Complexity & Emergence

The phenomenon

Six or seven neighbors, and nothing else

Start with what we actually know, in the careful Day 1 sense. In the mid-2000s, the STARFLAG project put stereoscopic cameras on a rooftop museum in Rome, photographed real starling flocks of up to 2,700 birds, and reconstructed the 3-D position of every single one. Then they asked a deceptively simple question: when a bird adjusts to its neighbors, which neighbors?

The intuitive answer is metric: a bird reacts to everyone within, say, three meters. The data said no. Birds track a fixed number of neighbors: six to seven, regardless of how near or far those neighbors happen to be. Pack the flock tighter and each bird still watches the same six or seven; spread it out and it watches the same six or seven, now farther away. The interaction is topological, not metric: it counts neighbors, not meters.

“Each bird interacts on average with a fixed number of neighbors (six to seven), rather than with all neighbors within a fixed metric distance.” Ballerini et al., PNAS, 2008.

Why does this matter? Because a topological rule is exactly what you would want if the goal were to hold a group together under attack. When a falcon stabs into the flock and local density explodes, a fixed-radius bird would suddenly track dozens of neighbors and lose the thread. A “watch my nearest seven” bird keeps its cohesion no matter how density warps. The rule is robust. And it is purely local. No bird knows the flock’s shape. No bird is solving for the global pattern. The pattern is a side effect of thousands of birds each minding a tiny neighborhood.

The model

Three rules, written in 1986

The simulation is essentially Boids, built by Craig Reynolds in 1986 and presented at SIGGRAPH in 1987. Reynolds wanted realistic flocking for animation and discovered he needed three local rules: separation (do not crowd), alignment (steer with neighbors), and cohesion (stay with the group). From those three, the repertoire of flocking falls out: wheeling, splitting around obstacles, fluidly merging again. Boids has since flocked movie bats and countless digital armies. It is the canonical demonstration that complex global behavior need not have a complex global cause.

And Boids is one specimen of the same trick: simple parts, surprising wholes.

• Ant colonies find short paths to food with no surveyor and no map. Each ant lays a chemical trail and follows others’ trails; shorter routes get reinforced faster. Biologists call this stigmergy: coordination through traces left in the environment rather than direct command.
• A pot of water can spontaneously organize itself. Heat it from below and, past a threshold, smooth liquid breaks into ordered convection cells: structure sustained by energy flow, a preview of the physics-of-life arc on Days 83-85.
• Your brain is reading this sentence with no homunculus inside doing the reading. Meaning emerges from neurons that individually understand nothing, the steep version of the puzzle we return to on Days 123-126.

When the parts compute

The maddening case of the slime mold

If the murmuration is emergence as beauty, slime mold is emergence as intelligence arising in something with no brain. Physarum polycephalum is a single giant cell, a sprawling yellow blob with protoplasm and an appetite for oats.

In 2010, a team led by Toshiyuki Nakagaki and Atsushi Tero laid out oat flakes in the geographic pattern of Tokyo and 35 surrounding towns, then dropped a slime mold on Tokyo. The blob spread out to engulf the food, then pruned itself. Redundant tubes withered; efficient ones thickened. Within about a day it had sculpted a network similar in total length, efficiency, and fault tolerance to the actual Tokyo rail system.

The model “quantitatively mimics phenomena that can be neither captured nor quantified by verbal description alone.” Wolfgang Marwan, Science Perspective, 2010.

The slime mold is not clever. It is running a simple local rule: reinforce tubes where protoplasm flows well, abandon the rest. Applied everywhere at once, that rule computes a near-optimal network. The colony solves a problem no ant understands. The blob designs a railway it cannot conceive of. The computation lives in the interactions, not in the components.

The model, sharpened

The phrase that launched a field

For a long time, emergence was a word physicists said with suspicion: vague, faintly mystical, the sort of thing you invoked when you lacked equations. Then in 1972, the condensed-matter physicist Philip W. Anderson published a short essay in Science with a title that became a slogan: “More Is Different.”

