There is a threshold beyond which quantity becomes quality. In sufficiently large learning systems, capabilities appear that are absent in smaller ones and were never explicitly designed. Emergence, long a theme in the study of complex systems, has entered the study of machines.
The character of emergence
Emergence names the abrupt arrival of an ability with scale. A capacity absent at one size appears, seemingly discontinuously, at another. The whole exceeds the sum of its parameters.
Why it unsettles
The phenomenon unsettles because it resists prediction. If capabilities arrive unbidden, the behavior of a system cannot be fully anticipated from its parts. Control presupposes foresight that emergence denies.
Continuity beneath the surface
Some argue the discontinuity is an artifact of measurement. Smooth underlying improvement can appear abrupt when judged by a threshold metric. The debate concerns whether the leap is real or apparent.
Scale as a variable
What is not in dispute is the centrality of scale. Size, data, and computation together act as a dial that summons behavior. The dial is powerful and poorly understood.
The interpretive burden
Emergence imposes an interpretive burden. To trust a system one must understand it, yet emergent competence outruns explanation. The gap between capability and comprehension widens.
A humbling prospect
The prospect is humbling. If intelligence can arise from scale in ways we cannot foresee, our theories trail our artifacts. We build faster than we understand.
The larger point
Emergence describes capabilities that appear with scale rather than by design. Whether abrupt or merely so in appearance, it strains prediction and control. It marks the point where our machines outpace our theories of them.
