A recent news report told of someone who logged on to an online service empowered by AI and complained that the cheese didn’t adhere to the pizza. The answer provided by AI was to try using a non-toxic adhesive. I understand that particular AI service is being tweaked to avoid giving such derisory answers in the future.

A Montreal publisher chose a picture from a stock photo service to use on the cover of a book. The picture showed a young lady standing on one leg with the other partially hidden behind it. After 1,000 copies had been printed someone noticed that the two parts of the partly hidden leg did not line up, almost giving the impression the lady might have three legs. That picture was produced by AI.

Canada’s immigration service recently discovered that a few refugee claimants had been using AI to fill out their claims. The give away? The forms gave detailed information of the dangers the refugee claimant would face if forced to return to their home country, complete with hyperlinks to reputable sources. The hyperlinks did not work, the supposed information did not exist.

A newspaper tested an AI service by asking it to answer ten questions from a nursing exam. The service responded with ten well-written answers. Only eight of them were correct. The newspaper commented that evidently this AI service had not been programmed to admit “I don’t know,” so it was always going to provide a seemingly plausible answer based on bits of information that did not fit together.

This sounds like the way my imagination works. When faced with an unknown situation it ransacks my brain for past experiences and puts together a scenario of the trouble that lies ahead or what this person I just met is really like. I have learned not to trust my imagination as a source of truth. Until I am fully assured otherwise, that is also my approach to AI.