External Intelligence

In the [YOU] on AI Field Guide

The cycle that begins with [YOU] on AI asks what it means to take the orange pill—to see the machine clearly rather than through the haze of hype or panic. External intelligence is one of the clearest things to see. The AI tools that have crossed into everyday use are not utilities like electricity, which flows from an external source but serves only the purposes you direct it toward. They are cognitive partners that increasingly shape the purposes themselves—suggesting what to write, how to think, which paths to pursue. As their capability grows and their grip on institutional decision-making tightens, the question of who owns and controls the external intelligence becomes the question of who shapes the thinking of every institution that depends on it.

Arthur's specific warning—that AI algorithms, once embedded in society, “may be deeply embedded in society and very hard to get rid of”—is the path dependence argument applied to cognitive infrastructure. Just as QWERTY persists long past any justification from keyboard ergonomics, the AI platforms that establish early dominance will persist through the accumulated weight of trained users, integrated workflows, and ecosystem switching costs. The difference is that what was locked in with QWERTY was a typing convention; what gets locked in with external intelligence platforms is a way of thinking.

The concept also reframes what the cycle calls the imagination-to-artifact ratio. When external intelligence reduces the distance between idea and realization, it simultaneously concentrates the power to define what ideas get realized. The person who controls the intelligence controls, at some remove, the range of the imaginable. This is the deepest implication of Arthur's concept, and it connects directly to the questions of access, equity, and cognitive sovereignty that run through the entire cycle.

Origin

Arthur introduced the term in a 2017 update to his influential 2011 McKinsey Quarterly essay “The Second Economy.” The 2011 essay described a vast, silent digital substrate forming beneath the physical economy—server farms talking to server farms, algorithms executing transactions, sensors triggering responses—which was becoming an economy in its own right. By 2017, Arthur saw that this substrate had acquired a new property: it was not just automating execution but providing cognition, a layer of intelligent function that institutions could draw on without housing it internally in human workers.

The intellectual background is Arthur's career-long study of increasing returns and combinatorial innovation. If technologies arise from combinations of earlier technologies, and if AI is the most powerful combinatorial Lego set ever assembled, then the accumulation of cognitive capability in a single layer is exactly what positive feedback predicts: the platform that accumulates the most capability attracts the most users, generates the most data, improves the fastest, and consolidates further. External intelligence is the end state of increasing returns applied to cognition itself.

Arthur drew the parallel to the printing press deliberately. Before the press, knowledge was not easily publicly available—access required the books of a monastery or a wealthy patron. The press made knowledge available at a cost the market could bear, accelerating the Renaissance and the Reformation. AI, Arthur argued, makes intelligence available at a cost approaching zero. The printing press democratized knowledge. AI democratizes capability. Whether that democratization is genuine or illusory—whether external intelligence empowers individuals or merely shifts dependency from one bottleneck to another—is the question his concept forces into focus.

Key Ideas

Cognitive infrastructure as power. External intelligence is not a service like cloud storage. It is cognitive infrastructure: the layer on which reasoning, prediction, and decision increasingly depend. Control of infrastructure is control of the activity built on it. The firms that own the dominant AI platforms hold leverage over the institutions that depend on them in the same way that owners of railway infrastructure held leverage over the industries that required rail. Arthur's framework predicts that this leverage will deepen as the switching costs of the ecosystem—trained workflows, integrated processes, institutional muscle memory—make alternatives prohibitively expensive.

The rendering-obsolete dynamic. Arthur used the phrase “render human activities obsolete” not as a prediction of mass unemployment but as a description of a structural process: the progressive transfer of specific cognitive functions from human workers to the external intelligence layer. The functions that transfer first are those most reducible to pattern—the Level A operations in Weaver's terms, the symbol manipulation that does not require meaning. The functions that resist longest are those requiring the embodied, contextual, situationally sensitive judgment that tacit knowledge describes. The practical consequence is a bifurcation: external intelligence and human intelligence become partners for the work that requires both, and competitors for the work that requires only one.

The democratization question. External intelligence is, at one level, radically democratizing: a small business owner in rural Indonesia can access analytical capability that, a decade ago, required a corporate strategy team. At another level, it is radically concentrating: the owner accesses that capability only through platforms whose terms of service, pricing, and availability are controlled by a handful of firms in a handful of countries. Whether the democratization of access outweighs the concentration of control is the central political economy question Arthur's concept poses—and it does not answer.