346 | Erica Cartmill on How Human and Animal Minds Think and Play

What It Covers

Anthropologist and cognitive scientist Erica Cartmill joins Sean Carroll to examine how human intelligence differs from animal cognition — not through a linear hierarchy, but through a unique constellation of overlapping abilities. The conversation covers numerical cognition, social hierarchy awareness, playful teasing, animal humor, and what comparative cognition reveals about evaluating AI systems.

Key Questions Answered

• •Intelligence as constellation, not hierarchy: No single cognitive ability separates humans from other species. Instead, humans possess a unique configuration of partial abilities — syntax, reference, compositionality — each found individually in at least one other species. The distinction is that humans combine all these components simultaneously, potentially producing something qualitatively different rather than simply quantitatively superior to other animals.
• •Chimpanzee working memory superiority: Chimpanzee Ai at Kyoto's Primate Institute outperforms untrained adult humans at sequencing numbers one through nine after they flash briefly on screen. However, trained human experts can match her performance. The key cognitive difference is that chimpanzees never internalize the cardinal number principle — that each successive number is exactly one greater — while human children grasp this pattern around age four or five.
• •Playful teasing as social calibration tool: Cartmill's research frames teasing as a low-stakes mechanism for continuously recalibrating social relationships. Because social dynamics shift daily with mood, status, and context — unlike stable physical properties — animals and humans use teasing to test how much a partner values the relationship over momentary annoyance, functioning as an ongoing, honest signal of relational investment and tolerance.
• •Laughter induces measurable optimism in bonobos: In controlled experiments, bonobos exposed to recorded laughter subsequently approached ambiguous gray reward boxes — never previously seen — at higher rates than bonobos exposed to neutral water sounds. This cognitive bias test, originally developed for animal welfare assessment, suggests auditory positive-emotion signals shift expectation states toward optimism without providing any direct informational content about actual reward availability.
• •Orangutan vs. chimpanzee problem-solving styles reflect ecology: Orangutans, largely solitary in the wild, observe problems at length before acting and often solve them on the first attempt. Chimpanzees and bonobos, living in large competitive social groups, immediately manipulate objects physically and repeatedly. These behavioral differences reflect ecological pressures — in competitive groups, delayed action means losing resources — rather than differences in raw cognitive capacity between the species.
• •Apply animal cognition methods to evaluate LLMs: Because LLMs produce fluent verbal responses, researchers risk treating their outputs as transparent self-reports of internal processing — the same error avoided when studying nonverbal animals. Cartmill advocates applying comparative cognition methodology to AI: systematically varying question framing, modality, and context to probe actual understanding rather than accepting verbal output at face value, since stated reasons may not reflect underlying computational processes.