What Everyone Heard, and What the Paper Actually Says
The version that travelled was simple: seventy percent of bumblebees spontaneously invented a tool, with nobody training them and nobody showing them anything. That version is false, and the paper says so plainly. The study — lead author Aniket Bhambore, senior author Olli Loukola, University of Oulu, published in Science on June 4, 2026, using Bombus terrestris, the buff-tailed bumblebee — ran three pretraining groups, and the pretraining turned out to be the whole ignition system.
Nineteen bees had met a ball as a movable object and, separately, met the blue artificial flower as food; once, the ball had been sitting on the flower, so they'd had to shove it aside to eat. Twenty-one bees got the flower and the reward but never saw a ball. Twenty got nothing at all. The first group beat both of the others enormously, at P below 0.001 twice over. The second and third groups were indistinguishable from each other: P = 0.972. Bees without the prior experience did no better than bees given nothing.
What nobody did was teach the solution. No bee was ever shown a ball being used to reach anything. The pretraining supplied two unrelated ingredients — balls move, blue ring is dinner — learned on separate days and never connected by the people who taught them.
The Ceiling, Not the Slippery Flower
A lot of the coverage said the flower had been made slippery so the bee couldn't land on it. Nothing was slippery. The blue ring, with sugar water in it, was fixed to the underside of a white acrylic ceiling, and that ceiling was simply too low: there isn't room in the gap for her to hover and reach up at the same time. It's geometry, not a prank.
The floor had four pits carved into it, exactly one of them directly under the flower, sized so a small Styrofoam ball drops in and stops rather than rolling away when something climbs on. The ball started at the centre of the arena, the flower's position was randomised so no bee could inherit an answer from the one before, and each bee was tested exactly once. No practice run, no second attempt — whatever she does, she does the first time she has ever seen the problem.
As for why a creature with wings solves it on foot: the practical answer is the low ceiling, and the second answer is cost. Flight is enormously expensive, and hovering is close to a sprint for her. If there's a way to eat that doesn't involve running the engine, she'll take it.
The Control That Failed — and Got Published Anyway
The hard question is how you tell "she put it there on purpose" from "she blundered into it," and the back half of the paper is a fight with that question. Round one put up a single barrier so the flower wasn't visible from where the ball started: sixteen of twenty-two bees succeeded, seventy-three percent — which is where the stray "seventy" in the headlines actually lives. The authors then argued against their own result, noting that as the ball nears the gap she might catch a glimpse of the flower, and conceding they could not fully rule out perceptual guidance.
Round two added three barriers, with seventy-seven bees split into treatment and control groups. It produced nothing: P = 0.368, and control bees with no flower in the arena at all pushed the ball to the same place at the same rate. In the authors' own words, they could not distinguish intentional placement from random exploratory behaviour. They published it anyway — and essentially none of the press coverage mentioned it.
Round three was the decisive one: two visually sealed compartments, with the flower genuinely invisible from the ball's starting position. Twenty-three of thirty bees moved the ball to the correct compartment, and sixteen of those went straight there. Then came the best check in the paper. Success was not predicted by how cleanly a bee pushed — trajectory straightforwardness predicted nothing. It was predicted only by how often she had looked toward the compartment where the flower was, before she moved anything. Predicted success rate 0.77, confidence interval 0.59–0.88, P = 0.006 against chance. That isn't a motor skill. That's memory doing work: holding a target in mind while she was somewhere else.
Insight-Like, Not Insight — and the Nine-Year Thread
The paper is careful in ways the coverage wasn't. Its word is "insight-like." The classic signature of insight — the aha moment, the transition from stuck to solved — was never observed; the camera angles wouldn't let the researchers track her gaze, so they don't claim it. Loukola's own line: "We are not claiming that bees think like humans." And the paper's phrase is "novel object manipulation," not tool use; the tool-use framing came from the editorial blurb and the press. It's defensible — she uses an object to reach something she otherwise couldn't — but it isn't the language the researchers chose.
This story also has a spine. On February 24, 2017, Science published a study from Lars Chittka's lab at Queen Mary University of London whose first author was a young researcher called Olli Loukola — the same man who is senior author on the 2026 paper. Nine years apart, same species, same ball. In that one, trained bees learned to move a small yellow ball to a target circle for sugar water, and untrained bees watched under four conditions: a live trained bee, a plastic model bee on a stick, a "ghost" where the ball moved on its own pulled by a magnet under the floor, and no demonstration at all. The ghost didn't work and neither did nothing; the live bee and the model bee both did. That's social learning.
The sharper detail is what the observers did with what they saw. The demonstrators had been trained to always move the ball furthest from the target — the long way round. The observers instead used the ball closest to the target, even when it was a different colour from the one they'd watched. They didn't copy the demonstration; they extracted the goal and discarded the rest, in a brain of roughly a million neurons. And in 2022, in Animal Behaviour, Chittka's lab asked "Do bumble bees play?" — logging 910 separate ball-rolling actions by forty-five bees, with thirty-seven of the forty-five going back to roll balls on at least one additional day after feeding, with nothing to gain. A conditioned place preference test showed the rolling was intrinsically rewarding: the bees later preferred the colour of the room where it had happened. One correction on that study too — males didn't play more; they rolled individual balls for longer and pushed them further, about 106 mm versus 40 mm.
The Electric Conversation Nobody Can See
All of this happened while the bee and the flower were already talking to each other electrically. A bumblebee in flight builds up a static charge and arrives at a flower carrying it. The number you've probably heard — two hundred volts — is wrong: what was measured, at Bristol in 2013 with fifty-one bumblebees flown into a Faraday pail, is charge, in picocoulombs. Ninety-four percent were positive, with a mean of about thirty-two picocoulombs. Pollen really does leap the gap before contact, but the demonstrated gap is on the order of half a millimetre — sub-millimetre, not centimetres, whatever the retellings say.
The Bristol team built fake flowers: 35 mm steel disks in purple epoxy, half carrying a small charge and sugar water, the other half visually identical, grounded, and filled with bitter quinine. Over their final ten visits, eleven bees hit eighty-one percent accuracy on a cue they cannot see, smell or hear. Then came the control that could have killed the result — ground both kinds of flower, remove the electrical difference, change nothing else. Accuracy fell to fifty-four percent. Chance. The field was the only thing they were reading. A weaker bias failed to work at all, which is how a real detection threshold shows up. And in 2016 the same lab found the sensor: both the antennae and the body hairs deflect in an electric field, but the antennae produce no neural response. It's hair deflection that fires neurons. She's covered in instruments that look like fluff.
One more correction, and it's the one that hurts. When a bee lands, the flower's electrical potential changes — electrodes in petunia stems recorded a shift of about twenty-five millivolts lasting roughly a hundred seconds, often beginning before she physically touches the plant. The popular next step is that this acts as a "closed" sign telling the next bee the nectar is gone. That has not been demonstrated. The paper says only that the change potentially carries information for other visiting pollinators; no experiment has shown a later bee reading the altered field and skipping the flower. Reported, plausible, not settled. The real punchline everyone walks past is speed: every other signal a flower gives about whether it's been visited — scent marks, colour shifts, humidity — changes over minutes to hours. This one changes in seconds. It's the only channel in the meadow that updates in real time.