The Free Lunch Test
Forty pet dogs watched one person feed a fellow dog and another person refuse to feed it, then met both humans face to face. They showed no preference for the generous one. Not a slight lean, not a delay before deciding. Nothing measurable at all.
The study, led by Hoi-Lam Jim at Kyoto University with colleagues Kadisha Belfiore, Eva B. Martinelli, Mayte MartÃnez, Friederike Range, and Sarah Marshall-Pescini, ran two versions of the setup. In the eavesdropping condition, a test dog watched a demonstrator dog get fed by one stranger and turned away by another. In the direct condition, the test dog was the one being fed or refused. Afterward, researchers tracked which person each dog approached first, whether it jumped up, and how close it stayed. The paper, published in Animal Cognition on June 28, 2025, covered dogs across three age groups: young, adult, and senior. None of them, at any age, picked the generous human over the selfish one at a rate beyond chance.
Why This Should Have Worked (and Didn’t)
To pass this test, a dog needs a specific chain of reasoning. It has to notice an act, feeding or refusing, and attach that act to a particular person rather than to the moment itself. Then it has to carry that memory forward and use it to choose between two people later. That whole chain is called reputation tracking, and it is not a trivial skill to have. Chimpanzees have it. Earlier research cited in the Kyoto paper found that chimps discriminate generous humans from selfish ones after watching a single interaction, and again after experiencing it directly. The dog study was built to mirror that same logic almost exactly, with two humans, one generous act, one selfish act, and then a choice.
That design is what makes the flat result worth pausing on. Researchers were not fishing with a vague setup that happened to fail. They aimed at a known target, using a method that already works on a different, comparably clever species, and it still came up empty.
It’s Not Because They Don’t Know Humans
There is an obvious rescue for the null result, and the researchers had already ruled it out before they started. Earlier work at the Wolf Science Center in Austria found that pack-living dogs and wolves also failed to form reputations of individual humans. But those animals live under fairly controlled conditions, with limited, structured contact with people, so their failure came with a built-in excuse: maybe they simply hadn’t clocked enough humans to learn who’s who.
The 40 dogs in the 2025 study are ordinary household pets with years of daily human contact, some of them for a decade or more. If exposure were the missing ingredient, this is exactly the population that should have had it. The result held steady anyway. That closes the loophole. It pushes the question from “did they get enough practice” to something harder: can dogs do this at all, under any amount of practice?
The Myth: Dogs Don’t Read Your Character, They Read Your Hands
People routinely treat a dog’s approach or avoidance as a character reference. Gravitate toward someone, and dog owners take it as a small vote of trust. Flinch from someone, and the same owners start wondering what that person is hiding. This dataset contradicts that assumption directly. Dogs that watched a stranger withhold food from another dog, or experienced it themselves, did not later avoid that stranger in favor of the generous one.
None of this means dogs are bad at reading people. Dogs are excellent at following a pointed finger, a glance, a shift in tone, in the moment. What this study shows is that turning one observed act into a lasting judgment about a person’s character looks like a separate skill entirely, and it did not show up here. The chimpanzee comparison matters for exactly this reason: it proves the ability exists in at least one other cooperative, socially sophisticated species. Dogs, for all their history alongside us, did not demonstrate it under these conditions.
What Nobody Knows Yet
Hoi-Lam Jim, now on the faculty at Kyoto University, put it plainly: reputation formation may be more complex than previously thought, even in an animal that cooperates with humans as closely as the dog does. What the study cannot settle is whether that complexity means dogs lack the capacity outright, or whether the test itself was too blunt to catch something real. A first approach, a jump, a few inches of proximity, these are coarse measures. A dog could be tracking a person’s history in some entirely different way, one that never shows up as “which stranger do I walk toward first.”
The next step, per the research team, is to look at dogs whose lives actually depend on reading a specific handler or a specific stranger correctly: free-ranging dogs, service dogs, police dogs. If reputation tracking exists in this species, that is where the stakes might finally run high enough to bring it into view. One image circulating with the coverage shows Jim’s own dog Jasper posing with his friend Sahibu, a personal photo the team used to illustrate the story, not a data point from the experiment. For now, the honest answer is that nobody knows.