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

The traditional view of intelligence as a linear progression with humans at the top is inaccurate. Each species has evolved a unique 'constellation of abilities' that allows it to thrive in its specific ecological and social environment. Humans are equally evolved as other species, just good at different things.

Erica Cartmill states, 'Humans are equally as evolved as every other species. And so the things that humans have evolved to be good at are, you know, helping humans fit into the social and ecological niches that we have. But the same is true of every other species.'

Chimpanzees, particularly one named Ayumu, demonstrate superior working memory compared to untrained human adults in a task where numbers flash on a screen and must be pressed in sequence after disappearing. This highlights a specific cognitive ability where non-human apes can outperform humans.

The experiment involves numbers 1-9 flashing on a screen; as soon as the first number is pressed, all others disappear, requiring quick mental mapping. Ayumu is 'much better than untrained adults, human adults.'

While chimpanzees can be trained to recognize and sequence numbers, their learning is laborious for each new number. Human children, however, experience an 'aha moment' around numbers four or five, grasping the 'cardinal number principle' that each new number is exactly one higher than the last, allowing for abstract understanding and manipulation.

Cartmill explains, 'teaching her [Ayumu] a new number is just as difficult as teaching her the last number... that's very different from what happens with humans... once they learn four or maybe five, then they suddenly get it and they realize that each new number is exactly one number higher.'

Cognitive strengths among great apes vary based on their social environments. Highly social chimpanzees and bonobos excel in 'Machiavellian intelligence' (understanding motivations, alliances, gossip) crucial for navigating complex group dynamics. More solitary orangutans, conversely, show superior skills in material culture and manual problem-solving.

Chimpanzees and bonobos 'are very interested in paying attention closely to who's doing what with whom when.' Orangutans 'spend a lot of time... in a semi-solitary state' and 'really seem to excel is in material culture, material manipulation.'

Humans are fundamentally a cooperative species, willing to share information and help unrelated individuals, often at a personal disadvantage. This high level of trust and cooperation is a key differentiator from other apes and is essential for complex social structures and communication like language.

Cartmill states, 'humans are fundamentally a cooperative species. We share information with others that doesn't benefit us. Sometimes it disadvantages us.' She contrasts this with apes who would not share the location of a resource.

Playful teasing in animals, especially great apes, involves 'offer and withdrawal' or 'provocative non-compliance.' This behavior, which often violates expectations or social norms, acts as a 'proto-joke' and serves to build and test the strength of social relationships by gauging how much the other individual values the relationship over momentary annoyance.

An infant orangutan repeatedly offered a stick to its mother then pulled it back, and the mother reciprocated. Cartmill notes, 'it has this structure of a very basic joke... this setup, the offer, and then this... punchline... the withdrawal.' Teasing 'might be serving both to build relationships but also to test those relationships.'

Bonobos, when exposed to laughter as an auditory stimulus, exhibit increased optimism. In a cognitive bias test, they were more likely to approach ambiguous 'gray boxes' (intermediate between consistently rewarded black and unrewarded white boxes) after hearing laughter, suggesting a positive emotional state influencing their expectations.

Bonobos 'when they've listened to laughter, they're more likely to approach the gray boxes than when they've listened to a control sound.' This indicates 'listening to laughter, leads to them expecting more positive things.'

To truly understand the cognitive capabilities and 'understanding' of Large Language Models (LLMs), researchers must apply rigorous methods from comparative cognition and developmental psychology. These fields specialize in assessing intelligence in non-verbal subjects (animals, young children) and can help probe beyond an LLM's verbal output to determine if it genuinely understands or merely predicts responses.

Cartmill advocates for 'applying more sort of guidelines and methods from animal communication or animal cognition to the study of AI and LLM specifically' because 'we can't just take them at their word.'