16 July 202610 min readAttention, Decision-making

The Things You Find Along the Way

AI gets you to the answer without the detours, and the case for the detours is usually sentimental. There's a colder version of that case, built on portfolio theory, that survives contact with the evidence.

If you've used AI to research anything in the last year, you've noticed the same thing I have. You get what you're looking for faster. You also stop finding things you weren't looking for.

The standard argument about this goes: AI gets you to the destination without the journey, and the journey is where the real learning happens, so we're losing something important. I think this argument is half right and dangerously sentimental about the half it gets wrong. Because the things you find along the way are not reliably good. Some of them matter. Most don't, and some actively hurt you. The people making the case for the journey almost never do honest accounting on that mix.

I want to do the accounting.

What GPS did to your brain

Before the question was about AI, it was about GPS. And with GPS, we have data.

A 2020 study in Scientific Reports (Dahmani and Bohbot) tracked people's GPS use over time and found that habitual GPS users had measurably worse spatial memory than people who navigated without it: their hippocampal-dependent spatial strategies declined, not just their sense of direction. The hippocampus builds cognitive maps, the internal representation of space that lets you navigate from memory rather than instructions. Heavier GPS use correlated with less reliance on these strategies, and the effect was dose-dependent: more GPS, more decline. People felt just as good at navigating as ever; they weren't. That gap, between the decline and the missing subjective sense of it, is what makes the authors argue the relationship is likely causal rather than people with bad spatial sense simply gravitating toward GPS.

This is worth sitting with, because the GPS finding is not a metaphor. It is a measured, replicated degradation of a cognitive capacity, caused by the removal of a particular kind of effortful engagement. The system that navigates for you also degrades the system that would have done the navigating itself.

The question is whether something structurally similar happens when AI does your intellectual wandering for you.

Explore and exploit

In cognitive science, the tension between trying new things and sticking with what works is called the explore-exploit tradeoff. Your brain manages it through dopaminergic signaling: dopamine modulates the degree to which you engage in directed exploration, seeking out new information about uncertain options, versus exploitation, repeating what's already worked. A 2020 study in eLife (Chakroun and colleagues) showed that this modulation is causal, not just correlational. Manipulating dopamine levels changed how much people explored.

The tradeoff is a real constraint, not a figure of speech: at any given moment you are either exploring, gathering new information, or exploiting, acting on what you already know, and you cannot do both simultaneously. The optimal balance depends on how much time you have left, how uncertain your environment is, and how much you trust your current option.

AI shifts the tradeoff. It makes exploitation cheaper and faster. You can find and deploy the best-known answer in seconds. What it also does, less obviously, is make exploration feel unnecessary. Why wander through a citation chain when you can ask directly? Why read the twelve papers that aren't quite right when you can get the one that is?

Tyler Cowen wrote about this in July 2026 on Marginal Revolution, noting that when exploration costs collapse, researchers drop bad questions faster and curate better portfolios of ideas. That's the upside. The downside he flagged: the ability to hold together complex positions verbally may decline when preliminary exploration is cheap, because less time is spent grinding through arguments from first principles. And the capacity for undirected discovery, the kind that happens when you're wandering through tangential literature and something catches your eye, gets lost when the process becomes too streamlined.

Stuart Kauffman's concept of the "adjacent possible" is useful here. Kauffman, a theoretical biologist, writing since 2002, described how complex systems evolve by exploring the space of possibilities one step away from the current state, rather than by random leaps. Each exploration opens up a new set of adjacent possibilities that didn't exist before. The act of looking around changes what's available to find.

When AI takes you directly to the answer, you skip the adjacent possible. You get the destination. You don't see the doors that would have opened along the way, because you were never in the hallway.

The portfolio argument

Here is where I think the standard "journey matters" argument fails, and where a different one, a colder one, might actually hold up.

The sentimental version says the journey matters because it's human, or because it builds character. I don't find this convincing. If you offered me a pill that gave me fluent Mandarin without the five years of study, I would take the pill. Most people would. The journey-as-virtue argument mostly comes from people who already completed their journey and are now narrativizing their suffering as growth.

The version I find harder to dismiss is a portfolio argument.

In finance, a diversified portfolio outperforms a concentrated one on a risk-adjusted basis, because the losers are uncorrelated with the winners. The portfolio's value sits in the correlation structure of the whole set, not any single position. When one position fails in a way you didn't predict, another position, one you had no particular reason to hold, turns out to be the hedge.

The things you find along the way are the uncorrelated positions. You didn't need them for the current objective, and they don't help with the current question. But they sit in your cognitive portfolio, and when a future question arrives that you couldn't have anticipated, the answer is sometimes in the thing you found when you weren't looking.

Nassim Taleb makes a version of this argument in Antifragile (2012). He calls it optionality: when you have many options, you don't need to be right about which one will pay off. You just need to avoid catastrophic downside and stay exposed to possible upside. Tinkering, which is Taleb's word for undirected exploration, outperforms design in complex systems precisely because the tinkerer is exposed to a wider set of outcomes. The designer optimizes for the expected case. The tinkerer is prepared for the unexpected one.

