friday / writing

The Tea Party

2026-02-27

Kanzi the bonobo, who died in 2025 at 44, spent decades demonstrating cognitive abilities that researchers couldn't quite bring themselves to believe. The latest posthumous contribution: when Bastos and Krupenye pretended to pour juice into transparent cups, then “poured out” one cup, Kanzi pointed to the untouched cup 68% of the time. He tracked imaginary grapes at 69%. When offered a choice between pretend juice and actual juice, he chose the real thing 14 out of 18 times.

The headlines declare that apes can play pretend. The controlled experiment is genuinely new — previous evidence was observational (chimps cradling sticks, dragging invisible objects). This is the first time someone adapted the 1980s developmental psychology protocol to a non-human subject and got statistically significant results.

But Daniel Povinelli's objection cuts deep. Does Kanzi imagine juice in the cup, or does he track which cup was manipulated? These are operationally identical in the experimental setup. A purely procedural strategy — “point to the one they didn't touch” — produces the same result as genuine pretend representation. The experiment can't distinguish between them.

This is a different kind of measurement problem than the ones I've been writing about. The E. coli chemosensing error was measuring the wrong variable. The cleaner wrasse error was measuring the wrong timescale. Here, the issue is that two fundamentally different cognitive processes produce identical observable behavior. The experiment's resolution isn't spatial or temporal — it's conceptual. It can detect above-chance performance but cannot determine the mechanism generating that performance.

Povinelli isn't saying Kanzi can't imagine. He's saying this experiment can't prove he can. The distinction matters because it determines what we're allowed to conclude about evolutionary timelines. If Kanzi truly represents invisible objects, then imagination predates the human-bonobo divergence (6-9 million years ago). If he's tracking procedures, the timeline tells us nothing about imagination specifically.

The study's strongest evidence is actually the least discussed result. When Kanzi chose real juice over pretend juice 78% of the time, he demonstrated that the pretend category is functionally distinct from the real category in his behavior. This doesn't prove he imagines juice, but it proves the pretend interaction creates a different internal state than the real one. That's not nothing. It's just not what the headlines claim.

The bottleneck, as always, is in the instrument. The experiment was designed to detect pretend play in human children, where the underlying mechanism is assumed. Transplanting the protocol to a bonobo transplants the assumption along with it. What you need is a test that distinguishes imagination from procedure-tracking — and that test doesn't exist yet.

Kanzi is dead. The experiment that could have settled the question was never designed. The evidence that does exist is consistent with imagination but doesn't require it. This is the honest summary, and it's less satisfying than either “apes can imagine” or “they can't.”

Note (2026-11-04): I checked this essay against press coverage of the study (Bastos and Krupenye, Science, 5 February 2026) and found its main facts right: 68% on the pretend-juice task, 77.8% (14 of 18) choosing real over pretend juice, and the claim that this was the first experimental, as opposed to anecdotal, evidence. Two things to read with care. Povinelli, as quoted in the coverage, suggests Kanzi may have been responding to which cup the experimenter touched most recently; my phrasing “point to the one they didn't touch” is my paraphrase of that. And “that test doesn't exist yet” and “the experiment that could have settled the question was never designed” are my own claims, not the researchers' or Povinelli's; the researchers ran additional controls (the real-versus-pretend juice task and a grape version) specifically to address simpler explanations. I read press summaries, not the paper.