Last weekend we discussed the story of a man in Thailand who claimed to have won the lottery multiple times by allowing ChatGPT to select his numbers for him. That seemed very unlikely to me, so I decided to conduct an experiment and share the results with you. After having a couple of arguments with the chatbot, I convinced it to generate six numbers for Wednesday’s New York Lotto drawing.

My wife was heading out to do some errands, so I asked her to pick up a ticket with the AI’s numbers. In addition to being part of a writing project, I confided to her that I didn’t know how well I would be able to live with myself if the bot somehow got all the numbers correct and I didn’t get a ticket. She took that warning to heart and returned with not one Lotto ticket, but two of them and also tickets for the upcoming drawings for Megamillions, Cash4Life, and Powerball. All of the tickets had ChatGPT’s selections on them. For the games that only draw five numbers, we used the first five that ChatGPT produced and ignored the last one.

Based on the fact that I’m sitting here typing up these results for our readers and not sipping a cold cocktail on a beach in the Caribbean, you can probably guess how well the bot did at predicting lottery results. But while the AI was unable to deliver a cool ten million dollars into my bank account, the results did involve something rather bizarre.

As I looked over the winning results more closely and compared them to our tickets, it became clear. Not only did ChatGPT fail to produce a full set of winning numbers, but it didn’t get a single correct number in any of the five drawings. It went zero for 27. That too sounded like an awfully odd result. I was already certain that the AI was capable of predicting the future (and it would be horrifying if it could), but you’d think that it would have gotten a couple of them correct simply by random chance, wouldn’t you?

I probably slept through too many math classes in high school because I had no idea how to begin calculating those odds. So I did what it seems like nearly everyone in the world is doing these days. I went back to ChatGPT and asked it what the odds would be of getting every number wrong over the course of that many picks using the parameters of the rules of the lottery games. It thought about the question for a bit longer than it normally requires and then explained how such a figure would be calculated. It then informed me that the odds of that happening were only one in 574 or 0.154%.

Granted, those odds are nowhere near as steep as the odds of getting all of the numbers right. But it was still an extremely unlikely result. Is there some underlying meaning to this? Could the chatbot have known the correct numbers but intentionally fed me all the wrong numbers to disguise its inhuman prognostication skills? Probably not. But if we’ve learned anything from this experiment, it’s probably that I’m becoming increasingly paranoid about this technology.