Before AI, intelligence was housed inside human workers. Institutions purchased that intelligence through employment contracts, coordinating it through management hierarchies. Arthur identified a fundamental shift: AI provides intelligence that is external—residing not in employees but in the digital substrate, accessed through interfaces rather than hired through contracts. This externalization transforms the nature of organizational capability. The institution's cognitive capacity becomes detachable, scalable independently of headcount, accessible on-demand. The competitive advantage shifts from possessing human intelligence to effectively directing external intelligence toward productive ends.
Arthur's 2017 McKinsey piece specified the mechanism: 'Business processes can now draw on vast libraries of intelligent functions that greatly boost their activities—and bit by bit render human activities obsolete.' The phrasing was precise—not that AI makes humans obsolete but that relying on external intelligence renders internal human activity obsolete. The difference is structural: the intelligence remains essential, but its location migrates from inside the organization to outside it, from embodied in workers to accessed through subscription.
The externalization creates novel organizational dynamics. When intelligence is internal, scaling requires hiring—more capability means more people, more people mean more coordination costs, more coordination costs limit scaling. When intelligence is external, scaling is nearly frictionless—adding capability requires purchasing more API calls, not navigating labor markets, training new employees, managing larger teams. The coordination constraint that limited organizational growth for a century has been relaxed, enabling individual builders to operate at scales that previously required institutions.
But externalization also creates dependencies that internal intelligence never produced. The organization relying on external intelligence no longer controls its own cognitive capacity. Platform availability, pricing changes, model updates, policy shifts—each affects capability without the organization's consent. The intelligence is more scalable but less controllable, producing a new species of organizational fragility masked by surface productivity. The institution optimized for external intelligence may discover during a platform outage that it has outsourced capabilities it can no longer provide internally.
Arthur connected external intelligence to his broader argument about technology becoming biological: 'Technology is developing sensory capabilities, interconnectedness, and learning capacity that increasingly resemble living systems.' The trajectory from rule-based expert systems to self-improving neural networks to autonomous agents is technology becoming alive—'so diverse, so distributed, that they cannot be managed in a top-down manner, but must now be taught to learn from their experience.' External intelligence is not a static library but an evolving ecosystem, and the institutions depending on it are in symbiotic relationship with something that has its own developmental trajectory.
The concept emerged from Arthur's decade-long study of the second economy as it developed through the 2000s and 2010s. Initially, the external intelligence was narrow: recommendation algorithms, fraud detection, inventory optimization. Each was a point solution addressing a specific domain. The qualitative shift came when large language models crossed the threshold to general cognitive capability—handling not merely specialized tasks but the broad range of reasoning, analysis, and creative production that had been human-exclusive.
Arthur drew on his Santa Fe Institute work studying distributed intelligence in complex adaptive systems. The key insight: intelligence distributed across a network can produce capabilities exceeding what any node possesses individually. Applied to AI, this meant external intelligence was not merely an aggregate of specialized tools but a genuinely new form of cognitive infrastructure—networked, learning, adaptive—that institutions would come to depend on as they depend on electrical power: always available, rarely thought about, catastrophic when absent.
Intelligence has been externalized. Cognitive capability once housed in human workers now resides in the digital substrate, accessed rather than employed.
Scaling becomes frictionless. Adding capability no longer requires hiring and coordination but purchasing access, relaxing the constraint that limited organizational growth.
Dependency replaces ownership. Organizations gain scalability but lose control, creating novel fragilities masked by surface productivity.
External intelligence evolves autonomously. The substrate is not static but learning, adaptive, developing according to its own trajectory rather than institutional direction alone.
The intelligence gap widens structurally. Institutions with sophisticated external-intelligence integration pull ahead of those relying on internal human capability, and the gap compounds through positive feedback.