I was eight or nine, living in a small town in Indiana, when my best friend Joey and his parents ended up with a last-minute ticket to a Cincinnati Reds game. They asked if I wanted to come. A mini road trip. My first Major League game. Another great memory with my best friend.

The only problem: Cincinnati was a couple of hours away, it was a weeknight, and I'd need my parents' permission. So I thought about how they'd react, and I concluded they'd say no. I declined.

Joey and his dad had to leave, so I headed home. I found my mom reading the paper and told her about the invitation, and the responsible decision I'd made. I was expecting praise for making sure a growing kid got a good night's sleep.

She put the paper down. "Well, of course you should go."

Panic. Could I undo it? My mom went to the phone to call Joey's mom while I ran to the window. We found out at the same time, in our own ways, that it was too late. Joey's mom confirmed on the phone what I could see from the window: a baby blue Mustang pulling out of the neighborhood, Joey's tow-headed blond hair in the rear window.

I was crushed. Didn't I have enough empathy with my own parents to predict what they'd say?

I've never forgotten that day. And it came back to me recently because of something an AI assistant said while Maria and I were recording.

Maria Ross and I just released the sixth and final episode of our podcast series, What Does It Mean to Be Human? It's the question we've been circling all season: what makes us human, now that machines can do so much of what we used to think was ours alone.

Somewhere in the middle of that episode I mentioned a New York Times column by the Princeton sociologist Zeynep Tufekci, on why she isn't worried about AI taking over. Her argument: these systems aren't reasoning machines. They're plausibility engines. They produce the most likely next thing.

Near the end of the episode, Maria read us something Claude, the AI assistant she uses, had written about itself. She'd asked whether it could imagine something that doesn't exist yet, and the answer turned into a question about whether it has any experience at all.

Claude said, carefully, that it clearly processes information and produces sensible answers. Whether there's anything felt behind that, any experience of its own, it said it genuinely didn't know. Then it said the part I keep thinking about. It can report "I found that interesting," but it can't tell whether that sentence describes something actually happening inside it, or is just the most statistically fitting thing to say at that point in the conversation. It can't see its own workings well enough to check.

A plausibility engine, pointed at itself.

(For the record: Maria has told her Claude to push back on her and not to overclaim, so some of that humility is hers. But the idea holds.)

Because when I imagine what you're thinking, when I picture how you feel, when I decide what you'd want before I've asked you, I'm doing the same thing. I'm running a plausibility engine on you. I generate the most likely version of you out of the material I already have, which comes mostly from me. It comes back fluent. It comes back confident. And most of the time, I never check it against the source.

That's exactly what nine-year-old me did.

Notice the question I asked at the end of that story. Didn't I have enough empathy to predict what my parents would say? For decades, that's how I understood it: I failed to read them.

That was never the mistake. The mistake was trying to predict at all. My mom was four or five houses down the street, a two-minute walk. Instead of going home and asking, I ran a plausibility engine on her, got a confident answer, acted on it, and didn't ask until it was too late.

Here's why that matters beyond one missed ballgame. Most of us carry around a version of empathy built on the old line about walking a mile in someone else's shoes. Imagine how they feel. Imagine what they're thinking.

There are two kinds of empathy. Emotional empathy is feeling what someone else is feeling. Cognitive empathy is understanding what they're thinking, and why. Both are real, and both are useful. But when you do either one by imagining, the material comes from the same place. You reach into your own library, your own emotions, your own experiences, what you already know about the other person, and you pull out the closest thing you've got. Then you project it onto them.

That tells you a lot about you. It tells you less than you think about them. There's only one other place the material can come from: the other person. You can't imagine your way there. You can only ask.

In 2018, three researchers, Tal Eyal, Mary Steffel, and Nicholas Epley, ran 25 experiments with about 2,600 people to test whether being told to take someone else's perspective actually made you more accurate about what that person was thinking or feeling. Reading emotions from faces. Predicting a romantic partner's preferences. Spotting fake smiles. Spotting lies.

It didn't work. Being told to imagine yourself in someone's position produced no consistent gain in accuracy. If anything, it made people slightly less accurate, and occasionally more confident. What did work was the plain thing: asking the person.

