ChatGPT can’t climb mountains, but it can give you bad advice on how to do it.
Relying on AI-generated advice proved to be a “critical misstep” for three climbers whose planned 8-hour ascent of California’s Mount Shasta spiraled into a grueling three-day ordeal. The climbers were finally rescued in a complex, multiday operation following an injury and two failed helicopter airlifts.
Setting off at 3 a.m. on Sunday, the group reached the mountain’s summit 16 hours later, only to get lost on their descent after one of them fell and injured their leg. Rescuers safely evacuated the group on Monday.
Hikers admitted to rescuers that they’d used several tech tools to plan their trip, including YouTube, the outdoor fitness and recreation app AllTrails and Google Gemini AI. Relying heavily on Gemini was a major mistake, the Siskiyou County Sheriff’s Office said in a Facebook post.
“The men told a Sheriff’s Deputy on scene that they had relied heavily on Google’s Gemini AI to provide them with information about the route as well as what to pack for their trip,” the post reads in part. “This was a critical misstep, as they were advised by Gemini to bring far less food and water than their group required, especially when their planned 8-hour ascent became a multiday ordeal. “
The sheriff’s office went on to warn people to “never rely solely on AI for your trip planning.”
Representatives for the US Forest Service and the Siskiyou County Sheriff’s Office didn’t immediately respond to requests for comment.
There were other problems with the trip. The hikers reported that their phone batteries died and that their backup batteries wouldn’t function. Without offline maps available, the disoriented group lost its way on the descent. Further reflecting their lack of proper gear, the injured hiker navigated in total darkness without a headlamp to save battery power.
A Google representative told CNET in a phone call that its trust and safety team is investigating the climbing advice generated by Gemini. The company has been unable to reproduce the answers the climbers received. The representative acknowledges that Gemini can make mistakes and says that people using it should always check Gemini’s cited sources if they’re unsure about a result.
Although the climbers have not said exactly what Gemini suggested they bring for food and drink, or how much to bring, the sheriff’s department has specifically mentioned this as an issue.
I attempted to prompt Gemini to give the kinds of responses the climbers say they received, but was unsuccessful. When I asked what foods it recommended for a mountain climb, Gemini recommended a mixture of carbs, fats and proteins. Even if I asked for a low-fat solution, Gemini still mentions that climbers often bring high-sodium, high-fat and calorie-dense items like cheese and cured meats. It did not suggest amounts until I specifically said how long I expected my fictional hike to last.
Artificial intelligence often overlooks critical details, leaving users ill-prepared for real-world situations. Sometimes the fallout is minor. Late last year, CNET editor Corin Cesaric-Epple used ChatGPT to meal prep for a week and noted on multiple occasions that the AI-provided recipes were heavily underseasoned. But other times, the consequences can be severe. Lawyers who have used AI to assist with legal filings have jeopardized their careers, with some receiving formal reprimands for error-laden work.
Generative AI tools simply predict patterns in the data they were fed, much of which comes from unverified web sources. As a result, experts urge users to independently verify any AI-generated advice before relying on it. Since AI doesn’t actually eat food, practice law or climb mountains, it’ll never be the best place to learn how to do those things.
