This essay grew out of a conversation that began with a simple naming question: whether “killer whale” and “orca” referred to different animals. From there, the discussion moved quickly from terminology to behaviour, especially the striking contrast between wild orcas and captive incidents involving humans, and then to the broader question of where orcas sit in the marine food web.
The first major conceptual turn came when the comparison shifted from orcas as predators to orcas as mobile social groups. Their long-range movement, cooperative hunting, learned traditions and stable social structures suggested a useful comparison with human hunter-gatherer bands. That comparison raised a more difficult question: if humans coordinating in groups are assumed to be communicating richly, why are similarly coordinated animal groups so often described in thinner terms such as “calls”, “signals” and “instinct”?
The discussion then widened to birds, especially corvids and parrots, because research over the past half-century has repeatedly shown that sophisticated cognition can arise in brains organised very differently from the human brain. This weakened the old intuition that intelligence must look anatomically or behaviourally human in order to count as intelligence.
A second major turn came from thinking about how an outside observer would judge unfamiliar intelligence. The conversation used the example of isolated or recently contacted human groups to expose an asymmetry in our assumptions. When humans speak an unknown language, we assume there is a rich system present even when we cannot decode it. With animals, researchers have often been more willing to treat undecoded communication as simple until complexity is proven.
That led directly to large language models. Neural language systems represent text as high-dimensional numerical states. Individual numbers do not carry intrinsic meanings, yet relations among large patterns of numbers can encode stable semantic and syntactic structure. Older word-embedding examples such as the relationship between “king”, “queen”, “man” and “woman” made the point especially concrete, while multilingual models suggested that different surface languages can converge on partially aligned internal representations.
From there, the essay’s central proposal emerged: animal communication might be studied less like a dictionary problem and more like a representation-learning problem. Rather than starting with labels such as “food”, “danger” or “turn left”, one could record vocalisations together with the full behavioural and environmental context, then ask what latent structure appears and how strongly it predicts subsequent action.
The walkie-talkie analogy sharpened this idea. If a group of humans performed a complex hunt while exchanging radio messages, researchers would naturally treat the radio traffic as a likely part of the coordination mechanism. The essay uses that asymmetry to question why equally structured acoustic exchanges among orcas are not more readily investigated as possible conversation.
Wittgenstein then supplied the philosophical bridge. His later view that meaning lies in use rather than in an intrinsic property of words fit naturally with the machine-learning picture. If meaning is relational and embedded in practice, then understanding another species may require reconstructing the structure of its use rather than identifying one-to-one translations.
The conversation then moved from communication to emergence. Human intelligence is amplified by external memory: writing, mathematics, libraries, institutions, software and accumulated culture. Once that is recognised, comparisons between “human intelligence” and animal intelligence become less straightforward, because a modern human is usually being evaluated together with an enormous inherited cognitive infrastructure.
The same perspective was then applied back to animals. Orca hunting traditions, birdsong, tool use and other socially transmitted behaviours can be understood as forms of knowledge stored partly in groups rather than solely in individual nervous systems. This made culture and collective intelligence central to the essay rather than side observations.
The final turn concerned individual differences among humans. The conversation rejected both the idea that all people are cognitively interchangeable and the idea that “intelligence” must be treated as a single mysterious substance. The cleaner formulation was simply that people vary in information-processing performance: in how reliably they extract structure, predict outcomes, learn, generalise and correct mistakes. Culture changes which abilities are visible and rewarded, but does not erase individual variation.
The finished essay was deliberately written without the technical vocabulary, diagrams, labels and working concepts used during the conversation. It presents only the public-facing argument: that the boundary between human and animal intelligence may have been drawn partly by historical assumptions, that AI provides a new computational analogy for meaning and representation, and that the most interesting next step is empirical—record enough of animal communication in context to discover its structure before deciding in advance what it can or cannot be.