Proximate and Ultimate Causes

In the AI Story

The distinction was developed in twentieth-century evolutionary biology, notably by Ernst Mayr, to separate mechanism-level explanations (how does this behavior work?) from evolutionary-level explanations (why did this behavior evolve?). Mayr argued that confusing the two produced systematic analytical errors: proximate answers to ultimate questions miss the causal structure, and ultimate answers to proximate questions miss the mechanism.

Diamond adapted the distinction for historical analysis in Guns, Germs, and Steel (1997). The proximate cause of Spanish victory over the Inca was specific: Pizarro's horses, steel weapons, and the smallpox that had already devastated indigenous populations. But these proximate factors required explanation — why did Europeans have horses, steel, and exposure to smallpox while the Inca did not? The ultimate causes, Diamond argued, ran back through millennia of agricultural history, continental geography, and the east-west versus north-south orientation of continental landmasses that had determined the spread of domesticable species.

The analytical power of the distinction is that it forces explanatory depth without allowing either level to collapse into the other. Proximate causes are not 'mere' mechanisms — they are real, specific, and historically consequential. Ultimate causes are not 'deep' truths that render proximate causes irrelevant — they are structural conditions that shape the space of proximate possibilities. Both matter; neither is sufficient.

Applied to the AI transition, the distinction produces specific analytical clarity. The proximate cause of the December 2025 capability threshold was specific: Claude Code's improvements in tool use, the architectural advances in agentic systems, the training run completions at frontier labs. The ultimate causes run deeper — the accumulation of computing infrastructure in a handful of jurisdictions, the historical concentration of research capacity in specific institutions, the positive feedback loops that Diamond's framework identifies as the structural mechanism of divergence. The proximate picture looks like a technology breakthrough; the ultimate picture looks like the continuation of thirteen thousand years of geographic and institutional accumulation.

Origin

Mayr formalized the distinction in his 1961 Science paper 'Cause and Effect in Biology,' which distinguished four causal levels (mechanism, ontogeny, phylogeny, adaptation) and argued that biological explanations require attention to all four. Diamond, trained as a physiologist and evolutionary biologist before turning to historical questions, brought the distinction into geography and history systematically in Guns, Germs, and Steel.

The adaptation of the distinction to technology policy analysis is consistent with Diamond's method even when not explicitly made by Diamond himself. The framework forces analysts to distinguish between the specific technological events (proximate) and the historical, institutional, and geographic conditions that made those events possible (ultimate).

Key Ideas

Both causal levels are real. The distinction is analytical, not hierarchical — proximate causes are not 'mere' mechanisms and ultimate causes are not 'deeper' truths.

Different questions require different answers. 'What happened?' calls for proximate explanation; 'why was this possible?' calls for ultimate explanation; confusing the two produces systematic error.

Ultimate causes shape proximate possibilities. The space of proximate events available to a society at any given moment is constrained by the ultimate conditions — geographic, institutional, cultural — that have accumulated over centuries or millennia.

AI requires both levels of analysis. Technology policy that addresses only proximate events (specific capabilities, specific products) misses the structural conditions that will shape future proximate events.

The framework resists technological determinism. By foregrounding the geographic, institutional, and historical conditions that shape technological possibility, the distinction avoids the flattening narrative in which technology simply 'happens to' societies.

Debates & Critiques

Some scholars have argued that the distinction is too easily manipulated — that any proximate cause can be attributed to some ultimate cause and vice versa, producing circular explanation. Diamond's response has been empirical: when applied to comparative cases, the distinction generates testable predictions about which societies face which outcomes under which conditions, and the predictions have held up across his archaeological archive. The contemporary application to AI is uncontroversial as method but contested as application — analysts disagree about whether specific AI capabilities should be treated primarily as proximate events of transient significance or as proximate expressions of ultimate structural change.