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

Episode Summary

Executive Summary: Sean Carroll and Erica Cartmill explore intelligence as a constellation of species-specific abilities rather than a single ladder toward humans. They discuss comparative cognition in apes, pointing, number sense, social hierarchy, teasing, humor, play, and animal joy, then extend the framework to AI: language-like outputs are not proof of understanding, so LLMs should be studied with tools from animal and developmental psychology.

Main Topics: Intelligence as a constellation, not a linear scale (Priority: 5/5): Cartmill argues against the old Scala Naturae view. Different species excel at different tasks depending on their ecological and social niches, so human intelligence is not a single superiority switch but a unique combination of abilities. Language and the limits of human exceptionalism (Priority: 5/5): The discussion breaks language into syntax, reference, compositionality, and other parts. Many components appear in some form across species, but humans may be unique in combining them all into one system. Comparative cognition in great apes and animals (Priority: 5/5): Examples include chimpanzee numerical sequencing, ape tool-related planning, hierarchy tracking, and communication studies that reveal complex social understanding in nonhuman animals. Teasing, humor, and play as social cognition (Priority: 5/5): Cartmill frames teasing as a gray zone between aggression and play that can function like a proto-joke, testing social bonds, expectations, and relationship strength in apes, dogs, cats, and possibly other animals. Social intelligence, cooperation, and role structure (Priority: 4/5): Human cooperation and trust are contrasted with ape social strategies. The conversation covers dominance hierarchies, alliances, mating access, and how animals track who matters in a group. Joy, emotion, and behavioral biomarkers (Priority: 4/5): Cartmill discusses ongoing work measuring positive emotion in apes, parrots, and dolphins using laughter playback, thermal imaging, and cognitive bias tasks to test whether animals become more optimistic. AI and the problem of apparent understanding (Priority: 5/5): The podcast closes by linking animal cognition methods to LLMs. Cartmill argues that AI systems cannot be trusted at face value just because they produce human-like verbal explanations; they need behavioral probing and interdisciplinary study.

Key Arguments: Human intelligence is best understood as a unique constellation of partial abilities rather than one monolithic trait. Many cognitive components once thought uniquely human can be found in other species, but not usually all together in the same package. Chimpanzees can outperform humans on trained memory-sequencing tasks, showing that comparison depends on expertise and task design. Numerical competence differs qualitatively between humans and chimpanzees: humans often grasp the cardinal number principle, while chimpanzees can learn sequences without the same abstract leap. Apes and other animals can interpret social information embedded in calls, rank, and context, not merely simple signal categories. Teasing and play may function as low-stakes tests of trust, social boundaries, and relationship strength, making them more than mere aggression. Humans are unusually cooperative and trusting in communication, which is part of what makes language and culture scale so effectively. LLMs should not be treated as if their verbal outputs directly reveal understanding; their answers require skeptical, methodical validation similar to animal cognition research.

Data Points: Chimpanzee number sequence: 1 through 9 (and reportedly into the teens) - Ai the chimpanzee was trained to press numbers in order after seeing them briefly on a screen. Human child number-learning pattern: Around 4 or 5 - Cartmill describes the human developmental shift where children often suddenly grasp the cardinal number principle after learning a few early numbers. Orangutan infant dependency: 10 to 12 years - Female orangutans travel with one infant at a time for many years before the older offspring becomes independent. Great ape study species: 3 species - Cartmill mentions a project comparing joy-related measures in kias, dolphins, and great apes. Species in joy study: Kia parrots, dolphins, great apes - The collaboration aims to identify positive emotion across distant lineages. Human-like hierarchy experiments: 3 individuals (A, B, C) - A playback setup is described where calls are rearranged to simulate a lower-ranked individual defeating a higher-ranked one. Social group structure: Multi-male, multi-female groups - Chimpanzees and bonobos are described as living in large socially complex groups similar in some ways to humans. Physical comparison: 1 dog - Cartmill refers to her dog Bonnie as part of a private game involving a stone turtle toy.

Pivotal Quotes: "Humans have evolved to be good at helping humans fit into the social and ecological niches that we have, but the same is true of every other species." — Erica Cartmill: Explaining why intelligence should not be ranked on a single linear scale. "What we really have is this unique constellation, this unique configuration, that pulls them together in a way that really helps us build on each other and gives us something that is really powerful." — Erica Cartmill: Describing the distinctive combination of human abilities. "The big question is, does an LLM understand anything?" — Erica Cartmill: Transitioning from animal cognition to artificial intelligence and the limits of language-like output.

Implications: Listeners should be cautious about ranking minds on a single human-centered scale. The same comparative tools used for animals can sharpen how we test AI, and future progress will depend on interdisciplinary, behavior-based methods rather than verbal self-report alone.