Krux
March 20, 2026
DeepMind Proposes 10-Faculty Test to Measure AGI Progress
Published: March 20, 2026 at 12:58 AM
Updated: March 20, 2026 at 12:58 AM
100-word summary
DeepMind published a framework to measure how close AI systems are to general intelligence, replacing single-task benchmarks with tests across 10 cognitive faculties, from memory to social cognition. Models get graded against demographically representative adults, not PhD researchers or kindergarteners. The twist: DeepMind is crowdsourcing the actual tests through a Kaggle hackathon with a $200k prize pool, asking participants to design evaluations for five faculties. Winners get announced June 1. The bet is that cognitive profiles showing where models excel and struggle will matter more than one headline score. The framework itself hasn't tested any models yet, but it signals the industry wants to know if AI is truly general...
What happened
DeepMind published a framework to measure how close AI systems are to general intelligence, replacing single-task benchmarks with tests across 10 cognitive faculties, from memory to social cognition. Models get graded against demographically representative adults, not PhD researchers or kindergarteners. The twist: DeepMind is crowdsourcing the actual tests through a Kaggle hackathon with a $200k prize pool, asking participants to design evaluations for five faculties. Winners get announced June 1. The bet is that cognitive profiles showing where models excel and struggle will matter more than one headline score.
Why it matters
The framework itself hasn't tested any models yet, but it signals the industry wants to know if AI is truly general or just really good at pattern matching.