The Chimp Test: Why Young Chimps Beat Humans at Memory

Sana Inomata and Tetsuro Matsuzawa trained three young chimpanzees — Ayumu, Ai and Cleo — on a task the primates had already practiced as a game since infancy: numerals scattered on a screen, touch them in order. The decisive manipulation was speed. In the key condition, the numerals appeared for just 650 milliseconds before being masked. Ayumu completed sequences of five numbers at around 80 percent accuracy; the human students, seeing the same 650 ms flashes, came in far below that — and on the briefest exposures the humans' performance fell apart while the young chimps' barely moved.

Two details matter for interpreting the result. First, the chimps had years of practice with the numeral task — though the students were also trained until their performance stabilized, so practice alone does not explain the gap. Second, the effect was age-bound: an older chimp in the study performed more like the humans. Whatever advantage the young chimps held, it was strongest in childhood — which is exactly where the interesting hypothesis begins.

Matsuzawa's explanation is a trade-off hypothesis: somewhere in human evolution, we traded raw, image-like working memory for language and symbolic processing. Our species' edge is meaning — chunking the world into words, categories and stories — and the price may have been the chimps' kind of memory: fast, high-resolution capture of exactly what was on screen, unfiltered by interpretation. In this view, Ayumu is not smarter; he is running a different, less annotated memory system that never got overwritten.

The competing, more deflationary explanation is attention and strategy: the chimps, raised on this task since infancy, may simply encode the numeral positions with a specialization the students could not match in weeks of practice — an extreme version of the expertise effects seen in chess masters and radiologists. The honest summary is that the gap is real, replicable, and its cause is still debated. What is not disputed: it says nothing about general intelligence in either species.

On the humanbenchmark-style version — eight numbers, click ascending, each correct press hides its number — most people reliably clear the first several levels and start failing somewhere in the mid-to-high teens of total numbers when the grid gets crowded and the flash gets harder to sweep. Scores swing enormously with strategy: players who fixate the center and encode relative positions ('3 upper-left of 5') hold sequences far better than players who try to read numbers one by one. If your score is mediocre, the bottleneck is almost always encoding speed, not capacity.

Age matters, in the direction the chimp data predicts: this kind of rapid visual capture peaks in childhood and adolescence and declines gently through adulthood — a decade earlier and more steeply than verbal span does. So a fourteen-year-old clearing levels you cannot is not practicing more; their visual working memory is genuinely faster. The fix is not envy, it is strategy, which is what the next section is about.

The single biggest gain comes from giving up on reading order: you cannot recite '1, 2, 3…' faster than the flash, but you can capture the whole shape at once. Encode the numbers as a path — a connect-the-dots route across the screen — because spatial paths are what your visual memory stores best. Center-anchor first: note where 1 sits, then link each number to the previous one as a direction ('2 up-right, 3 down, 4 far-left'). With practice the whole sequence becomes one shape instead of eight items, and shapes survive the hidden-numbers phase that items do not.

Second: stop the mouse from doing memory's job. Players who hover number by number, waiting to click, offload the sequence onto their cursor path and collapse the moment a click lands wrong. Plan the full route before the first press, then execute it. Third, drill the underlying span separately — Sequence Memory trains exactly this store-sequence-reproduce loop in a purer form, and alternating it with the chimp test is the fastest combination we have measured for improving chimp-test levels specifically.

• Encode as a path, not a list. One connected shape beats eight separate numbers, every time.
• Center-anchor. Fix number 1's location first; everything else gets encoded relative to it.
• Plan the whole route before clicking. Hover-clicking is offloading, and offloading fails.
• Alternate drills. Chimp Test for the real task, Sequence Memory for the raw span.

Practice the chimp test and you will get strikingly better at the chimp test — within weeks, scores can double. But the improvement is heavily task-specific, consistent with the wider working-memory training literature: near transfer (similar visual-span tasks) is real, far transfer (general intelligence, everyday memory) is close to zero. If you are playing because the game is fun and the score is satisfying, that is a perfectly good reason — the focus it demands is genuinely engaging.

If your goal is everyday memory — names, where you put things, what you walked into the room for — spend the majority of your time on tasks closer to life: the games in our daily workout rotate through spatial, verbal and sequential families, which is a fairer portfolio than any single fast-flash task. Use the chimp test as what it is: a sharp, well-validated probe of one fascinating, narrow corner of memory, with the best origin story in cognitive science.