The paper, authored by Tom S. Roth of Utrecht University, Ilyena Hirskyj-Douglas of the University of Glasgow, and Juan Olvido Perea-García of the University of Las Palmas de Gran Canaria, does not call for a halt to digital technologies in animal research. Instead, it delivers a systematic critique of how those technologies are deployed, drawing on case studies that range from parrots tapping tablet screens with their tongues to orang-utans failing reaction-time tasks that their energy-conserving biology makes them poorly suited to perform. The authors argue that many null results in comparative cognition—cases where an animal “fails” a test—may reflect a mismatch between the method and the species rather than an absence of the cognitive ability under investigation. Null findings, in other words, are too often read as evidence of what animals cannot do, when they may instead signal that the instrument itself was never fit for the purpose.

The technical foundations of the critique rest on six interlocking problems. First, most comparative studies today are effectively ports of well-established human paradigms, a top-down transfer that ignores species differences and constrains interpretation. Second, these studies rely heavily on extrinsic motivation—usually food rewards—which measures an animal’s drive to obtain a treat as much as its cognitive capacity, and individual differences in food preference can obscure genuine performance differences. Third, participation typically demands extensive training regimes that reshape behavior before data collection even begins, hiding the very reasons animals do or do not engage with stimuli. Fourth, data collection usually requires the physical presence of researchers or caregivers, a well-documented source of confound; prior work has shown that even minor differences in experimenter behavior measurably alter how animals respond. Fifth, the necessary equipment is sophisticated, proprietary, and expensive, favoring wealthy institutions and effectively monopolizing research topics. Sixth, testing is compartmentalized around researcher availability, which reduces the animals’ perceived control over their environment—a factor widely regarded as fundamental to captive animal welfare.

Nowhere are these tensions more visible than in eye-tracking, the method the authors use as their central test case. Restraint-free eye-tracking systems measure attention by detecting infrared reflections from the cornea, allowing researchers to record gaze direction and fixation duration with millisecond precision. The method transformed primate social cognition research, contributing to landmark findings such as evidence that great apes anticipate others’ actions based on false beliefs. But stabilizing a chimpanzee’s head without physical restraint posed an early challenge, and the field’s solution—the juice nozzle—has become standard: participants hold a fixed nozzle in their mouth and sip juice rewards while viewing stimuli. The innovation solved the stabilization problem while introducing new ones. Juice administration is often controlled manually, its timing and duration vary, and reward preferences shift over time and between individuals. A decline in performance may reflect habituation to the reward rather than anything about attention or cognition.

Even refinements designed to fix these problems can import fresh biases, and the review illustrates this with a study of predator gaze detection in chimpanzees. Researchers there automated juice delivery with a dosing pump, dispensing rewards only when a participant attended to the screen—a clear improvement over manual administration. But during calibration, the chimpanzees were rewarded only for sustaining gaze for at least 400 milliseconds on a shrinking video, a procedure that may have inadvertently trained them to look toward the center of the screen. Heatmaps of the resulting gaze data showed chimpanzees fixating centrally despite no salient information there, raising the possibility that the human-chimpanzee differences reported in the study stemmed from the methodology itself rather than from divergent visual perception. The case demonstrates, the authors argue, how a succession of amendments to a human-designed method can progressively obscure what the data actually mean.

The cost and accessibility of commercial eye-tracking hardware compounds these scientific problems with structural ones. Commercial systems are designed for human users, encouraging uncritical transfer of human paradigms, and they carry substantial financial burdens: specialized hardware, proprietary software, plexiglass installations, and protective casings for electronics in zoo environments. Long acquisition lead times pressure researchers to publish findings without fully questioning whether the apparatus suits their research question. Open-source alternatives point toward a different future. An open toolbox for calibrating eye-trackers with nonhuman primates now exists, some laboratories have published complete tutorials for portable touchscreen setups, and machine-learning algorithms can extract gaze estimates from ordinary camera footage, relaxing the physical constraints that currently force animals into juice-stabilized postures. Custom systems can be dramatically cheaper: one automated testing platform built on Raspberry Pi microcomputers, which cost between 35 and 75 dollars, with custom reward dispensers assembled for under 45 dollars each, collected more than five million trials from socially housed rhesus and squirrel monkeys over two years—an order of data volume impossible under researcher-dependent paradigms.

