I grew up planning road trips with an AAA TripTik, a stack of folding maps that never folded back the way they came, and a Thomas Guide on the passenger seat. Before I ever left the driveway, I knew how to get from A to B. I didn’t think of it as a skill at the time. It was just how you planned a trip.
Now I get in the car and do whatever the GPS tells me.
The same thing happened with research. In high school, I learned to use the library: pulling reference books, reading through multiple documents, and digging until I found the one fact I needed to support my hypothesis and my thesis. College reinforced that habit, except the library was replaced by online searches. The tools changed, but the work didn’t. I still had to read the sources, compare them, and decide for myself which one actually supported my argument.
Today, research looks like a conversation with AI. I ask a question, it answers, I ask a follow-up, and it answers again. It’s fast, it’s convenient, and if I’m being honest, I’m not sure I’m actually learning anything.
I’m not here to argue that GPS or AI is bad. Both are excellent at getting people information quickly, and for someone who just needs the final answer — the fastest route to a restaurant, a quick definition, a number to drop into a report — that’s exactly what they should do. But there’s a difference between someone who needs the answer and someone who needs to understand how the answer was reached. When we’re developing professionals, or anyone whose job depends on the underlying knowledge, the tool can’t replace the fundamentals. And when the tool does the thinking, the thinking stops happening, for people who already have the skill as well as the ones who never built it.
What the Research Says Happens When the Tool Does the Thinking
The navigation research is surprisingly clear. In a 2017 study published in Nature Communications, researchers at University College London scanned the brains of 24 volunteers as they navigated a simulation of London’s Soho district. When people chose their own route, the hippocampus and prefrontal cortex lit up every time they entered a new street, and activity increased as the number of route options grew. When they were simply told which way to turn, that activity disappeared. As senior author Hugo Spiers put it, “these parts of the brain simply don’t respond to the street network.”
A few years later, McGill University researchers Louisa Dahmani and Véronique Bohbot studied 50 regular drivers and found that people with more lifetime GPS experience had worse spatial memory when navigating on their own. When they retested a small group three years later, the people who had used GPS more in the meantime showed steeper declines. Importantly, the heavier GPS users didn’t rely on it because they had a poor sense of direction to begin with, which points to the tool contributing to the decline rather than simply attracting people who were already struggling. The follow-up group was small, only 13 people, so the authors are careful about drawing hard conclusions, and I will be too.
The early research on generative AI is pointing in the same direction. A 2025 study from Microsoft Research and Carnegie Mellon University surveyed 319 knowledge workers about 936 real tasks. The more confidence people had in the AI, the less critical thinking they applied. The more confidence they had in their own ability to do the task, the more critical thinking they applied. Separately, Michael Gerlich of SBS Swiss Business School surveyed and interviewed 666 people and found that frequent AI use was associated with lower critical thinking scores, largely through what researchers call cognitive offloading, which means handing mental work to an external tool. That study is correlational and relies on self-reports, so it can’t prove that AI causes the drop.
There’s an honest caveat here, too. The most famous study in this space, Betsy Sparrow, Jenny Liu and Daniel Wegner’s 2011 “Google effect” paper in Science, found that people who expect to be able to look information up later remember the information itself less well. A large replication project published in 2018 did not reproduce that result. The research on digital tools and memory is still developing, and anyone who tells you the science is settled in either direction is selling something.
Why This Should Worry All of Us, Not Just Novices
Look closely at that Microsoft finding again. Critical thinking didn’t disappear for everyone who used AI. It held up for people who were confident in their own ability to do the work.
It would be easy to read that and conclude that experienced professionals are safe. I’d like to believe that. When a GPS sends me somewhere strange, I notice, because I know roughly where I should be. When AI hands me a statistic that sounds too clean, I still know how to go find the source.
But knowing how to find the source and actually doing it are two different things. Those hours in the library weren’t just a slow way to get to a fact. The reading, comparing, and discarding were where the learning happened. I absorbed context I didn’t know I needed, ran into arguments I hadn’t considered, and built a sense of what a credible source looks like. A conversation with AI hands me the destination and skips the trip. I can recognize a good answer because I spent years learning the hard way. I’m much less sure I’m building anything new.
If that’s true for someone with 25 years of experience, it’s even more true for a new employee. Someone who has only ever followed turn-by-turn directions has no mental map to check them against. A junior designer who has only ever asked AI for a learner analysis has no way to tell whether the analysis is any good. Experienced people risk letting their skills erode. Novices risk never building them at all.
So What Do We Do About It?
I’ve argued before on this blog that a lot of training should be a job aid instead, and I still believe that. A GPS is the ultimate job aid. If you’re driving to a restaurant you’ll visit once, learning the map would be a waste of time, and the same is true for anyone who just needs the final information to get their job done. But a delivery driver, a paramedic, or a city planner needs to understand the roads, not just follow directions. The same distinction applies to everything we train. The question isn’t whether people should use the tool. It’s whether they need the answer or need to understand it. In practice, that means a few things.
Decide deliberately who needs the answer and who needs the understanding. For every audience and task, ask whether learners just need the final information, where a job aid or AI tool is enough, or whether they need to understand the underlying knowledge, where we have to teach the fundamentals. A customer service rep can look up a return policy. A customer service trainer needs to understand why the policy exists, so they can explain it, defend it, and recognize when it doesn’t fit.
Build the skill before you hand over the tool. Have learners work through the analysis, write the first draft, or solve the problem manually before showing them how the AI tool does it. Psychologists Robert and Elizabeth Bjork have spent decades studying what they call desirable difficulties: conditions that make learning feel harder in the moment but produce stronger, longer-lasting learning. Doing it the slow way first is one of those difficulties.
Teach verification as a skill in its own right. The Microsoft researchers found that AI shifts critical thinking toward checking information and overseeing the work. That’s a real skill, and we can teach it. Give learners AI output with planted errors and ask them to find them. Make “where did this come from?” a required question in every activity that uses AI.
Watch for misplaced confidence. If the research is right that confidence in the tool lowers critical thinking, then our job includes building learners’ confidence in themselves, not just their comfort with the software.
Back to the Thomas Guide
I don’t miss the TripTiks, and I certainly don’t miss trying to refold a map in a moving car. But the library taught me something a chatbot can’t: the answer you dig for is the one you keep. The skills I built the slow way are the reason I can still question a GPS or an AI, and I’m learning that if I don’t keep using them, I won’t keep them.
Our learners deserve the same foundation, and we owe ourselves the same discipline. Give them the map first. Then hand them the GPS. And every once in a while, put the GPS away and find your own way there.