Algorithmic Principal Agent, Automated Evaluation
The AI Performance Review
The AI Performance Review is an Algorithmic Principal Agent and Automated Evaluation scenario. The core lesson: When algorithms evaluate you on metrics you didn't know existed, optimizing for performance becomes guesswork. Your annual review is now algorithmically pre-scored before your manager sees it. Communication tone, response times, meeting participation patterns, and cross-team collaboration metrics are scored and summarized. DecisionPlay maps the players, payoffs, and equilibrium dynamics that shape how this situation typically resolves.
The situation
Your annual review is now algorithmically pre-scored before your manager sees it. Communication tone, response times, meeting participation patterns, and cross-team collaboration metrics are scored and summarized. Your score dropped 14 points from last year. You don't know which metric changed, and the scoring methodology is proprietary.
Background
Automated performance evaluation creates a novel principal-agent problem: the agent (you) is being evaluated by a system whose criteria are opaque and may shift without notice. Unlike a human manager whose preferences you can observe over time, an algorithm's weighting is invisible, possibly updated without announcement, and not subject to the normal social accountability that makes feedback useful. The rational response to opaque metrics is gaming observable proxies, which produces the metrics the algorithm wants rather than the work the organization needs.
What this reveals
Evaluation without explanation
Performance reviews have always been imperfect. Algorithmic performance reviews add a new kind of imperfection: they're systematically opaque in a way that human evaluation isn't. A human manager who rates you poorly can be asked why. An algorithm whose weighting shifted between cycles can't explain itself, and the organization may not know what changed either. The result is evaluation without accountability, which breaks the feedback loop that makes performance management useful in the first place.
How to counter it: Ask for the criteria before the evaluation, not after. The question 'how will my performance be measured this year?' should have a specific answer before you're scored on it. If it doesn't, if the criteria are proprietary or undefined, that's a significant professional risk worth surfacing early. You can't improve at something you can't see, and you can't defend yourself against judgments that weren't disclosed before they were made.
A question to sit with
How do you defend your performance against a judgment that no human consciously made, and that the organization may not fully understand itself?
Frequently asked questions
- What game theory concept does The AI Performance Review illustrate?
- The AI Performance Review illustrates Algorithmic Principal Agent, Automated Evaluation. When algorithms evaluate you on metrics you didn't know existed, optimizing for performance becomes guesswork.
- What is the situation in The AI Performance Review?
- Your annual review is now algorithmically pre-scored before your manager sees it. Communication tone, response times, meeting participation patterns, and cross-team collaboration metrics are scored and summarized. Your score dropped 14 points from last year.
- What does The AI Performance Review reveal about how people decide?
- Evaluation without explanation. Performance reviews have always been imperfect. Algorithmic performance reviews add a new kind of imperfection: they're systematically opaque in a way that human evaluation isn't. A human manager who rates you poorly can be asked why. An algorithm whose weighting shifted between cycles can't explain itself, and the organization may not know what changed either. The result is evaluation without accountability, which breaks the feedback loop that makes performance management useful in the first place.
- How do you avoid the trap in The AI Performance Review?
- Ask for the criteria before the evaluation, not after. The question 'how will my performance be measured this year?' should have a specific answer before you're scored on it. If it doesn't, if the criteria are proprietary or undefined, that's a significant professional risk worth surfacing early. You can't improve at something you can't see, and you can't defend yourself against judgments that weren't disclosed before they were made.
- What is the research behind The AI Performance Review?
- Automated performance evaluation creates a novel principal-agent problem: the agent (you) is being evaluated by a system whose criteria are opaque and may shift without notice. Unlike a human manager whose preferences you can observe over time, an algorithm's weighting is invisible, possibly updated without announcement, and not subject to the normal social accountability that makes feedback useful. The rational response to opaque metrics is gaming observable proxies, which produces the metrics the algorithm wants rather than the work the organization needs.
- How long does The AI Performance Review take to play?
- About 8 min, at core difficulty, across 4 decision points. It runs in your browser with no account and no sign-in.
Keep exploring
More Future Stakes scenarios, or browse all scenarios. New to this? Start with how DecisionPlay works or the game theory glossary.
Topics: futures, AI, workplace, algorithmic-evaluation, employment, season-3