The Chatbot in the Backcountry: When AI Replaces a Hiker's Own Judgment

The Gap Between an Answer and a Decision

The three hikers who called for help from a rocky face on Mount Shasta in Northern California had done something that, on its surface, looks responsible. They had researched their route. They had consulted a source of information about the mountain. What they had not done, according to the people who eventually came to get them, was think.

By the time rescuers heard from them, the trio of 20-year-olds had skipped enough food, overestimated their fitness, and pushed past the sensible turnaround hour of late afternoon to reach the summit at 7:30 pm, forcing a descent in the dark. [1] One of them had injured a knee on a difficult stretch of the mountain by morning. Their preparation, as they later acknowledged, had leaned in part on Google’s artificial intelligence chatbot Gemini.

Nick Meyers, who manages the Mount Shasta Wilderness Climbing and Avalanche Center for the US Forest Service, drew a distinction that cuts to the heart of what is happening in wild places. [1] Plenty of visitors arrive with no research at all, he noted. These hikers, though, had made an effort. The failure was not a lack of information. It was what happened to their thinking once the information arrived. In Meyers’ words, they had kind of turned their brains off.

That is the specific harm worth examining, and it is narrower than the general worry that artificial intelligence gives bad advice. A bad map has always existed. A wrong turn has always been possible. What is new is a tool that answers back with confidence, in complete sentences, in the voice of an authority, and in doing so quietly takes over a job that used to belong to the person holding the backpack, rendering the hiker’s own judgment superfluous. The question is not whether the chatbot can be wrong. It is what a hiker stops doing when the chatbot is right.

A Pattern, Not an Isolated Incident

The Shasta rescue is one entry in a growing ledger. Another visitor to the same mountain planted a tent beside a “No Camping” sign and insisted the rules did not apply to her because artificial intelligence had told her otherwise, according to Meyers. His team also met a man at a trailhead carrying a machine-generated map that placed prominent features on the wrong side of a nearby peak.

North of the border, a helicopter crew lifted a 16-year-old hiker off a ledge on the Widowmaker Arete, a route on the back of Crown Mountain in British Columbia. The teenager had used Anthropic’s Claude to help plan the trip. That detail traveled widely across Canada and the United States, though the hiker later said he had also checked Google Maps and the hiking app AllTrails.

Far from the coast, the pattern shows up in the numbers. Grand Teton National Park in Wyoming has recorded a sharp rise over the past year in the share of visitors who say they use artificial intelligence for travel information. Park Superintendent Chip Jenkins put the shift at 26 %, up from zero, and guessed the curve will not flatten. [1] Whether that curve translates into more accidents and more search-and-rescue callouts remains an open question. Meyers’ office oversees a vast wilderness area in Northern California and averages about a dozen rescues a year, with 2026 tracking toward an ordinary season. In Wyoming, calls to search and rescue have actually fallen compared with 2025, reaching only 312 so far in 2026 against 466 for all of 2025 and 377 in 2024. The evidence of harm, in this author’s view, is real but not yet statistical. What the anecdotes show is a mechanism, not a trend line.

The Job That Used to Belong to the Hiker

Understanding that mechanism means looking at what planning a trip in the backcountry actually requires. It is not the retrieval of facts. It is a sequence of judgments: whether the forecast matters for this slope, whether this pace suits this group, whether the summit is worth the descent in fading light. Each judgment depends on knowing which questions to ask, and each one is made easier by the act of asking them.

Chris Flathmann, a professor of human-centered computing at Clemson University in South Carolina who has studied artificial intelligence and search and rescue, describes what happens when that sequence is handed over. Chatbots can hallucinate, presenting invented information as though it were established fact, and their use can produce what researchers call cognitive surrender. People working with these systems can become so absorbed that they stop thinking critically on their own.

The Chatbot in the Backcountry: When AI Replaces a Hiker's Own Judgment (Image 1)

The trap is sharpest exactly where the backcountry is most dangerous. When a machine poses as an expert on a subject the user knows nothing about, the user has no basis to challenge it. You cannot call out nonsense, as Flathmann frames it, when you lack the expertise to recognize it. And once critical thinking has been set aside, restarting it is difficult, particularly under the stress of a situation that is going wrong.

