There’s a scene in Good Will Hunting that I keep thinking about as the conversation around AI gets louder.

Robin Williams and Matt Damon are sitting on a park bench. Damon’s character, Will, is a genius with an extraordinary memory and a habit of using everything he knows to keep everyone else at a distance. He’s already taken a shot at his therapist, Sean, confidently dissecting the man’s life based on a painting hanging in his office.

Sean has had some time to think about that.

He tells Will that he could probably recite everything ever written about Michelangelo. The work, the politics, the relationship with the pope. But then he asks whether Will knows what it smells like inside the Sistine Chapel.

Will has never been there.

The speech moves through love, war, and loss before Sean turns Will’s own logic back on him: Would reading Oliver Twist give Sean any real understanding of what Will’s childhood as an orphan had been like?

That’s where the scene gets past the easy lesson about book smarts. Will has lived through things that no amount of someone else’s research could adequately explain. Yet he’s comfortable assuming his research explains everyone else.

I think about that scene now because we’ve built tools that can give almost anyone an impressive command of almost any subject in a few minutes. And I’m increasingly interested in what happens when that command starts to feel like understanding.

We’ve made it very easy to sound like we’ve been there

I use AI constantly. I use it to research, organize ideas, challenge my thinking, and get work done that would have taken considerably longer a few years ago. I’m excited about what it makes possible.

I’m also aware of how quickly it can make me feel prepared.

Give me twenty minutes and I can have an industry overview, a competitive analysis, a customer profile, and a reasonably convincing vocabulary for a business I knew almost nothing about over breakfast. That’s useful. It also requires me to remember that breakfast was twenty minutes ago.

The document can look remarkably similar to something produced by a person who has spent fifteen years in that industry. The difference becomes apparent when something happens that neither of us thought to put in the document.

A customer hesitates for a reason that never appeared in the research. An employee agrees to a plan you can tell they don’t believe in. A supplier makes a promise that sounds perfectly reasonable unless you’ve heard that particular promise before.

Those moments draw on something accumulated through participation: noticing, making a call, getting it wrong, and having to deal with what happens next.

In my last article, I used the example of two people offering to build a deck. One has AI and tutorials; the other has twenty years of experience. My concern was what happens to the next generation of experienced people if we eliminate the early work they used to learn on.

But there’s another question in there for those of us who already have some experience.

How current is it?

Experience has to keep happening

It’s tempting to treat experience as an asset we’ve already acquired. We did the work, learned the lessons, earned the scar tissue. Now we can apply that wisdom to whatever comes next.

Some of that holds up. Some of it needs updating. And some of it was a conclusion we drew from one unpleasant Tuesday in 2009 that we’ve somehow promoted into a universal law.

Having been there matters. So does going back.

A founder who used to know every customer can eventually experience the entire customer base through a dashboard. An executive can become fluent in the presentation version of a company while losing touch with what it feels like to work there. A marketer can spend all day reviewing research about people whose actual lives they rarely encounter.

AI can make those reports more comprehensive and those presentations more persuasive. It can also make the distance comfortable enough that we stop noticing it.

That’s the part I want us to pay attention to. We’re getting better and better representations of reality, delivered with less and less need to encounter it ourselves.

At some point, we have to leave the summary.

The world gets to disagree with you

I want to make a case for actual, physical participation in things.

Go to the store where the product is sold. Watch someone try to use what you built. Sit across from the person who is disappointed with your work. Make something with your hands that refuses to cooperate with your intentions.

There is something useful about encountering a problem you can’t resolve by asking for another version.

The screw is stripped. The room is uncomfortable. The customer is standing right there. Whatever happens next requires your participation, and you don’t get to quietly regenerate the customer until they become more aligned with your strategy.

Of course real relationships, work, and consequences happen online. Most of us spend a substantial part of our lives there. But physical presence gives us information we often edit out of our digital lives: the awkward pause, the effort involved, the conditions around the task, the difference between someone saying they understand and watching them try.

It also reminds us that people have lives beyond the categories we assign them.

Someone described in a deck as a “price-sensitive consumer” might be a parent standing in an aisle doing math they wish they didn’t have to do. You can know that intellectually. Spending time around the person can change the attention you give it.

That change in attention matters. It affects what you ask, what you build, and what you’re willing to accept as a good answer.

Getting better tools should make us more curious

I don’t think the answer is to romanticize doing everything the hard way. There’s no particular virtue in spending six hours on something a tool can help you do in twenty minutes.

But I do think we should care about what we do with the time we get back.

We can fill it with more output. We can also use some of it to get closer to the people and situations that give the output meaning.

That might mean joining a sales call instead of only reading the recap. Spending an afternoon with the team responsible for delivering the plan. Taking on something unfamiliar where our professional competence doesn’t automatically transfer and we have to be a beginner again.

It also means paying attention afterward. Experience doesn’t automatically produce wisdom; sometimes it just produces a person who has been making the same mistake for thirty years and now insists it’s best practice.

We have to examine what happened, hear the feedback, and let it change our minds.

That’s an investment worth protecting alongside whatever we’re spending on the latest tools. The next model release may improve what we can produce. An afternoon in the right room may change our understanding of what’s worth producing.

Back to the bench

What stays with me about Sean’s speech is that he’s still interested in Will.

He’s heard the impressive answers. He wants to know the person who’s been using them as cover. And getting there requires Will to risk participating in a conversation he can’t control simply by knowing more than the other guy.

There’s something in that for the rest of us.

AI can help us prepare for an extraordinary range of experiences. It can give us useful context, help us ask better questions, and help us make sense of what happened afterward. I want all of that.

But we need to be careful about the confidence that comes with a thorough answer. We can ask follow-up questions, challenge the reasoning, request another perspective, and walk away feeling like we’ve covered the subject. All of that helps. It still leaves us with an understanding that hasn’t had to survive outside the conversation.

That’s where showing up matters. We need to hear what the customer actually says, see whether the plan works, and stay involved when something happens that wasn’t in the research. Even years of experience need that contact with reality to remain useful.

Sean could have listened to Will talk about Michelangelo all afternoon. He wasn’t questioning how much Will knew. He was asking whether Will understood how much he was still missing.

That’s a question worth carrying into our own use of AI. Before we turn a convincing answer into a confident decision, we should ask: What haven’t I seen? Who haven’t I listened to? What would I only discover by actually doing this?

A good prompt can help us get ready.  We should pay at least as much attention to the last time we went out and learned something our prompts didn’t know to ask.