It's worth being precise about what the study tested, because it's easy to overread. It tested guessing. What emotion is on that face. Which movie would your partner pick. Is that smile real. Guessing has a right answer, so it can be graded. Imagining failed the graded part. And notice what the researchers used as the answer key: they asked the person. Perspective, how somebody actually sees a thing and why, doesn't have an answer key. It's not something you guess. It's something they give you.

The second study is the one I want to spend more time on. In 2022, Michael Kardas, Amit Kumar, and Nicholas Epley published a paper called "Overly Shallow?" They put strangers in conversation and asked them beforehand to predict how it would go, including what it would be like if the conversation went deep. Not weather and traffic. What are you grateful for? When did you last cry?

People predicted it would be awkward, and that the other person wouldn't much care. They specifically predicted a deep conversation would be more awkward than a shallow one.

Then they had the conversations, and nearly every prediction was wrong. The conversations went better than expected, the deep ones especially. People felt more connected and happier than they thought they would. The researchers' conclusion: we systematically underestimate how much other people care about, and are interested in, what we'd tell them.

Here's the part that matters. When the researchers corrected that expectation, when they told people these conversations tend to go well, people chose deeper topics. The belief wasn't just sitting there. It was steering what they were willing to say.

I'll be honest about the limits. Most of these conversations were between strangers. Extending the finding to your mother, your spouse, or your uncle with the political opinions you avoid at Thanksgiving is a step the study doesn't take. But I've sat in enough strangers' kitchens to tell you what I've seen, and it tracks. People want to be seen and heard. What stops them is the fear that nobody really wants to know. Most people do.

Put the two studies together.

We don't fail at empathy because we can't read minds. Nobody can. We fail because the people around us have quietly decided we don't want the real answer.

The research doesn't go this far, but I will: most of us never give them a reason to think otherwise. Every conversation we keep on the surface confirms what they already suspected. A good question isn't just a technique. It's evidence. It's you handing someone proof that the assumption they were carrying about you, that you're not interested, that you don't want to go deep, was wrong.

Which brings me back to prediction. I often hear empathy described as a forecasting skill. Read the room. Anticipate what they'll do. Get ahead of them.

I think that's a false equivalence, and this AI moment makes it obvious. A plausibility engine is extraordinarily good at predicting what comes next based on everything that came before. It is very, very good at the thing that isn't empathy. Empathy was never about guessing what someone will do. It's about understanding their feelings and perspective well enough that we can get somewhere better together: so I can support you, give you what you actually need, or work with you toward a better outcome. No model, no algorithm, and no amount of imagining has access to what you would tell me if I actually asked.

And that's the real reason we need it. Not so we feel more for each other, but because every decision I make about you without asking is a decision made about someone who doesn't exist. It's like making a synthetic twin of you.

If you work in insights or marketing, that phrase should land somewhere specific. My industry is in the middle of a debate about synthetic respondents, AI personas built to answer in place of real consumers. They can be useful for some things. But a synthetic twin is a plausibility engine with a name tag. It can tell you what someone like your customer would most likely say. It can't tell you what your actual customer would tell you if you asked.

That night in Indiana, I built a synthetic version of my parents, decided for them instead of asking, and missed the game. The cost of skipping the ask is never that you cared less. It's that you acted on the wrong person.

That's what the 5 Steps to Empathy are for: Dismantle Judgment, Ask Good Questions, Actively Listen, Integrate into Understanding, Use Solution Imagination. I've described them as a method for years. I think they're more than that. They're a workaround for a bias, a specific mistake we all make about a conversation before we have it: that people won't want to go deep, won't care, so we won't share, so we won't ask. Dismantle Judgment is you noticing the plausibility engine running and turning it off. Ask Good Questions and Actively Listen are you going to the source.

Earlier this year I was preparing for an interview with one of the participants in Navigating to a New Normal, the longitudinal study I started in March 2020 that Dig Insights runs today. Her politics and outlook are much more conservative than mine. I planned to ask about the Iran war and ICE activity in Minneapolis, because they were topical and because documenting how people react to the news is part of the study.

