Emergent Behavior or emergence occurs when an entity is observed to have certain properties , that its individual parts do not have on their own. Certain properties or behaviors might come about only when the parts that make up that entity interact as a wider whole , or when the number of parts making up that entity reach a certain minimum threshold.

It is hypothesized that an AI system might display emergent properties simply as a function of scale. This is supported by the fact that various complex systems exist in nature that have displayed such emergent properties due to quantitative differences. Examples include :

Heart : Individually cells cannot pump bood whereas the whole heart can.
Uranium : Small amounts, nothing special, large amounts = nuclear reactions
Civilization : Individuals, nothing special, collective specialization = human civilization

This raises concerns that due to the quantitative increases in the size of AI models , we might soon hit an unexpected threshold and see unexpected differences in qualitiative behaviors and capabilites. This provides motivations to study complex systems with emergent phenomena to be able to draw inferences for when , and what types of capabilites might emerge.

References