AI Safety

Why Brackets Fail: Inefficiency in Regulation from Speed Limits to Tax Codes

From 18th-century hanging to modern tax brackets — why we need to rethink categorical regulation.

Deep Dive

Regulation often requires treating different degrees of a trait differently — going 1 mph over the limit versus 150 mph in a 40-zone are both 'speeding,' yet warrant vastly different punishments. The default solution is to chop continuous distributions into discrete brackets and apply uniform rules within each. This works poorly.

Brackets create inefficiency because the top and bottom of each bracket experience the same rule differently. For example, a 20kg airline baggage limit charges the same fee for 20.001kg and 22kg, incentivizing travelers to pack right up to the threshold. In 18th-century England, the death penalty applied to theft of 12 pence or more, which meant petty thieves had no marginal disincentive against murder — they already faced hanging. This led to escalation and juries lying about theft values to avoid the threshold.

Similar dynamics appear in tax brackets, insurance deductibles, and software subscription pricing. The essay argues for graduated, continuous regulation instead — e.g., fines proportional to speed, not binary brackets. While harder to implement, such systems would reduce waste and gaming, producing fairer outcomes across the board.

Key Points
  • Brackets cause clustering at the top or bottom of each threshold, wasting resources and distorting behavior.
  • Historical example: 18th-century England’s flat capital punishment threshold for theft of 12 pence led to murder escalation and jury nullification.
  • Modern parallels include airline baggage fees, hotel late-checkout charges, and tax bracket edge effects.

Why It Matters

This framework exposes hidden inefficiencies in everything from personal finance to AI regulation — continuous rules beat binary cutoffs.