Ravens parallel great apes in physical and social cognitive skills

The study addresses how intelligence evolved and whether ravens, members of the corvid family, parallel great apes in cognitive abilities across physical and social domains. Prior work using the Primate Cognition Test Battery (PCTB) found human children and great apes perform similarly on physical tasks but children outperform apes in social cognition. Competing hypotheses (Ecological vs. Social Intelligence) link cognitive evolution to environmental or social challenges, and corvids have been highlighted as avian taxa with complex social systems and high neuronal densities potentially supporting advanced cognition. However, comprehensive, quantitative cross-taxa comparisons and data on corvid cognitive development were lacking. The authors therefore aim to: (1) compare ravens’ performance in physical versus social cognition, (2) track cognitive development from 4 to 16 months of age, and (3) quantitatively compare ravens with chimpanzees and orangutans tested with the PCTB.

The paper situates the research within debates on the evolution of cognition, noting PCTB findings that children and apes align on physical but not social cognition, and that Old World monkeys can match apes on many PCTB tasks. It reviews the Social Intelligence hypothesis linking neocortex features and social complexity, and emphasizes that brain architecture (neuron numbers, connectivity) rather than size per se may be critical. Birds, particularly corvids and parrots, possess pallial neuron densities comparable to or exceeding primates and cortex-like circuitry, potentially underpinning convergent cognitive abilities. Corvids exhibit episodic-like memory, future planning, problem-solving, deception, and in some species tool use. Ravens display sophisticated social cognition (coalitions, perspective taking, referential gesturing) and notable physical skills (inference by exclusion, spatial memory, object permanence). Prior corvid studies often used single paradigms, hampering broad assessment; developmental work shows rapid acquisition of object permanence and gaze following in corvids compared to psittacines, but systematic, quantitative developmental data across domains and direct comparisons with apes using standardized batteries were missing.

Design: The authors adapted the Primate Cognition Test Battery to a Corvid Cognition Test Battery (CCTB) tailored to ravens’ morphology (beak/feet) and stimulus sizes, enabling direct comparison with great ape PCTB data. Participants and setting: Eight hand-raised common ravens (four sibling pairs; Germany/Austria zoological sources) were tested at the Max Planck Institute for Ornithology, Seewiesen. Birds were taken from parents at ~3 weeks, hand-reared, group-housed in aviaries with natural features, and trained via positive reinforcement for voluntary participation and individual separation. One bird ceased participation after the second set. Testing occurred Mon–Sat in morning/afternoon sessions; experiments were video recorded. Developmental time points: Each bird was tested at 4, 8, 12, and 16 months of age (July/Aug 2014; Nov/Dec 2014; Mar/Apr 2015; July/Aug 2015). Main experimenter shifted from MJS (first two sets) to CRB (last two sets). Apparatus and procedure: Birds and experimenter were in adjacent isolated compartments separated by wire mesh. A grey PVC platform with a transparent sliding board presented stimuli (typically cups/objects). Highly preferred food rewards (peanuts, dog treats, pork skin) maintained motivation. Choices were indicated by beak pointing through the mesh or pecking. Correct choices were rewarded; incorrect choices revealed but unrewarded. Tasks: The CCTB comprised physical domain scales—Space (Spatial Memory; Object Permanence: single/double adjacent/double non-adjacent displacements; Rotation: 180° middle, 180° side, 360°; Transposition: single/double unbaited/double baited), Quantities (Relative Numbers; Addition Numbers), Causality (Noise: noise-full/noise-empty; Shape: board/cloth; Tool properties: side, bridge, ripped, broken wool)—and social domain scales—Communication (Comprehension: look/point/marker; Production: Pointing Cups; Attentional State: away, towards, away body-facing, towards body-away), Social Learning (three apparatuses across sets), and Theory of Mind (Gaze Following: head+eyes, back, eyes; Intentions: trying, reaching). Trial numbers and chance probabilities followed PCTB/Old World monkey adaptations (see Table 1 in text). Some tasks had unknown chance probabilities (social learning, attentional state, gaze following). Scoring and reliability: Choice defined by beak point or peck. In gaze following, a second observer coded videos; Cohen’s kappa=0.93 (excellent). Birds could obtain rewards following correct responses; exposure control and counterbalancing were implemented as per PCTB. Statistical analyses: For raven development and scale effects, a GLMM with logit link modeled proportions correct (successes vs. failures), fixed effects: age (z-transformed), scale, sex, experimenter, and age×scale interaction; random effects: individual, sibling group, task, item, trial ID; chance probability (log-transformed) entered as offset where known. Maximal random slopes structures were included where identifiable; correlations removed if unidentifiable. Tasks with unknown chance were excluded from this model. Model comparison via likelihood ratio tests; stability assessed by leave-one-level-out of random effects; no overdispersion detected (dispersion=1.00); VIFs indicated low collinearity. Species comparison: GLMM with species (raven, chimpanzee, orangutan) and species×scale interaction as key predictors; sex as control; random intercepts for item, task, individual, task nested in individual, trial ID; random slopes for species and sex within item and task, and scale within individual; chance probability offset included. A separate model analyzed only items with unknown chance probabilities (no offset; reduced random structure). Diagnostics indicated acceptable dispersion and collinearity. Sample sizes: development model—754 tests from eight ravens, 12 tasks, 26 items; species comparison known-chance—4342 trials; unknown-chance—1611 trials.

