A reward changes not only how much of the desired behaviour you get, but also which new strategies become rational.
The famous cobra anecdote is weakly documented historically, but the mechanism of perverse incentives is strong. Once an intervention changes payoffs, strategic actors respond to what is actually rewarded.
A reward changes not only how much of the desired behaviour you get, but also which new strategies become rational.
The popular story says that a government introduced a bounty for each killed cobra, prompting people to breed cobras. The historical reliability of that specific anecdote is uncertain; it nevertheless illustrates a robust incentive-design problem.
Perverse incentives, principal–agent problems and gaming are supported by research. The cobra story itself must not be used as empirical evidence for the mechanism.
An intervention changes payoffs; actors adapt to rewarded observables; divergence between reward and objective may therefore increase.
AI transfer to investigate: AI may search for loopholes, exploits and reward-maximising strategies more quickly. This is amplification, not a new mechanism.
Proposed test, not yet conducted: Compare incomplete incentives with and without AI assistance and with hidden or rotating independent audits. Measure rewarded score against objective quality.
The cobra story illustrates the problem. Incentive theory and the work of Goodhart and Campbell provide starting points for research.
Use the cobra story as an illustration, not as historical evidence. Investigate whether a reward rule causes behaviour that improves the proxy without improving the independent objective.
Campbell’s 1976 working paper (pp. 49–54) supports the incentive argument, not the historical cobra anecdote. The 1979 journal version is a separate publication whose full text has not been assessed here. The AI application has not been tested.
Help improve this case
We review your contribution before updating the text or sources.