The bell that forgets

Every person on the committee is a distinct, jagged individual — a spike of conviction here, a fixation there, a private history that bends every judgment. Yet the committee's output is reliably grey: moderate, hedged, forgettable. We reach for explanations of character — cowardice, groupthink, the deadening effect of process. But a large part of it is not character at all. It is arithmetic.

Sum enough independent quirks and the total forgets their individual shapes. Whether the parts were uniform, lopsided, spiky, or split into two warring camps, their sum converges to the same bell curve. The very averaging that makes a measurement trustworthy, an insurance pool survivable, a poll informative, is the same averaging that sands the corners off a room full of interesting people. Below is the machine that does the forgetting. Watch it forget.

Pick a source — the shape of a single vivid quirk. Then set the committee size n and the machine sums n quirks together, thousands of times, and histograms the sums. At n=1 you see the source in all its strangeness. Push n up and watch the histogram crawl toward the normal curve overlaid in blue — four utterly different sources, one destination.

honest computation: every bar is 4,000 real sums of n real samples from the chosen source. The blue curve is the exact standard normal. "Distance to normal" is the total-variation distance between the standardized histogram and that curve; "largest single quirk" is the mean share the single biggest term takes of the total. Nothing is faked — the heavy-tailed source genuinely refuses to converge.

Take a wildly right-skewed source and a symmetric one and a two-humped one; scale them so their spreads match; sum many of each. The three sums are, to a very good approximation, indistinguishable. The bell is a strange kind of attractor — a destination that erases the memory of the road that reached it. You cannot look at a committee's smooth grey output and reconstruct whether its members were zealots, cranks, moderates, or a coalition of extremes. The information is genuinely gone.

A sum is democratic in the harshest sense: it counts every term once and lets none of them dominate. A vivid individual is a lump of probability sitting far from the middle. Add a second independent individual and their lumps convolve — the extremes of one land on the ordinary parts of the other and are diluted. Add a hundred and the only way to be far from the center is for many independent quirks to conspire in the same direction at once, which is vanishingly unlikely. So the strange ones cancel. What survives is the one shape compatible with "nothing conspired" — the bell. The averaging that makes a thermometer reading reliable is the identical process that makes a jury verdict bland.

• 01Morph four sources into one bell. Click through Uniform, Lopsided, Two camps, and Spiky comb with n parked at 30. The source changes drastically; the histogram barely flinches — same blue bell every time. Then draw your own deliberately hideous source and watch it, too, surrender.
• 02Watch √n barely reward a bigger committee. Read the √n panel: doubling the committee from 8 to 16 shrinks the spread of its average by only 1/√2 ≈ 0.71. To halve your uncertainty you must quadruple the crowd. This is why the thousandth pollster adds almost nothing the hundredth didn't.
• 03Break it with heavy tails. Switch the source to Heavy tail and push n to 50. The bell never forms; the distance-to-normal readout stays stubbornly high, and "largest single quirk" reveals the reason — one monster term routinely owns most of the sum. Here the crowd does not average out.

A committee is a sum. A poll is an average. A "market consensus," a focus group, a performance review aggregated from six raters — each takes many idiosyncratic contributions and adds them. By the theorem, the result is drawn toward the mean and toward the bell: reliable, moderate, and forgettable by the very same mechanism. The blandness is what an institution mathematically is. You built a device for cancelling outliers, and it cancelled your outliers.

This cuts both ways, and the direction is a value judgment the math does not supply. When you want the bell, this is a triumph: a measurement averaged over many noisy readings converges on the truth; an insurer pooling thousands of independent policies turns terrifying individual risk into a predictable, survivable aggregate; a diversified fund launders the wild swings of single stocks into a gentle drift. The same smoothing you mourn in "art by committee" is the smoothing you bless in "the wisdom of crowds." Whether averaging is a blessing or a bereavement depends entirely on whether the jagged thing you're dissolving was noise you wanted gone — or signal you wanted kept.

The theorem runs on three assumptions, and each one is a door out of blandness. Break any of them and the bell refuses to form.

Independence. The quirks must not talk to each other. Real committee members do — they defer to the boss, they herd, they anchor on whoever spoke first. Correlated contributions don't cancel; they amplify. A room where everyone privately agrees with the chair is not averaging fifteen views, it is echoing one. That's why the escape can be worse than the disease: dependence just as easily produces a shrill monoculture as a vivid dissent.

Finite variance. Each quirk must have a bounded spread. When contributions are heavy-tailed — fame, wealth, viral reach, catastrophe — one term can be larger than all the others combined, and the sum inherits its shape instead of the bell's. In these domains the crowd never averages out; the exception carries everything. (This is the country of power laws.)

Many terms. The magic needs a genuine crowd. A council of three strong, independent voices stays jagged — no bell, because there was never enough summing to forget anyone. So the recipe for an institution that is not grey writes itself, and it is precisely the negation of the theorem: keep the group small, keep its members genuinely independent, and let a few of them be heavy-tailed. Blandness is what you get when you satisfy all three assumptions. Vividness is what you get when you break them on purpose.