Why protecting incumbents is the cardinal sinWhy protecting incumbents is the cardinal sin
The dominant instinct of every government facing technological pressure is to protect what currently exists. The taxi medallion. The radiologist's reading queue. The voice-actor's residuals. The newspaper. This instinct feels like compassion. Darwin would call it the most dangerous thing you can do, and he would be precise about why.
Consider his actual mechanism. A population carries variation. Selection pressure culls the unfit and concentrates the fit. The rate at which this happens determines how fast the population tracks a moving environment. Now intervene: artificially shield the unfit variants from the pressure. What have you done? You have not made them fit. You have decoupled the population from the environment and let the mismatch accumulate. The environment keeps moving; the protected configuration does not. The gap between them — call it adaptive lag — grows silently behind the shield, and when the protection finally fails (it always fails, because you cannot subsidize a mismatch forever), the correction does not arrive gently. It arrives all at once, sized to the entire accumulated gap.
This is not a metaphor I'm stretching. It is the literal logic of suppressed adjustment, and it shows up everywhere selection is dammed rather than resolved. Kodak invented the digital sensor in 1975 and spent twenty-five years protecting the film business from its own discovery. The pressure did not go away. It compounded in silence and then took the entire company in a span that looked, from outside, instantaneous. The protection did not save the workers. It deferred and concentrated their extinction.
At Bridgewater we had a blunter version of this: you don't avoid a deleveraging by refusing to mark your book to market. You just convert a slow, survivable repricing into a sudden, fatal one. Minsky said the same thing about whole economies — stability suppressed long enough manufactures the instability that ends it. Suppressed adjustment is not safety. It is the manufacture of fragility, paid for later with interest.
So the most destructive AI policy on the table — the licensing regimes, the compute thresholds designed to freeze the current frontier, the sector-by-sector slowdowns sold as "responsible deployment" — is destructive precisely because it works in the short run. It lowers the selection pressure. It keeps the unfit configuration alive a little longer. And every quarter it does so, the gap between the protected economy and the real fitness landscape widens.
The dominant instinct of every government facing technological pressure is to protect what currently exists. The taxi medallion. The radiologist's reading queue. The voice-actor's residuals. The newspaper. This instinct feels like compassion. Darwin would call it the most dangerous thing you can do, and he would be precise about why.
Consider his actual mechanism. A population carries variation. Selection pressure culls the unfit and concentrates the fit. The rate at which this happens determines how fast the population tracks a moving environment. Now intervene: artificially shield the unfit variants from the pressure. What have you done? You have not made them fit. You have decoupled the population from the environment and let the mismatch accumulate. The environment keeps moving; the protected configuration does not. The gap between them — call it adaptive lag — grows silently behind the shield, and when the protection finally fails (it always fails, because you cannot subsidize a mismatch forever), the correction does not arrive gently. It arrives all at once, sized to the entire accumulated gap.
This is not a metaphor I'm stretching. It is the literal logic of suppressed adjustment, and it shows up everywhere selection is dammed rather than resolved. Kodak invented the digital sensor in 1975 and spent twenty-five years protecting the film business from its own discovery. The pressure did not go away. It compounded in silence and then took the entire company in a span that looked, from outside, instantaneous. The protection did not save the workers. It deferred and concentrated their extinction.
At Bridgewater we had a blunter version of this: you don't avoid a deleveraging by refusing to mark your book to market. You just convert a slow, survivable repricing into a sudden, fatal one. Minsky said the same thing about whole economies — stability suppressed long enough manufactures the instability that ends it. Suppressed adjustment is not safety. It is the manufacture of fragility, paid for later with interest.
So the most destructive AI policy on the table — the licensing regimes, the compute thresholds designed to freeze the current frontier, the sector-by-sector slowdowns sold as "responsible deployment" — is destructive precisely because it works in the short run. It lowers the selection pressure. It keeps the unfit configuration alive a little longer. And every quarter it does so, the gap between the protected economy and the real fitness landscape widens.