Wetness is not a property of any individual H₂O molecule. Consciousness is not a property of any individual neuron. Market prices are not a property of any individual trader. Traffic jams are not a property of any individual car. In each case, a qualitatively new property appears at the aggregate level that cannot be straightforwardly read off from the parts. Philosophers and scientists have used the word emergence for this phenomenon since at least G. H. Lewes in 1875, but the term has carried very different meanings in different mouths. The 20th century split it into two: weak emergence — aggregate behavior is unexpectedly hard to predict from the parts, but in principle derivable from lower-level laws; strong emergence — aggregate behavior follows new laws not derivable from lower-level laws even in principle. The first is uncontroversial — every domain in this curriculum exhibits it. The second is philosophically contested — most physicalist philosophers reject it; some philosophers of mind accept it for consciousness. The polymath-relevant point is to keep these distinctions clear: emergence is a real and important pattern, but strong emergence is a much stronger claim, and many popular invocations of the word are motte-and-bailey moves between the two.
Weak emergence — the uncontroversial kind that runs through nearly every domain — has a few characteristic features. Aggregate properties show up that are not present at the component level: liquidity is not a property of single H₂O molecules, market clearing is not a property of any individual trader, and a glider in Conway's Game of Life is not a property of any single cell. Multiple realizability allows the same aggregate property to be implemented on radically different substrates: computation runs equally on silicon, neurons, billiard balls, or DNA, and markets work with corn, currency, or kidneys. There is also some sense in which the aggregate level constrains what the components do — a market price is caused by trader behavior but also causes trader behavior — though whether this downward causation is genuine causation or just convenient description is a philosophical question. The defining feature, in practice, is that the collective dynamics are hard to predict even with full knowledge of individual behavior. Conway's Life with its four local rules and Reynolds's Boids with its three (separation, alignment, cohesion) are the canonical small demonstrations that simple components plus simple interactions can produce surprise.
The philosophically loaded question is whether strong emergence is also real — whether some aggregate properties follow new laws not derivable from lower-level laws even in principle. C. D. Broad argued for it in the 1920s; some contemporary philosophers of mind keep the option open for qualia or biological function; most analytic philosophers and most working physicists reject it as a confession of ignorance about the underlying physics. The argument has not closed in a hundred years and probably will not close soon, because it is intertwined with the Hard Problem of Consciousness and with what counts as scientific explanation. The practical lesson is to keep the two senses straight. Weak emergence is a description of computational and explanatory difficulty — real, important, and not philosophically loaded. Strong emergence is a metaphysical claim with consequences for what reduction can do, and many popular invocations of "emergence" are motte-and-bailey moves that defend the weak claim and reach for the strong one. When someone says X emerges from Y, the first question is which kind they mean.
The word has surged back into circulation around artificial intelligence. Large language models display emergent abilities — capacities like arithmetic or chain-of-thought reasoning that appear abruptly at scale and were absent in smaller models — and a 2022 wave of papers treated this as a deep fact about scaling. A 2023 rebuttal (Schaeffer and colleagues) argued that much of it is a mirage: an artifact of harsh all-or-nothing metrics that hide smooth underlying improvement. The dispute is the weak-versus-strong distinction playing out in real time — is the new behavior merely hard to predict from the parts, or genuinely irreducible? The same question shadows debates over machine consciousness, the origin of life, and AI alignment, where the stakes of getting the distinction right are no longer merely academic.