The Vietnam Draft, AI, and Who Can You Trust
The other day, some friends and I were discussing the Vietnam draft lottery (they televised the thing. Talk about must see TV!) because one of the guys is older and we were asking what his draft number was. For those of you who don’t know, for four years from 1969 - 1972, the US government would dump 366 birthdays into a jar, pull them out one by one, and whichever spot your birthday fell was the order in which you would be called to serve.
I was curious. What would my numbers had been if I weren’t 2 years old but instead 19 at the time. I asked an AI bot – Claude.AI in this case. Turns out some years I would have gone, and others my draft number would be high enough to keep me at home. After satisfying that curiosity, I tried my sons’ birthdays (one of them had a draft number of 2. Ugh.) and then my father and grandson’s shared birthday. And here’s where it got weird. Claude returned a “fact” that dad and grandson’s birthday was chosen in the #9 spot – FOR ALL FOUR YEARS! I was floored by that statistical anomaly. That’s almost an 18 billion with a B chance of that happening.
I texted my father about it. Texted my son about his son’s birthday’s crazy rarity. And thought, with something so statistically difficult and unlucky as that, how had I never heard of it before? It’s not like the Vietnam draft wouldn’t have been front page news. That date coming up again and again would certainly have been significant from the odds alone.
And do you want to know why I hadn’t heard about it (or you hadn’t}? Because the AI was wrong. And it was wrong all the way down the rabbit hole. Built me a great backstory about other birthdays that had duplicates and that nothing had come close to 4 in a row. Etc. Etc. Etc.
And all wrong.
So, I thought, well maybe Claude isn’t as powerful as some of the others. I asked the same question to Google Gemini and ChatGPT and Microsoft Copilot. Surely they would give me the right answers. Four queries, four AIs, and each one gave me different sets of numbers, and each had at least one if not several mistakes. We’re talking about a birthday. Where did {insert date here} end up ranked in all four Vietnam draft lotteries? This would seem like a well documented set of numbers. How could all four be wrong, each in its own way? I asked each bot how they got it wrong. I gave them correct numbers, and they’d fix it but then have others wrong. It was frustrating as well as misleading.
Consider this a word of warning. No matter which AI tool you rely on, how much should you really trust its output? I'm not arguing against using it — I've leveraged it myself to analyze open-ended survey responses, though it consistently fell short on the kind of nuance a human reader catches instinctively, reducing rich answers to surface-level summaries. The bottom line: if you're outsourcing your thinking to AI, don't be shocked when the facts unravel and the bill comes due.