Building long-term empathyA large-scale comparison of traditional and virtual reality perspective-taking

Let's start with a big idea that's been bouncing around tech and psychology for almost a decade: can a headset make you kinder? When people called virtual reality, or VR, the "ultimate empathy machine," they were betting that stepping into someone else's shoes—literally looking through their eyes and moving as their body—would change not just how we feel in the moment but how we act weeks later. Herrera, Bailenson, Weisz, Ogle, and Zaki took that claim and put it under a microscope. They focused on homelessness, a tough, emotionally charged outgroup where empathy isn't abstract; it might mean signing a petition, backing a housing measure, or donating money. They didn't just pit VR against a written story. They layered in a desktop version and even an information-only condition to ask a sharper question: what's the power of the medium over and above the message? The first study set the tone: a head-to-head, eight-week comparison between a virtual reality perspective-taking experience—let's call it VRPT—and a traditional narrative-based task, or NPT. The mechanics were simple and controlled. People were randomly assigned, they spent about fifteen minutes with the intervention, then reported how they felt and what they thought right away, again two weeks later, and finally eight weeks out. The project recruited four hundred fifty-two people and analyzed data from four hundred thirty-nine after removing incomplete cases. This was not a narrow lab sample: ages ranged from teenagers to folks in their eighties, with a mean around thirty. That diversity mattered because VR novelty and prior exposure can shape how people respond. Immediately after that first encounter, VR did what its champions would hope. Participants who tried VRPT felt more empathy in the moment and more personal distress than those who read and imagined the narrative. However, self–other overlap—the sense of "we are in this together"—didn't budge between the two. Crucially, those elevated feelings didn't stay elevated. Over the next two months, empathy and distress settled to about the same level in VRPT and NPT. Where the curves pulled apart was in attitude and action. As weeks went by, people who had been in VR grew more positive toward the homeless, and they were less likely to dehumanize them—judging them, for example, as "less evolved"—than those who did the traditional exercise. They also behaved differently. A larger share of the VRPT group signed a petition supporting housing initiatives; one test of that difference yielded a t-value just over three and a half with a vanishingly small p-value, a strong signal that it wasn't a fluke. Money told a different story. Donations hovered in the seven to eight dollar range across the board, and because about seventy percent of people gave the maximum available, the measure simply couldn't distinguish much between conditions. There were subtler ripples too. Two weeks after the intervention, VRPT participants wrote letters to elected officials that used more first-person plural pronouns—"we," "our," "us." That's a tiny linguistic clue, but it's revealing. Instead of describing the problem at arm's length, they were framing it as something we, collectively, could act on. Four weeks out, they also voiced more support for a local affordable housing measure, often referred to as Measure B. By eight weeks, their attitudes toward people experiencing homelessness remained reliably more positive than those who'd done the narrative task. So the immediate heat of empathic feeling cooled, but a different kind of warmth stuck around: a steadier sense of concern and a willingness to back it up. But was that VR doing the work, or the specific story it told? To pry those apart, the second study added two more arms. One was a desktop version of the very same interactive homeless narrative—no headset, just a screen. The other was an information-only condition, a packet of facts about homelessness. Now there were four routes through the material: Information, NPT, Desktop, and VRPT. If immersion was the magic, VR should feel more "there" than the desktop, which in turn should feel more "there" than the written narrative. That's exactly what the presence checks showed. Coders read people's open-ended descriptions and rated whether they reported feeling transported into the scene. Those ratings were remarkably consistent—a Cohen's kappa of about 0.93—and the pattern was clear: significantly more VR participants felt present than desktop users, and desktop beat narrative too. On the self-report measures, the picture got more nuanced. Compared to facts alone, any perspective-taking—traditional, desktop, or VR—made people feel closer to the homeless, boosted their empathy, and nudged up personal distress. However, the three perspective-taking formats were surprisingly similar to one another on those immediate feelings. The headset didn't add a reliable extra dollop of empathy over the desktop or the written task in this study. There was a hint that social presence—the sense of interacting with another mind—trended higher in VR than on desktop, but it didn't reach the usual statistical cutoff. Dehumanization measures moved only modestly, with no clean split between conditions right after the intervention. Action, again, told a sharper story. When it came to backing a specific policy—Proposition A—support didn't differ between information and the perspective-taking conditions overall. However, when researchers compared mediated