Siskiyou County Sheriff’s Office
There is a new AI cautionary tale making the news, and it isn’t about a chatbot writing a bad blog post or blundering through trivia.
It involves three hikers, a mountain, and a very real rescue.
Recently, three novice climbers from Roseville, California, had to be rescued from Mount Shasta after relying heavily on Google’s Gemini to help plan their climb. According to the Siskiyou County Sheriff’s Office, the hikers used AI for route information and advice about what to bring, and they ended up with less food and water than the prolonged trip ultimately required.
Underpacking might seem like a minor oversight when you’re sitting on your couch planning a trip.
It becomes a critical error at 14,000 feet.
The Plan vs. Reality
The group set up camp at about 8,400 feet and began their summit push around 3 a.m., operating on an estimated eight-hour timeline.
But the mountain didn’t agree with the estimate.
- Delayed Summit: They didn’t reach the peak until about 7 p.m. – well past the recommended noon turnaround time.
- Dangerous Descent: Descending in the dark, they lost the trail and wandered into Mud Creek Canyon, where one hiker fell and injured his knee.
- Overnight Survival: The group had to spend the night on the mountain until U.S. Forest Service climbing rangers reached them the following morning.
The Siskiyou County Sheriff’s Office called their reliance on AI a “critical misstep.”
And that’s really the heart of the story.
The problem isn’t necessarily that they used AI to help plan the trip.
It’s that they made the dangerous leap from:
“AI gave me an answer.”
to:
“Therefore, the answer must be correct.”
Two very different things.
The Illusion of Expertise: AI Sounds Like It Knows
AI systems are extraordinarily good at producing answers that sound authoritative.
That’s one of the biggest issues here.
A search engine generally gives you links to sources that you can evaluate yourself. An AI chatbot gives you a synthesized response – one that can be beautifully written, logically organized, and entirely wrong.
We’ve seen the absurd side of this before.
Remember when Google’s AI suggested adding Elmer’s glue to pizza sauce to help keep cheese attached?
It’s funny when the subject is pizza, and it’s obvious (or should be) that it is so wrong.
It’s a very different story when the subjects are hydration, navigation, weather, elevation, and survival.
An Eight-Hour Hike… For Whom?
Consider a simple question:
“How long does this hike take?”
An AI might give you an answer such as:
“Approximately eight hours.”
Sounds useful.
But eight hours for whom?
A seasoned mountaineer?
A novice?
Someone carrying a light pack?
Someone carrying extra equipment?
What happens if the weather changes?
What happens if the trail is difficult to follow?
What happens if someone gets injured?
What happens if the trip takes twice as long as expected?
An experienced mountain guide isn’t just thinking about the average time to reach the summit. They’re thinking about the conditions, the people, the terrain, the equipment, the turnaround point, and what could go wrong.
That’s the difference between information and judgment.
AI can provide information. Sometimes accurate, sometimes not.
And it doesn’t automatically give you the judgment of someone who has actually been there.
Outsourcing Judgment Has Real Consequences
And this problem extends far beyond backcountry hiking.
People are increasingly asking AI for guidance on things that can have real-world consequences – health questions, legal matters, finances, cybersecurity, home repairs, travel, business decisions, and technical problems.
It can become a huge problem when AI becomes the final authority without anyone checking its work. This is why there are disclaimers like “Gemini is AI and can make mistakes.” and “ChatGPT can make mistakes. Check important info.”
This applies to business and web development too.
The current AI marketing narrative often sounds like this:
Let AI write your code.
Let AI generate your content.
Let AI manage your SEO.
Let AI build your website.
Let AI create your marketing strategy.
Put everything on autopilot.
We’ve explored this idea before in our article about the “AI Employee” illusion:
The “AI Employee” Illusion: Why Fully Automated SEO Is Still Marketing Fiction
AI can generate code and content at remarkable speed.
But generating something and knowing that it’s correct or good are two different things.
Someone still needs to understand the subject matter well enough to recognize when the AI gets it wrong.
AI Doesn’t Have to Be Bad to Be Dangerous
You don’t have to believe that AI is “stupid” to recognize its limitations. In fact, that’s almost the point.
AI can be extremely impressive.
It can write an article in seconds.
It can explain complicated programming concepts.
It can analyze enormous amounts of information.
It can brainstorm ideas that might not have occurred to you.
And it can still confidently give you an answer that shouldn’t be trusted.
That’s what makes the technology different from many older tools.
The danger isn’t necessarily that AI sounds foolish.
The danger is that it can sound extremely convincing when it is wrong.
The Takeaway
The Sheriff’s Office advised hikers not to rely solely on AI and to verify their plans with local ranger stations and knowledgeable human sources.
That’s good advice – not just for hiking, but for AI in general.
Use AI to brainstorm, research, outline, summarize, and explore possibilities.
Question AI when the stakes move beyond surface-level information.
Verify AI against reliable sources, current information, real-world conditions, and qualified experts.
AI can make a very useful assistant.
But it makes a terrible substitute for judgment.
And if you’re standing on a mountain at 14,000 feet, running out of food and water, and the chatbot’s eight-hour estimate turns out to be wrong…
That’s really not the time to discover that lesson.