Google DeepMind Proposes Cognitive Framework for Measuring Progress Toward AGI

2026-03-18T13:13:31.000Z·2 min read
Google DeepMind releases a paper proposing 10 cognitive abilities to measure AGI progress, alongside a $200K Kaggle hackathon for the community to build evaluations against frontier models.

Google DeepMind has released a new paper, "Measuring Progress Toward AGI: A Cognitive Taxonomy," that presents a scientific foundation for understanding the cognitive capabilities of AI systems. The framework draws on decades of research from psychology, neuroscience, and cognitive science.

The 10 Cognitive Abilities

The framework identifies 10 key cognitive abilities hypothesized to be important for general intelligence in AI:

#AbilityDescription
1PerceptionExtracting and processing sensory information from the environment
2Motor controlProducing outputs such as text, speech, and actions
3AttentionFocusing cognitive resources on what matters
4LearningAcquiring new knowledge through experience and instruction
5MemoryStoring and retrieving information over time
6ReasoningDrawing valid conclusions through logical inference
7MetacognitionKnowledge and monitoring of one's own cognitive processes
8Executive functionsPlanning, inhibition, and cognitive flexibility
9Problem solvingFinding effective solutions to domain-specific problems
10Social cognitionProcessing and interpreting social information

Three-Stage Evaluation Protocol

  1. Evaluate AI systems across a broad suite of cognitive tasks covering each ability, using held-out test sets to prevent data contamination
  2. Collect human baselines for the same tasks from a demographically representative sample of adults
  3. Map each AI system's performance relative to the distribution of human performance in each ability

$200K Kaggle Hackathon

To put the framework into practice, DeepMind is launching a Kaggle hackathon focused on five cognitive abilities with the largest evaluation gaps: learning, metacognition, attention, executive functions, and social cognition.

Why It Matters

Most AGI benchmarks focus on task-level performance (MMLU, HumanEval, etc.). DeepMind's cognitive taxonomy approach shifts the evaluation lens to underlying capabilities — asking not just "can the model do X?" but "does the model possess the cognitive ability required to do X, Y, and Z across domains?"

Source: Google Blog | HN Discussion

↗ Original source
← Previous: OnPrem.LLM: Launch Autonomous AI Agents with Sandboxed Execution in 2 Lines of CodeNext: Ryugu Asteroid Samples Contain All DNA and RNA Building Blocks →
Comments0