The Question

In 1902, French colonial officials in Hanoi offered a bounty for rat tails to combat a plague-driven infestation. Residents duly delivered thousands of tails. Rat numbers did not decline. Tailless rats were soon spotted scurrying through the city, kept alive to breed more tails. Entrepreneurs began farming rats specifically for the bounty. The measure had become the target, and in doing so had ceased to measure what it was meant to measure. John and colleagues (2023) argue that this pattern, which they call proxy failure, is not an amusing historical curiosity. It is a fundamental and universal risk in any goal-oriented system.

The Argument

The paper proposes a unifying mechanism underlying phenomena that have been named and described separately across economics, ecology, academia, machine learning, and neuroscience. These include Goodhart’s Law, the observation that any measure used as a target ceases to be a good measure; Campbell’s Law, which holds that the more a quantitative indicator is used for social decision-making, the more it will be distorted; and the Cobra Effect, named for a colonial Indian bounty on cobras that incentivised cobra breeding (John et al., 2023).

The authors argue that all of these reflect the same underlying dynamic. Whenever incentivisation or selection is based on an imperfect proxy measure of an underlying goal, pressure arises that tends to make the proxy a worse approximation of the goal. The proxy is imperfect by assumption: if it were a perfect measure of the goal, gaming it would require achieving the goal. It is precisely the gap between proxy and goal that creates the opportunity for proxy failure (John et al., 2023).

Where This Appears

The paper develops the argument across three domains. In neuroscience, dopamine has been studied extensively as a proxy for reward and learning, but the relationship between dopamine signals and actual motivational states is more complex than simple reward-prediction models suggest, and interventions targeting dopamine pathways may not reliably produce the motivational outcomes intended. In economics, performance metrics in organisations, financial targets, and standardised testing in education all show the same pattern: once the metric becomes the goal, behaviour optimises for the metric at the expense of the underlying objective. In ecology, evolutionary pressures on proxy traits such as peacock tails, which signal fitness without directly constituting it, illustrate how proxy failure can operate across biological timescales (John et al., 2023).

Why This Connects to Preregistration and GCA Debates

Regular readers of this series will recognise the dynamic immediately. The preregistration debate reviewed in an earlier post, where Klonsky (2024) argued that preregistration badges were becoming the latest target to be gamed rather than a genuine safeguard, is a direct instance of proxy failure in science. The long-running debate about whether validity coefficients for cognitive ability tests accurately measure what they claim to measure involves the same underlying issue: the gap between the proxy and the actual goal of prediction (John et al., 2023).

Why It Matters

John and colleagues (2023) are not arguing that measurement and incentives are bad. They are arguing that proxy failure is an inherent structural risk in any system that uses imperfect measures to pursue goals, which is to say, every goal-oriented system that has ever existed. Recognising this risk is the first step toward designing systems that are less vulnerable to it: by holding proxies lightly, maintaining direct contact with the underlying goal, and remaining alert to the ways that optimising for the measure can diverge from achieving the thing the measure was meant to represent.

Excerpt

When a measure becomes a target, it ceases to be a good measure. A 2023 paper argues this is not just a management aphorism but a fundamental and universal failure mode of all goal-oriented systems, from rat bounties to performance reviews to scientific publishing.

Reference

John, Y. J., Caldwell, L., McCoy, D. E., & Braganza, O. (2023). Dead rats, dopamine, performance metrics, and peacock tails: Proxy failure is an inherent risk in goal-oriented systems. Behavioral and Brain Sciences. Advance online publication. https://doi.org/10.1017/S0140525X23002753