Measuring Goodhart’s law
OpenAI just revealed how it tackles Goodhart's law — the metric that breaks when you optimize it too hard.
Why it matters
As AI companies scale optimization objectives, managing the tension between measurable KPIs and actual model quality becomes a core strategic problem. OpenAI's public grappling with this signals how fundamental metric gaming is to building reliable AI systems.
The key facts
8 to knowOpenAI addresses Goodhart's law in model optimization
Challenge: optimizing objectives that are difficult or costly to measure
Relevance to alignment and objective specification in AI systems
Published April 13, 2022 — foundational perspective piece
OpenAI addresses Goodhart's law application to AI objective optimization
Focus on measuring objectives that are difficult or costly to measure
Published April 13, 2022 — historical strategic commentary on AI safety/alignment
Original principle: 'When a measure becomes a target, it ceases to be a good measure'
Go to the source
OpenAI Blogopenai.com
Publisher excerpt: Goodhart’s law famously says: “When a measure becomes a target, it ceases to be a good measure.” Although originally from economics, it’s something we have to grapple with at OpenAI when figuring out how to optimize objectives that are difficult or costly to measure.