The research, conducted by Marie Barou-Dagues of the Centre d’Etudes Biologiques de Chizé, CNRS-La Rochelle Université, and Frédérique Dubois of the Université de Montréal, asked a deceptively straightforward question: does success in a simple colour discrimination task predict success in a more complex version of the same task? To find out, the team tested zebra finches (Taeniopygia guttata), small Australian estrildid finches that have become a staple of laboratory research on behaviour and cognition thanks to their readiness to breed in captivity and their quick, reliable learning.

The experimental design hinged on two visual discrimination tasks that differed in a single, carefully controlled dimension: reward predictability. In the simple task, one coloured cue always delivered five seeds while the alternative colour always delivered nothing. Birds could, in principle, learn the rule quickly, because every choice produced unambiguous feedback. In the complex task, the researchers scrambled that certainty. The rewarded colour now delivered five plus or minus one seed, while the less rewarding colour delivered two plus or minus one seed. Both options paid off; they simply differed in expected value. To choose optimally, a bird had to track differences in reward magnitude across trials in which random variation blurred the signal.

Rather than relying on a single measure of success, the authors extracted three complementary performance metrics from each bird’s choice history. The first was a conventional learning score: the number of trials required to reach a pre-defined learning criterion. The second was a behavioural reaction norm capturing learning speed, essentially how quickly an individual’s choices shifted in response to accumulated experience. The third was the best linear unbiased prediction, or BLUP, of individual choices, a statistical estimate of learning strength that quantifies how consistently an animal’s decisions tracked the underlying reward contingencies once noise from the experimental design was accounted for. Using several metrics matters, because each captures a different facet of what colloquially gets called intelligence, and studies that collapse cognition into a single pass-fail score risk missing meaningful variation.

The results confirmed that the manipulation worked as intended. Birds learned faster and achieved greater learning strength in the simple task than in the complex one, demonstrating that trial-by-trial variation in reward magnitude genuinely increased discrimination difficulty. Even more striking was the role of colour itself. Yellow and pink cues significantly enhanced learning speed in both tasks, regardless of which colour was actually associated with the larger reward. This stimulus bias, an artifact of how the birds perceive and respond to particular hues, illustrates a persistent headache for comparative cognition research: stimulus properties can drive performance independently of cognitive ability, potentially masking or inflating apparent differences among individuals.

The headline finding, however, concerns what psychologists call convergent validity, the expectation that different tests of the same underlying ability should produce correlated results. It largely failed here. Interindividual differences in learning score, learning speed and learning strength were all significantly greater in the complex task than in the simple one, and, crucially, simple-task metrics were poor predictors of complex-task metrics. A bird that cracked the easy discrimination in record time was not reliably the bird that mastered the stochastic version. In other words, the two tasks did not appear to be measuring a single, stable learning ability; they seemed to be tapping into different combinations of perceptual, motivational and cognitive processes.

Why would complexity amplify individual differences? One possibility is that simple tasks create a ceiling effect. When one option always pays and the other never does, most animals in good condition converge on the correct answer, compressing the range of observable scores and hiding genuine variation. Complex tasks, by contrast, demand integration of noisy information over many trials, so individuals with different perceptual acuity, attention, memory, risk sensitivity or persistence begin to diverge. The zebra finch data are consistent with this interpretation: variance among individuals expanded precisely when the task made the correct choice harder to identify. From a methodological standpoint, this suggests that easy screening tasks may systematically underestimate the cognitive diversity present in a population.

The study also carries a broader conceptual message. Individual choices, the authors emphasise, result from the interplay of cognitive abilities, reward structures and stimulus properties. Any observed performance therefore reflects all three, not cognition alone. A bird’s apparent slowness might stem from a preference for one colour rather than from inferior learning. An apparent genius might simply have received a cue that its visual system finds salient. Disentangling these factors requires experimental designs that deliberately manipulate task complexity and stimulus characteristics, rather than assuming that any discrimination task provides an equivalent window onto learning ability.

For the growing community of researchers studying animal personality and cognitive variation, and for evolutionary biologists interested in whether cognitive traits are heritable or linked to fitness, the implications are significant. If simple tests poorly predict performance in ecologically realistic scenarios, where rewards are variable and feedback ambiguous, then conclusions drawn from easy laboratory assays may not generalise to the challenges animals face in the wild. Cognitive ecologists have long debated how to measure cognition in ways that are both rigorous and relevant; this study adds empirical weight to the argument that task design is not a neutral choice but an active determinant of what gets measured.

The work, supported by a Natural Sciences and Engineering Research Council of Canada discovery grant and a scholarship from the University of Montreal, was approved by the University of Montreal’s Animal Care committee, with procedures designed to minimise stress, including dark capture conditions and maintenance of auditory contact with flock mates. Its findings, published under a Creative Commons licence, are likely to resonate well beyond ornithology. Anyone using associative learning assays, from comparative psychologists to behavioural ecologists and even researchers studying learning in other taxa, now has quantitative grounds for a familiar suspicion: the easiest test is not always the most informative one. Capturing the true breadth of cognitive diversity, the authors argue, requires embracing complexity rather than avoiding it.

Subject of Research: Associative learning and interindividual cognitive variation in zebra finches across simple and complex discrimination tasks

Article Title: Associative learning: does performance in a simple discrimination task predict success in a more complex one?

Article References: Barou-Dagues, M., & Dubois, F. (2026). Associative learning: does performance in a simple discrimination task predict success in a more complex one?. Animal Cognition. https://doi.org/10.1007/s10071-026-02103-y

Image Credits: AI Generated

DOI: 10.1007/s10071-026-02103-y

Keywords: associative learning, zebra finch, animal cognition, colour discrimination, reward magnitude, cognitive ecology, task complexity, interindividual variation, learning speed, visual discrimination, behavioral reaction norm, Taeniopygia guttata