. When it comes to fact responses I'm often cautious about full reliance.

11 comments