Incentive Mapping¶

Why It Matters¶

Systems routinely produce outcomes their designers didn't want because the incentive structure rewards something else. Diagnosing this requires mapping incentives as they are, not as hoped. "Why does this keep happening?" is usually answered by "What does this system actually reward?"

The Key Move¶

For any system, identify what actions are rewarded, what's punished, who monitors, and what the monitors' incentives are. Then predict behavior that maximizes reward within those rules. The behavior you get is the behavior you're paying for.

Classic Application¶

Compensation design. Salespeople paid on revenue maximize revenue, not profit. Employees measured on lines of code write verbose code. Teachers evaluated on test scores teach to the test. The system produces exactly what it measures.

Surprising Application¶

Content recommendation. Algorithms optimizing for engagement produce outrage because outrage engages. The system is working exactly as designed, just not as intended. The recommendation engine's incentive is maximizing clicks - if outrage maximizes clicks, outrage is what you get.

Failure Modes¶

• Assuming agents share your objectives - they respond to their incentives, not your wishes
• Ignoring that measurement itself shapes behavior (Goodhart's law) - when a measure becomes a target, it ceases to be a good measure
• Forgetting that monitors need incentives too - who watches the watchers?
• Treating stated incentives as actual incentives - official rewards often differ from real ones
• Ignoring non-monetary incentives - status, autonomy, meaning all drive behavior

Go Deeper¶

Kerr's "On the Folly of Rewarding A While Hoping for B." Principal-agent literature generally.