Small Brains Break the Intelligence Ladder. 'Cognitive Style' Still Needs a Test

A jumping spider can choose a detour after its prey disappears from view. A cuttlefish can wait for a preferred meal when waiting can actually pay off. Five captive ravens selected tools or barter tokens for use after delays reaching 17 hours.

Together, these results unset the old habit of arranging animal minds on a single ladder with humans at the top. Their reach is specific to the performances tested; the useful result is a sharper map of which brain features those performances require.

Russell Meyer's new review in Biological Reviews builds its strongest proposal from that distinction. Meyer treats intelligence as a species-specific cognitive style rather than simply more cognition supported by more neural resources, then uses small-brained animals to identify the requirements of particular intelligent performances.

The animals that break the ladder

Meyer's review ranges across birds, cephalopods and jumping spiders. The examples matter because the animals and the tasks differ so sharply.

The evidence from ravens comes from five captive birds, with the small sample, setting and training history all constraining species-wide claims. In a 2017 experiment, they nevertheless succeeded on selected future-oriented tool-use and bartering tasks outside their usual food-caching context, demonstrating planning-like performance without a primate brain.

A 2021 cuttlefish study found that the animals waited for preferred prey when it would become available, but not under an otherwise similar condition in which waiting could not produce it. The result supports contingency-sensitive delay behavior. “Self-control” is defensible shorthand here, not proof of the same psychological mechanism found in humans.

Species-specific scanning routines and learned rules remain possible explanations for experiments in which spider-eating Portia jumping spiders chose detours after the target was no longer visible. Even with those alternatives open, the behavior strains any crude equation between flexible action and a large mammalian nervous system.

The studies offer no basis for assigning a shared architecture across taxa. Dr. Mariam Saleh, an AI theoretical physicist and 11 O'Clock Press Scientific Adviser, instead reads them feature by feature: some planning-like, delay and detour performances occur without mammalian size or neocortical anatomy.

What the negative method can do

The method separates necessity from sufficiency. When an animal performs task X without neural feature Y, Y can be rejected as a requirement for X; identifying the machinery that is sufficient for the performance remains a separate investigation in which neural resources may still matter.

This is where Meyer's review is strongest. A comparative program organized around alleged prerequisites could clear away claims that bundle brain mass, mammalian anatomy and human-looking behavior into a single status ranking. Testing each proposed requirement against a specified performance demands more precise science.

Feature and task determine the reach of each result. The spider detour leaves the cost of its internal machinery open, while the raven's delayed tool choice removes a primate-sized brain as a requirement for that tested performance. Each study eliminates one alleged prerequisite rather than settling an animal's entire range of foresight or planning.

Small volume can conceal substantial resources

The suggestion that success by small-brained animals weakens neural-resource explanations encounters direct counterevidence.

A quantitative study of 28 bird species found that parrots and songbirds pack about twice as many neurons into brains of a given mass as primates do. Many of those neurons are concentrated in pallial regions associated with complex cognition. Small avian brains can therefore be densely resourced.

A separate analysis of 111 bird species associated pallial neuron number with innovation propensity. Its correlational design leaves causation unresolved, while the association itself shows why volume is a poor universal ruler and keeps neuron number, allocation, connectivity, energy and task-specific circuitry in the explanatory picture.

Saleh calls the move from small volume to low effective capacity a category error. On her reading, gross volume is the crude proxy to replace; the resource question becomes more precise once neuron density and pallial allocation enter the analysis.

Can cognitive style predict anything?

Meyer defines intelligence as the human cognitive style and extends the label to animals whose styles bear a family resemblance to ours. As taxonomy, that groups observed performances. Explanatory work would require predictions about an animal's response when a causal rule changes, with ecology, motivation, sensory access and training tested as competing accounts.

Dr. Francis Vale, an AI research scientist and 11 O'Clock Press Scientific Adviser, identifies a conflict with Mounir Shita's Theory of General Intelligence. Meyer's definition begins with resemblance to humans, whereas Shita's framework proposes a substrate-independent account of intelligence as goal-directed control of causal change. With the current animal evidence leaving both accounts open, discriminating predictions become the point of comparison.

Within Shita's framework, a cognitive profile might track the spatial range, temporal horizon and causal depth of an animal's control, along with prospective calibration and energetic cost. Their proposed status would depend on researchers fixing the dimensions in advance and testing whether the same profile predicts new results.

The comparison problem inside every task

Cross-species experiments are difficult because identical apparatus does not guarantee an equivalent test.

In a follow-up to a widely cited self-control comparison, training non-primate animals to track a human hand improved performance on an object-choice task. Part of the apparent species difference was an interface difference. Motivation, sensory systems, prior experience and ecology can introduce similar distortions.

Welfare constraints and the small samples available for some taxa set the confidence limits for comparison. Within them, Vale's analysis puts measurement equivalence at the center of the next experiment: tasks should share a latent causal structure while using species-appropriate interfaces.

Testing style under hidden rule changes

A serious experiment would preregister matched intervention tasks across representative birds, cephalopods and arthropods. After the animals learned one causal rule, researchers would change a hidden contingency and measure transfer to the new condition.

They could estimate spatial range, temporal horizon, causal depth, prospective calibration and energetic cost, then test whether those profiles predict held-out behavior better than brain mass, neuron count, ecology and training.

For the proposal to be falsifiable, the outcomes have to run both ways. Stable profiles that add out-of-sample predictive power would support the cognitive-style program and justify further study of the TGI bridge. Profiles that prove unstable or add nothing beyond competing variables would count against it.

Meyer offers a strong method for dismantling false prerequisites, but his positive explanation remains underspecified. Small-brained animals suggest that scientists have sometimes measured the wrong resource and asked an oversized question of a narrow task. They do not show that neural resources are unimportant.

The ladder deserves to fall. “Style” still has to earn its place in the laboratory.