CONCEPT
The Knowledge-Acquisition Bottleneck
Feigenbaum's diagnosis of the structural limit of hand-coded expert systems: the agonizing slowness and incompleteness of extracting expert knowledge into machine-readable rules—the precise obstacle that deep learning bypassed, honoring the knowledge principle by abandoning the method.
The knowledge-acquisition bottleneck names the binding constraint that
Feigenbaum identified within his own program: to build an expert system, you had to extract knowledge from a human expert and encode it as explicit rules, and this process was agonizingly slow, expensive, and structurally incomplete. The difficulty was not merely tedious. Much of what an expert knows is
tacit—compiled into intuition over years of practice, unavailable even to the expert for conscious inspection. The knowledge engineer sat with the expert, watched them work, and tried to reconstruct implicit reasoning as explicit if-then rules. The reconstruction was always partial, always brittle at the edges where the expert's implicit understanding covered ground the rules could not reach. As knowledge bases grew, their rules began to interact in ways no one fully understood, making them fragile outside their narrow domains and expensive to maintain. Feigenbaum named this bottleneck precisely because he understood it was the binding constraint on the entire expert-systems enterprise—and naming