perspective-taking—the desktop and VR—against the traditional narrative, the mediated formats led to higher support, and VRPT outpaced desktop on that same policy question. The petition numbers were even more decisive. Out of one hundred fifteen people in VRPT, ninety-eight signed. Desktop landed in the middle, with seventy-five of one hundred thirteen. The traditional narrative was lower, sixty-six of one hundred four. Those contrasts held up statistically; the difference between VRPT and desktop, for example, came in with a p-value of about 0.001. If you're keeping score, the headset pushed more people to put their name down in public support than either the same story on a screen or the classic "imagine this person" exercise. Donations, again, didn't differ. The same ceiling effect that flattened study one reappeared here, so money couldn't tell us much about modality. Follow those people out in time, and the pattern from the first study resurfaces. Self-reported empathy and distress leveled off; by the end of the eight-week window, VRPT didn't leave people feeling consistently "more empathic" than the traditional or desktop tasks. Once again, attitudes held up better in VR. They stayed more positive toward the homeless, and they were less prone to slide toward dehumanizing language. In the correlations that knit these measures together, you can see the contours of the experience. Feeling closer on the self–other overlap scale linked up with reporting more empathy, though the link was modest. Empathy and distress rose and fell together, which makes sense: taking on someone else's pain can be moving and aversive at once. A greater sense of social presence traveled alongside more empathy too. So what do we make of this? One way to think about it is that VR seems to work like a powerful ignition. It sparks a vivid, affect-laden first-person moment, but those feelings dissipate. What lasts are not the fireworks but the path the spark sets you on. You come away with a slightly different stance toward the group—warmer attitudes and a broader sense of "we"—and that stance makes it a little easier to act. Not to give more money in a lab setting where almost everyone maxes out, but to do the everyday civic things people actually do: sign a petition, say yes to a housing proposition, write to an official. Now, a few caveats matter, and the authors are frank about them. Many of the self-report effects are small. Presence was coded qualitatively, not on a standard scale, even though the coding was very reliable. Study two didn't include a pure no-intervention control, which limits some comparisons. Baseline attitudes toward the homeless weren't measured before the interventions, so while the longitudinal trends are clear, we can't calculate individual change from a true zero point. Novelty is the elephant in the headset. Lots of participants were new to VR, and first-time awe can add emotional voltage that isn't about the content at all. None of these limitations overturn the pattern, but they do put boundaries around how far we should extrapolate. It's also worth noticing where the medium didn't deliver. People didn't feel more self–other overlap in VR than in the other perspective-taking formats right after the experience. Dehumanization scores didn't split cleanly at the start. Social presence, by a standard rating, only trended higher in VR than on desktop. And money—arguably the most concrete behavior—wasn't sensitive to the differences the researchers were probing. In other words, VR is not a general-purpose empathy amplifier. It's a selective tool, and its edge shows up most clearly when the outcome is a policy-relevant, low-friction action. Across both studies, though, the arc is consistent. VR perspective-taking produces stronger immediate affect than a traditional narrative, but those feelings fade. What endures are attitudes and actions: more favorable views toward the homeless and more signatures and policy support in the weeks that follow. In the second study, simply making the perspective-taking mediated—shifting from a written exercise to an interactive scene—helped. But the fully immersive version helped most. When the same story moved from a monitor to a headset, more people crossed the line from good intentions to a public yes. If you're building tools to change minds and behavior, that's encouraging. And it's also a nudge to be precise. The medium matters, but so does what you ask people to do afterward. Petition signing showed crisp differences; donations didn't. That suggests standardizing behavioral endpoints that match the kind of real-world actions we want to see—especially those that don't max out immediately—will be key in future work. Where does this leave the "ultimate empathy machine" claim? Closer to earth and more interesting for it. VR isn't a magic wand for feelings. It looks more like a way to make an issue feel lived-in enough that when the moment to act arrives—on a petition sheet or in a ballot measure—you lean in. Herrera and colleagues close by calling for exactly the kind of studies that would tighten the bolts: separating message from medium more cleanly, measuring pre-intervention baselines, mixing in implicit measures that are hard to game, and tracking durability across different social targets, not just homelessness. Given the small but consistent edge for immersive experiences and the clear link to policy-relevant actions, that's a test worth running. In the meantime, the headset has earned a subtler title: not the ultimate empathy machine, but a reliable starter motor for prosocial momentum.