AI Dictionary › AI Fundamentals

Goodhart's Law

Legge di Goodhart

"When a measure becomes a target, it ceases to be a good measure." The law is named after economist Charles Goodhart, who stated it in 1975 about British monetary policy, and it applies wherever there is a number somebody has an interest in pushing up. Calls closed by a call center, lines of code written, benchmarks passed by a model.

Definition

For artificial intelligence it is the underlying law behind reward hacking and behind the benchmark race: a model trained to pass a test learns the test; a model graded on human preferences learns the preferences. The measure was good as long as nobody optimized it; the moment it becomes the target, whoever chases it finds the shortest road, which rarely runs through the intended result.

The operating rule: keep the measure and the goal distinct, and check the goal directly now and then, by hand, with a method the system does not know. A measure nobody verifies from outside is already a target.

Related terms

• Reward Hacking
• Benchmark
• Model Evaluation
• Drift
• Reward Function

More in AI Fundamentals

• LLM (Large Language Model)
• Token
• Context Window
• Temperature
• AI Hallucination
• Fine-tuning
• RAG (Retrieval Augmented Generation)
• Embedding

Put it into practice

• Real scenario: apply this concept with an instant AI verdict.
• Daily gym: short drills to train dictionary concepts.
• Playoff game: challenge other players, elimination-style.

Say what you mean. Get what you need.

Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.