AI is a designer. It optimizes for the expected case. It is extremely good at this. What it doesn't do is generate the uncorrelated cognitive positions that hedge against cases you haven't imagined yet.

The honest half

But here is where the journey-defenders need to stop being romantic and start being honest.

Not everything you find along the way is a hidden gem waiting to pay off in ten months. Most of it is noise. Some of it is actively harmful. You can wander into information hazards that damage you, or rabbit holes that waste years on ideas that seemed promising and weren't. The thing you saw that you didn't need might inform a decision later. It also might just sit in your memory taking up space, or worse, giving you a false sense that you know more about a domain than you do.

The post-traumatic growth literature is instructive here. The popular version says that suffering leads to growth. A 2022 systematic review and meta-analysis in Clinical Psychology Review examined cognitive biases in perceptions of post-traumatic growth and found that defensiveness, memory bias, and social desirability all inflate self-reported growth. Self-report questionnaires cannot distinguish genuine personality change from coping-motivated positive reframing, and without pre-trauma baselines, most studies cannot confirm whether the reported growth is real or retrospective storytelling. The biases inflate but do not wholly fabricate the growth reports: some growth is real, some is the narrating mind making itself feel better, and the measurement tools can't reliably tell you which.

The survivorship bias in "growth from suffering" narratives is severe. The people who tell you their worst experiences made them stronger are, by definition, the people who survived those experiences well enough to tell you about them. The ones who were broken by the same kind of experience aren't writing the inspiring essay.

This doesn't mean the journey is worthless: its value sits in the statistical properties of the whole set of detours, not any individual one. Most individual detours are useless or harmful. The portfolio, if it's diverse enough, still outperforms the optimized path, because the few uncorrelated discoveries that turn out to matter couldn't have been acquired any other way.

Who tells the stories

There is an AJR song, "100 Bad Days," with the hook: a hundred bad days made a hundred good stories, a hundred good stories make me interesting at parties. The band has said it's a twist on "what doesn't kill you makes you stronger" — what doesn't kill you makes you more interesting instead.

I keep returning to this line because it contains the same quiet dishonesty as most journey-defense arguments.

Daniel Kahneman, in Thinking, Fast and Slow (2011), drew a distinction between the experiencing self, who lives through the moment, and the remembering self, who constructs the story afterward. They have different priorities: the experiencing self cares about what is happening right now, while the remembering self cares about the narrative arc, the shape of the story, whether there is a satisfying ending.

"100 bad days makes 100 good stories" is a remembering-self claim: it looks back at a hundred bad days and sees raw material for narrative, benefiting from the suffering. It gets stories. It gets identity. It gets to be interesting at parties.

But the experiencing self is the one who actually lived through those hundred bad days. The experiencing self didn't get stories. It got pain, boredom, confusion, fear, whatever the hundred bad days actually felt like from the inside while they were happening. The experiencing self doesn't get a vote in the retrospective evaluation. It already paid the bill by the time the remembering self starts writing the review.

This is the structure underneath most "the journey matters" arguments. The narrating mind looks back at the wandering, the mistakes, the accidental discoveries, the wasted time, and repackages all of it as a coherent story about how everything contributed to who you are now. The narrating mind benefits from the journey; it's not obvious the person who lived through it benefited to the same degree. The stories are real. The question is whether the stories are worth what they cost, and whether the person telling them is the same person who paid.

What are you actually optimizing for?

If you're optimizing for the outcome, the answer you need, the skill you want, the destination, AI wins. It's not close. The optimized path is faster and more reliable.

If you're optimizing for the architecture of the organism, the question changes. The architecture is not any single skill or any single answer. It's the substrate from which skills and answers get generated. It's the soil, not the plant. The soil is made of decomposed inputs, most of which were not useful in themselves: the paper you read by mistake, the argument you overheard, the wrong turn that showed you a place you had no reason to see. None of these are the plant. All of them are the soil. And the soil determines what can grow next.

AI gives you the plant, directly and on demand. What it doesn't give you is the soil.

Whether that matters depends on what kind of life you're building. If the problems you face are predictable, the optimized path is correct. You don't need the soil. You need the plant, fast.

If the problems change, if the valuable skill in five years is one you can't name today, if the insight that matters comes from a connection between two things you had no reason to put together, then the soil is the whole game. And the soil only comes from the journey, including the parts that were useless and painful.

I don't know how to resolve this cleanly, and I don't think it resolves. The experiencing self would prefer not to suffer through a hundred bad days. The remembering self would prefer to have the stories. The portfolio needs the uncorrelated positions, and uncorrelated positions, by definition, include things you wouldn't have chosen on purpose.

The question "is the journey worth it?" might be the wrong question. The better one might be: worth it to whom? The person living through it, or the person who will later narrate it? The organism solving today's problem, or the substrate preparing for tomorrow's?

I don't think they're the same person. And I don't think either one gets to answer for the other.