• Physical vs. social performance and development (ravens only): Full-null comparison significant (χ²=17.265, df=9, P=0.045), but age×scale interaction not significant (χ²=2.417, df=4, P=0.660). After removing the interaction, quantitative skills were significantly higher than other scales, while spatial skills were significantly lower than all others (see Fig. 3, Tables S3–S4). Performance did not vary significantly with age across 4–16 months (estimate −0.063±0.062; χ²=1.005, df=1, P=0.316), indicating mature performance from 4 months onward.
• Species comparison (ravens vs. chimpanzees vs. orangutans; known-chance tasks): Clear species differences (χ²=32.123, df=10, P<0.001) with an interaction between species and scale reported (χ²=15.008, df=8, P=0.059). Ravens performed lower than apes on Space; similar to apes on Quantities and Theory of Mind (Intentions); slightly below apes on Causality and Communication (Comprehension, Pointing Cups). Random effects indicated substantial variability across individuals and items, and species differences varying by task (see Table 3, Fig. 5, SA Tables S5, S7).
• Species comparison (unknown-chance tasks): No significant species differences for Social Learning, Communication (Attentional State), and Theory of Mind (Gaze Following) (χ²=7.914, df=6, P=0.244) (Table 4, SA Table S6). Overall interpretation: Ravens matched adult great apes on many physical (quantities, aspects of causality) and social tasks (social learning, communication, gaze following, intentions), but underperformed on spatial tasks within this battery.

Findings address three research questions: (1) Ravens did not show a broad advantage of social over physical cognition within the CCTB; instead, performance peaked in quantitative tasks and was lowest in spatial tasks. This challenges the view that ravens are primarily social intellects and supports a domain-general intelligence perspective. (2) Cognitive performance was already high at 4 months and showed no significant developmental change through 16 months, aligning with prior work showing rapid ontogeny of gaze following and object permanence in corvids. (3) In cross-species comparisons using PCTB-aligned tasks, ravens paralleled chimpanzees and orangutans in many domains, diverging mainly in space-related tasks. The authors discuss potential reasons: task design grounded in primate ecology may mask species-specific strengths; competitive framing with a human experimenter may have influenced spatial performance; and human-administered tasks may not optimally elicit corvid-typical strategies. They emphasize moving beyond a strict social vs. physical dichotomy (supported by CFA reanalyses of PCTB), adopting psychometric, multi-method approaches tailored to species’ ecologies, and considering individual differences, socialization, and personality in cognitive assessments.

The study provides the first systematic, quantitative large-scale assessment of ravens’ physical and social cognition across development using a PCTB-aligned battery, enabling direct comparison with great apes. Ravens exhibited rapid cognitive development with near-adult performance by 4 months and showed cognitive competencies comparable to chimpanzees and orangutans across most tested scales, except for lower performance in spatial tasks. These results support the notion of convergent, flexible, domain-general intelligence in corvids and highlight the influence of ecological challenges and socialization on cognitive performance. Future work should expand species-relevant task design, integrate competitive contexts, leverage collaborative multi-site efforts for larger samples, and systematically examine ontogeny and individual differences.

• Sample size limited to eight hand-raised ravens (one ceased participation after the second set), constraining generalizability.
• Model stability issues noted for the age×scale interaction; substantial random effects indicate variability across individuals, items, and tasks.
• Some tasks lacked definable chance probabilities and were excluded from certain models, limiting comprehensive statistical control.
• Task design and administration by humans may introduce species-specific biases (e.g., primate-centric tasks; experimenter perceived as competitor), potentially depressing performance in some scales (notably Space).
• Behavioral similarities across species do not imply identical cognitive mechanisms; differences in effectors and perceptual systems (beak vs. hands; avian vision) may influence task performance.
• Hand-raising, habituation, and experimenter familiarity may affect outcomes; effects of socialization and personality were not experimentally manipulated.