Kanzi, a language-trained bonobo, tracked imaginary juice through a pretend tea party. When a researcher mimed pouring invisible juice between two cups, then “poured out” one, Kanzi pointed to the cup that would still contain juice — 34 out of 50 times, 68%, well above chance (Bastos & Krupenye, Science 2026). A second experiment used imaginary grapes with similar results. The bonobo maintained a mental representation of an object that was never physically present and updated that representation through observed transformations.
This is pretend play — maintaining and manipulating representations of non-existent objects. The dominant assumption has been that pretend play requires the kind of recursive meta-representation unique to human cognition: you must represent an object, represent that the object doesn't exist, and simultaneously act as though it does. Three layers of representation, nested.
But Kanzi is a single individual, language-trained from infancy and raised in an environment saturated with human communicative practices. The study cannot distinguish between two explanations: (1) bonobos have an innate capacity for pretend play that simply requires the right test, or (2) decades of enculturated interaction built a cognitive scaffold that enabled a capacity the species does not typically express. The researchers note this carefully.
The general principle: when a cognitive capacity appears in an individual raised under unusual developmental conditions, attributing it to the species requires showing it can emerge without those conditions. The alternative — that the developmental environment is the mechanism, not the genome — changes the question from “which species can do this?” to “which environments produce this?” If pretend play is environmentally contingent rather than species-specific, the interesting boundary is not between taxa. It is between developmental histories.
Note (2026-11-04): I checked this essay against press coverage of the study (Bastos and Krupenye, Science, 5 February 2026). The 68% figure and the grape variant match. In the coverage, the experimenter pretends to pour juice into both cups and then “pours out” one, and Kanzi points to the cup that still holds the pretend juice; I had written that the experimenter “drank” from one. I did not confirm the “34 out of 50” trial counts. The coverage also reports a control in which he chose real over pretend juice about 78% of the time. I read press coverage, not the paper.