All the talk these days is about AI and the dangers that come with it. Rightfully so. But I never thought I’d be writing an article warning folks about AI dangers in my own business so soon. Yet here I am, sharing a cautionary tale about how over-trusting an AI-enriched search for conditions could upend years of trust we’ve built in our field.
Backstory: We get a call from a customer wanting to finish her pre-teen’s scuba certification. No problem. Morning of the dive, they walk in, and the first words out of the pre-teen’s mouth are, “I don’t know why we’re here. I hate scuba diving.” Not exactly a great start. Despite our best efforts, we couldn’t get the pre-teen in the water. The customer enjoyed her own dives and even mentioned to a few people on the ride home that pushing her kid probably wasn’t the best call.
Fast forward four days, and I get an email like nothing I’ve seen in 20+ years running this business. Accusing my staff and me of taking customers offshore in life-threatening conditions. Endangering everyone. Demanding to know how I could expect her kid to finish a certification in weather like that. She wanted a full refund or threatened bad reviews.
I was confused. My captains and I debrief every single day, and that day’s reports were all positive. These customers were on our biggest boat, the Orion — Coast Guard certified, 50 feet with the swim platform, 16 feet wide, pushing 40 tons. We dove a public reef about 9 miles out. I knew the forecast that day, and it didn’t match her story one bit.
Since I wasn’t on the water myself, I didn’t just take my crew’s word for it — I dug in. Checked the nearest offshore buoy: 1.6 to 2.3-foot seas during our dive window. Pulled the NOAA forecast for that zone, GMZ853: winds SE at 5 to 10 knots, seas around 2 feet. Called three other customers from the same trip. All three said conditions were fine — a little bumpy on the way out, nothing over 2 feet, laid down to about 1 foot coming back in. Visibility around 15 feet. So where were these “life-threatening conditions” she described?
Here’s where it gets good. Turns out she’d used Gemini AI to check Gulf conditions, but her search was too vague. It pulled up a tropical system forming in the western Gulf — one that had zero bearing on where we dove and didn’t even start forming until after her trip was over. Gemini spit out a report of 6- to 8-foot seas and life-threatening conditions, matching her complaint almost word for word. Only problem: that forecast was for the Texas coast, two days after our dive. Turns out there’s a name for this — AI hallucination, when bad input leads to confidently wrong output.
I called her up and asked if she wanted to change her story. Nope. I told her I’ve got kids too, and learned the hard way that pushing them rarely ends well. I let her know that if she’d just called and asked for a refund since we couldn’t finish the certification, I’d probably have said yes. Then I sent her the real forecast, side by side with her account and the NOAA buoy data. Advised her to double-check AI next time — and thanked her for the AI lesson and for making it my topic for this article.