Anderson’s target was reductionism: the assumption that once you know the fundamental laws, everything else is “just” applied physics. He did not deny reductionism. He granted it, then drove a wedge through its smug corollary.

“The ability to reduce everything to simple fundamental laws does not imply the ability to start from those laws and reconstruct the universe… at each level of complexity entirely new properties appear.” P. W. Anderson, Science, 1972.

Knowing how two water molecules interact does not hand you wetness, fluidity, convection cells, or the exact way a tsunami breaks. Those are real, lawful properties of collectives. New levels of organization have their own rules, concepts, and sciences. Chemistry is not merely physics in inconvenient notation; biology is not merely chemistry with extra vocabulary. Each level is emergent from the one below and yet, in a meaningful sense, autonomous from it.

The debate

Weak vs. strong: the fault line

Everyone agrees murmurations and slime molds are emergent. The fight is over what that word is allowed to mean.

Weak emergence: the scientist’s version

The philosopher Mark Bedau gave the careful account most working scientists can use. A property is weakly emergent if you cannot predict it from the rules by a clever shortcut: the only way to find out what the system does is to run it or simulate it. The murmuration is weakly emergent. There is no simple formula that takes the three Boids rules and spits out tomorrow’s flock shape; you have to let it fly. Crucially, weak emergence is not spooky. It is fully grounded in the parts and their interactions.

Strong emergence: the radical version

A property is strongly emergent if facts about the whole are in principle not deducible from complete knowledge of the parts, not even by a godlike simulator with unlimited compute. The high level would possess genuinely new causal powers, including downward causation. David Chalmers argues that there may be exactly one candidate for strong emergence in nature: consciousness. Everything else looks weakly emergent: surprising, irreducible in practice, but not irreducible in principle.

Why not embrace strong emergence everywhere? Because it collides with Jaegwon Kim’s causal exclusion problem. Suppose a high-level state causes a physical effect. But that effect already has a complete physical cause at the level of parts. Either the high-level cause is redundant, the effect is overdetermined, or the claim violates the physics. Defenders have answers, but no knockout. This is why the strong/weak line remains live under biology, mind, and social systems.

Sit with what the Game of Life proves. Two rules, applied to a grid, are enough to build logic gates, memory, clocks, and a universal computer. The “computer” is nowhere in the rules; it is a pattern in the cells. This is the cleanest illustration of weak emergence: you can know the rules perfectly and still have no way, short of running them, to know whether a starting pattern will fizzle, freeze, repeat, or compute. Simple law, unboundedly complex behavior, no shortcut.

A necessary detour

Nobody can agree how to measure “complex”

Here is an embarrassing secret of complexity science: there is no single agreed definition of complexity. Seth Lloyd once compiled a famous list of dozens of proposed measures, and they do not agree with each other. This is not mere sloppiness. “Complex” is doing several jobs at once: how hard is it to describe, how hard is it to build, and how organized is it?

That last family hides the key idea. The most complex things are not the most random ones. Consider a crystal, a living cell, and a canister of gas. By a pure “hard to describe” measure, the gas wins because perfect randomness needs the longest description. But a gas is not more complex than a cell. A crystal is too ordered; a gas is too disordered. Complexity lives in the structured, surprising, neither-frozen-nor-random middle.

Open questions

What’s genuinely unsettled

• Is strong emergence real anywhere? Or is everything weakly emergent, surprising but reducible, with Kim’s exclusion argument quietly winning?
• Is downward causation a real force or a figure of speech? Can a whole genuinely push its parts around in a way the micro-laws do not already determine?
• Is there one principled measure of complexity? Or will we always have a drawer full of measures, each right for a different job?
• Does Assembly Theory carve nature at a new joint? Or is it useful compression and information theory wrapped in bigger claims than it can cash?
• What about AI emergence? When a large language model seems to gain a new skill at scale, is that genuine emergence or an artifact of the metric? Hold that thought for Day 139.