The parrot touchscreen studies offer perhaps the most vivid demonstration of why iterative prototyping matters. Capacitive touchscreens—the kind in every smartphone—detect the electrical conductivity of human fingertips. Parrot tongues are not wet, so when researchers trained parrots to interact with tablets using their tongues, recognition accuracy averaged a dismal 41 percent, with multitapping corrupting 46 percent of interactions. The failures were not cognitive failures; they were ergonomic mismatches rooted in anatomy. By enlarging button sizes and implementing first-touch recognition with 300-millisecond thresholds, the team raised accuracy to 47 percent in later iterations—and along the way discovered something genuinely new: that button spacing, which matters for human touch performance, made no difference to parrots, while the birds’ persistence in the face of interface challenges opened unexpected research questions about individual differences in tenacity. A parallel story unfolded with white-faced saki monkeys, for whom fine-motor push buttons proved incompatible with natural movement patterns. Iterative redesign toward swing and pull mechanisms aligned the interface with the sakis’ own interaction tendencies, revealing species-specific motor preferences in the process.

The review also scrutinizes what happens when animals are excluded for failing training criteria—a practice the authors link to what philosopher Cameron Buckner termed methodological anthropocentrism. Zoo-based touchscreen studies have a median sample size of just three individuals, in part because training exclusions shrink already small populations. In one study of gaze-cue visibility, ten chimpanzees entered training but only two completed all test conditions; the resulting conclusion—that human-like white sclera makes gaze direction more visible—may instead reflect the eye-contact tolerance of a self-selected minority, given that chimpanzees generally avoid mutual gaze and can perceive prolonged eye contact as threatening. Similarly, when Bornean orang-utans showed no attentional bias in a touchscreen dot-probe task, the researchers interpreted the null result biocentrically: orang-utans’ energy-conserving metabolism, shaped by fruit scarcity in their native peat swamp forests, may make rapid manual responses ecologically foreign. A follow-up study using eye-tracking, which captures eye movements with less delay than hand movements, did detect the bias the touchscreen task had missed. The paradigm, not the primate, had been the bottleneck.

The authors’ recommendations are deliberately practical. Researchers should treat high exclusion rates as warning indicators rather than statistical inconveniences, critically assessing how a paradigm may have shaped results before drawing species-level conclusions. They should reduce dependence on extrinsic rewards and researcher presence, designing autonomous or semi-autonomous setups that animals can engage voluntarily—improving both welfare, by restoring perceived control, and data quality, by removing social confounds. Where possible, tasks requiring minimal pre-training should be preferred, or training integrated into routine husbandry. Manual behavioral scoring, though labor-intensive, can capture intrinsically motivated attention without rewards, as a recent study of long-tailed macaques watching videos of familiar versus unfamiliar individuals demonstrated. Advances in computer vision, with automatic facial recognition in macaques now reaching accuracies of roughly 82 to 92 percent, promise to make such approaches scalable without invasive identification implants that many zoos will not accept.

The deeper message of the review is epistemological: comparative cognition should treat its methods as perpetually provisional. Just as human computing evolved from expert-operated machines in the 1950s into adaptive systems woven into daily life, animal research interfaces should evolve as understanding of each species deepens—a reflective loop in which every dataset, including every failure, becomes feedback for redesign. Results published today, the authors suggest, are better understood as preliminary stepping stones than as definitive verdicts on what animals can and cannot think. Embracing that humility, they argue, would yield findings that are simultaneously more scientifically robust and more ethically sound, aligning research objectives with the genuine interests and welfare of the non-human participants whose minds science is trying, at last, to meet on their own terms.