Attempts to plan mountain climbs using AI-powered chatbots are leading to dangerous incidents in the US. Hikers are camping in restricted areas, navigating using fake maps, and getting trapped on peaks due to algorithmic miscalculations.
Attempts to plan mountain climbs using AI-powered chatbots are leading to dangerous incidents in the US. Hikers are camping in restricted areas, navigating using fake maps, and getting trapped on peaks due to algorithmic miscalculations.
Business Insider reports on the new trend, citing the US Forest Service. According to Nick Meyers, the chief mountaineer at Mount Shasta and director of the Avalanche Safety Center, there have been at least three recent cases where people got into trouble due to overconfidence in AI.
The most serious incident occurred while three 20-year-old novice climbers were climbing 14,000-foot Mount Shasta. The group used Google Gemini, YouTube, and the AllTrails app to plan their route. The chatbot estimated the hike would take 8 hours, so the hikers took limited food and water.
In fact, the climb took 16 hours. The group reached the summit at 7:00 p.m., when it was getting dark. During the descent, one of the men injured his knee, and his smartphone with navigation went dead. The tourists had to spend the night on the slope without proper equipment until they were evacuated by rangers at 9:30 a.m. the next day.
«We relied too much on artificial intelligence and not on our own critical thinking,» the rescued climbers were quoted as saying by the US Forest Service.
Rangers say AI is also causing less serious but systematic violations. In one case, a hiker set up camp under a no-camping sign, saying, «ChatGPT told me I could camp here.» Another climber provided rescuers with an AI-generated map that showed terrain and routes on completely different sides of the mountain.
Technology developers have reservations about such scenarios. OpenAI and Google note on their resources that their language models may make mistakes or «dream,» so information related to safety and life should be checked with official sources.
Experts emphasize that large language models are not real-time geographic information systems and do not take into account current weather conditions, terrain complexity, and human physical fitness.