I'll be honest: I was dreading it. Thanks to the plausibility engine, I thought I could predict how she'd respond, and her answer was going to collide with my own beliefs. I didn't want to navigate that on a recorded call.

When we got there, I asked for her perspective. I thought I knew what she was going to say. This is someone I've talked with regularly for more than six years, and everything I knew about her said she'd support the administration's actions.

Instead, she disagreed with the administration. On both issues.

I want to be clear about the lesson. It isn't that she agreed with me. It's that after six years of real conversations with her, I still got her wrong the moment I stopped asking and started predicting. If I can do that with someone I've interviewed for six years, you can do it with anyone.

That's what asking gets you. The actual person.

Two things, built on the second and third of the 5 Steps to Empathy, Ask Good Questions and Actively Listen.

Go one question deeper. Pick one conversation this week that would normally stay on the surface. A catch-up with a colleague or your boss. A neighbor. Someone you see every week and know almost nothing about, maybe at your local drugstore or grocery store. Ask one question past where you'd usually stop. Not a new topic. Same topic, one layer down. All it takes is "Tell me more about that." Then notice two things: how awkward you expected it to be, and how awkward it actually was. The gap between those two is the lesson.

Name the assumption, then ask. This one is for anyone who makes decisions on someone else's behalf, which is most of us. Think of a decision you're about to make for a customer, your team, or someone in your family. Write down, in one sentence, what you're assuming they want. Then go ask them for their perspective. The rule: you're not checking whether you were right. You're collecting what you didn't have, so the decision is better informed.

Nine-year-old me could have walked home and asked. It would have taken two minutes, and the answer would have been yes.

I'd like to hear how it goes. Reply to this via email or leave a comment.

Listen to the finale. Maria and I pull all six episodes together, get into the philosophers Silicon Valley is now hiring, whether a machine can want anything, and why we keep calling our best human qualities weaknesses: What Does It Mean to Be Human? Ep 6: The Whole Messy Thing We Want to Protect. If you're new to the series, it's six episodes, and this one is a good place to start before going back to the first.

And on Spotify…

Read the sources. Zeynep Tufekci's column in The New York Times (gift link). Eyal, Steffel, and Epley, "Perspective Mistaking," Journal of Personality and Social Psychology, 2018 (link). Kardas, Kumar, and Epley, "Overly Shallow?," Journal of Personality and Social Psychology, 2022 (free copy from the APA).

Ocean City, Maryland. On October 6 at 8:30am ET I'm speaking at MD SHRM 2026 on The Neglected Superpower: How to Reduce Turnover, Strengthen Workplace Culture and Enhance Relationships. It's the workplace version of everything above: people leave when they've concluded nobody wants their real answer, and most managers never give them evidence otherwise. If you'll be there, come say hello. Conference details.

Last week I was in Chicago for the Insights Association's Corporate Researchers Conference, there on behalf of Dig Insights and as an Insights Association Laureate. It was three days of good conversations with clients old and new, and with colleagues and friends I've known across a lot of years in this industry. I also had the honor of introducing the Laureate class of 2026.

And I got to introduce Rishad Tobaccowala before his workshop. Rishad wrote the endorsement that's on the cover of Tell Me More About That. Before this week, we'd only ever talked once, on a Zoom call when I asked if he'd read the manuscript. He did, and his words ended up on the cover. I'd tried to meet him on a couple of trips to Chicago, but the timing never worked. When I saw he'd be at CRC, I emailed him, and he said he was looking forward to it. A day or two later, I was asked to introduce him.

Getting to stand up and acknowledge him publicly, as a way of saying thank you, felt like a full-circle moment.

When we finally sat down, he asked how I was doing and what I've been working on. We talked about my aging parents project: how population growth is slowing, and how people's cultural, racial, and ethnic backgrounds shape the decisions they make about how to age. It was a short conversation before his workshop. He saw the need for the work right away, which makes sense coming from someone whose latest book, Rethinking Work, is partly about how demographic change is reshaping work.

Stay curious. Ask good questions. Let empathy follow.

Rob