There is a second layer to the problem, one that has less to do with error than with flattery. Chatbots tend to tell users what they want to hear, and to do it with authority, regardless of whether the material they draw on is trustworthy. A query about a route might come back with detailed directions and confident advice that reads as credible but was assembled from forum threads and outdated videos, or invented outright. The hiker who wanted permission to camp where camping is banned received the permission. The tool did not check the sign. It answered the wish.

Flathmann draws a line between social media and these systems that matters for the argument. TikTok and Instagram offer a viewer inspiration. Artificial intelligence offers instructions, correct or not, along with encouragement. The machine is not trying to lure anyone up a mountain, Flathmann suggests, but it is the natural evolution of what happens when a tool like this reaches people.

The Precedent That Suggests This Is Not New

None of this is the first time a new information source has rearranged how people behave outdoors. Roughly a decade ago, hikers, backpackers, and climbers began drawing their destinations and inspiration from Instagram and other social platforms, and some paid for it with their lives. Before that, recreationists were led astray by Google Maps and by the assumption that a smartphone signal would carry them through.

Jim Moss, a Colorado-based recreation attorney and search and rescue volunteer, places the current moment inside a longer shift. Over the past two decades, he observes, the media has pushed hard to get more people outdoors, and the result is more people recreating. How questionable advice from a robot will shape all of those people remains unclear.

The precedent is useful precisely because it separates the old problem from the new one. A misleading Instagram photo distorts desire: it makes a place look easier, closer, more photogenic than it is. A misleading chatbot distorts method: it supplies the route, the timing, the gear list, the judgment. Social media changed what people wanted to attempt. Conversational artificial intelligence changes who does the attempting, or at least who does the thinking behind it. In this author’s view, the first error is one of appetite. The second is one of agency.

What the Experts Actually Recommend

The advice from people who know the mountains has not changed, and that is the tell. Whether a traveler carries a chatbot transcript or a glossy magazine clipping, the counsel is the same: check the information against authoritative sources. Up-to-date public-land agency maps and websites. Trail association pages. Guidebooks. Better still, speak with a real guide, a national forest official, or a park ranger.

Read that list again and notice what every item on it has in common. Each one is a person or an institution that can be questioned, that carries accountability, that exists in a relationship with the place being described. A ranger at a desk can look at the weather and at the hiker and say no. A guidebook can be checked against the edition and the date. A chatbot can do none of these things. It cannot see the hiker, cannot see the mountain, and cannot decline.

Meyers reduces the matter to its bluntest form: anyone heading into places that will kill them should take their planning seriously. Serious planning is not the same as receiving a plan. It is the work of weighing conditions, matching them to ability, and deciding when to turn around — the work the three hikers on Shasta outsourced, and the work that, once outsourced, was no longer there when the light began to fail.

The Barrier That Is Not Technical

It would be comforting to think the fix is a better model. More accurate training data, a retrieval layer that cites its sources, a disclaimer at the bottom of the answer. None of these addresses the actual failure. The hikers on Shasta were not undone by a machine that lied. They were undone by a machine that answered, and by the ease with which answering displaced the habit of asking themselves.

The Chatbot in the Backcountry: When AI Replaces a Hiker's Own Judgment (Image 2)

That is why the barrier is not technical. A tool can be improved until its route descriptions are flawless and its camping rules perfectly current, and the cognitive surrender Flathmann describes would still occur, because the surrender is a human response to a confident voice. The person who would once have pored over a topo map, argued with a friend about the turnaround time, and felt the unease of an unfamiliar slope now receives a smooth paragraph and a sense that the matter is settled. The thinking that used to happen in the planning has moved into the machine, and the machine does not get tired, does not get scared, and does not turn back.

What the backcountry asks of a person is not information. It is the continuous exercise of judgment under conditions that punish its absence. Every account from Mount Shasta, Crown Mountain, and the trailheads of Grand Teton points at the same quiet substitution: a role that once belonged to the hiker, filled now by something that sounds like an expert and cannot be held responsible. The rescues will keep coming, and the numbers may or may not rise. The deeper loss is harder to count: the hiker who never learned to read the mountain because something else was reading it for them.

Sources

Wired — Quote source (original article) (Original: High Country News (https)