A 2026 Science paper: insight in an insect brain
In 1917, German psychologist Wolfgang Köhler watched a chimpanzee named Sultan stack boxes to reach a banana on Sumatra, coining the term Einsicht — insight. For a century, that club of animals showing insight (crows, dolphins, elephants) shared one trait: large brains. A June 2026 paper in *Science* added a new member whose brain is 1/100,000th the size of a chimp's: a bumblebee.
How small is a bumblebee brain?
Humans have ~86 billion neurons, chimps ~28 billion, mice ~70 million. *Bombus terrestris* has about one million — a brain the size of a sesame seed — which must still handle flight navigation, flower recognition, social communication, and thermoregulation.
Loukola's lab at the University of Oulu had previously shown that bumblebees roll balls for fun (2022, meeting all five criteria of play) and can socially learn a skill too complex for any individual to invent alone (2024, *Nature*). But both findings had limits: play doesn't require thinking, and social learning has a demonstrator. The 2026 question: can a bumblebee face a novel problem alone and devise a solution?
The experiment: blue flower, ceiling, and a ball
Step 1 — Training. Bees learned two independent facts:
- Fact A: blue artificial flowers contain sugar water.
- Fact B: a small wooden ball is movable and harmless.
- Bhambore, A.A., Akmeşe, E.N., Häkkinen, E., Jussila, M.K., Kantola, J.-H., & Loukola, O.J. (2026). Spontaneous problem-solving in bumble bees. *Science*, 392(6802), 1046–1049. DOI: 10.1126/science.ady1618
- Köhler, W. (1917). *Intelligenzprüfungen an Menschenaffen*.
- Loukola, O.J. et al. (2022). Do bumble bees play? *Animal Behaviour*, 191, 1-8.
- Loukola, O.J. et al. (2024). Bumblebees socially learn behaviour too complex to innovate alone. *Nature*, 627, 572–578.
- University of Oulu press release, July 2, 2026.
No association was ever made between the two.
Step 2 — Challenge. The blue flower was placed under a transparent ceiling, out of reach. Only the ball was available.
Result: After a few exploratory passes, some bees performed an untrained action sequence:
> Push the ball directly beneath the flower, then climb on it and extend the proboscis to reach the sugar water.
"This is essentially the insect version of the 'box and banana' problem," said Loukola.
Ruling out the boring explanations
1. Chance: Successful bees' ball trajectories were significantly straighter and more directed than failed attempts. 2. Play by-product: Playful rolling is random and back-and-forth; solution-pushing was direct to a target — kinematically distinct patterns. 3. Gradual trial-and-error: Successful bees often got it right on their first few attempts, with no learning curve. 4. Visual tracking: The hardest to rule out — so researchers hid the flower. Bees still solved the task, pushing the ball to the correct location from memory of the flower's position.
"One moment the animal seems to be exploring without direction; the next, it executes a highly efficient sequence leading straight to the solution," said co-author Ece Nur Akmeşe. This is genuine goal-directed behavior: maintaining a goal in mind and organizing actions around it.
Combination, not creation
The task requires both prior facts — without "blue = food" or "ball = movable," the trick fails. The bee didn't create from nothing; it combined two independently learned experiences into a new path — structurally identical to Sultan combining "boxes stack" with "height reaches bananas."
The authors also draw a parallel to AI: the emergent abilities of large language models are likewise combinatorial — reassembling fragments of learned capability into new ones. As Backyard Brains' blog put it:
> "We're living in a moment when giant AI models take all the attention... but I often find myself equally impressed by models small enough to run on a phone or laptop. That's exactly the lesson bumblebees keep pushing in front of us."
Intelligence appears to be a function of organization, not hardware scale. The bee's million neurons form ~200 glomeruli with distinctive connectivity — a different architecture, not a miniature human brain.
Convergent solutions to intelligence
Köhler's insight and the bee's ball are behaviorally isomorphic but neurally different: cortex and prefrontal lobes in apes, mushroom bodies in insects. Like sharks and dolphins converging on streamlined bodies, evolution may have found convergent solutions to "combine existing information to solve novel problems." For AI, the implication is direct: we need not copy the human brain any more than we needed to copy bird wings to fly.
Loukola's closing attitude is instructive:
> "Knowing what your experiment cannot prove is part of the craft, not a retreat."
The bee proved: combine two experiences into a new plan. It did not prove: thinking like a human. Both are true. A sesame-sized brain can do far more than we assumed — intelligence doesn't care how big the hardware is, only whether the organization is right.
---
References: