For years, the race toward Artificial General Intelligence has been measured in vague proclamations and shifting goalposts. That changes now. Google DeepMind has introduced a rigorous cognitive framework that transforms AGI progress from speculation into measurable science—and they're putting [$200,000 on the table](https://blog.google/innovation-and-ai/models-and-research/google-deepmind/measuring-agi-cognitive-framework/) to prove it.

The Problem: We've Been Measuring AGI Wrong

Current AI benchmarks tell us something is wrong. Models ace standardized tests while failing basic reasoning tasks. They write poetry but can't plan a multi-step project. They solve complex math problems yet struggle with simple social interactions. This disconnect reveals a fundamental truth: we've been measuring AI capabilities in silos, not as integrated cognitive systems.

DeepMind's new framework, detailed in their paper ["Measuring Progress Toward AGI: A Cognitive Taxonomy"](https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/measuring-progress-toward-agi/measuring-progress-toward-agi-a-cognitive-framework.pdf), addresses this by drawing from psychology, neuroscience, and cognitive science to deconstruct general intelligence into 10 core cognitive abilities: perception, generation, attention, learning, memory, reasoning, metacognition, executive functions, problem solving, and social cognition.

The 10-Ability Taxonomy: A Technical Breakdown

The framework treats intelligence as a multidimensional "cognitive profile" rather than a single score. Each ability represents a distinct cognitive capacity:

  • Perception: Extracting and processing sensory information from the environment
  • Generation: Producing outputs such as text, speech, and actions
  • Attention: Focusing cognitive resources on relevant information while filtering distractions
  • Learning: Acquiring new knowledge through experience and instruction
  • Memory: Storing and retrieving information across time scales
  • Reasoning: Drawing valid conclusions through logical inference
  • Metacognition: Monitoring and understanding one's own cognitive processes
  • Executive Functions: Planning, inhibition, and cognitive flexibility
  • Problem Solving: Finding effective solutions to domain-specific challenges
  • Social Cognition: Processing social information and responding appropriately

This taxonomy builds on DeepMind's 2023 Levels of AGI framework, but with a crucial difference: it provides a systematic method for human-relative performance mapping across all dimensions.

The Three-Stage Evaluation Protocol

The framework proposes a rigorous evaluation methodology, [according to DeepMind's official blog](https://deepmind.google/blog/measuring-progress-toward-agi-a-cognitive-framework/):

  1. Test AI systems on broad cognitive tasks for each ability using uncontaminated datasets
  2. Gather human baselines from demographically representative adult samples
  3. Map AI performance against human performance distributions for each ability

This approach moves beyond binary "human-level or not" metrics to nuanced, multidimensional profiles showing where AI excels, matches, or falls short of human capabilities.

The $200,000 Kaggle Hackathon

DeepMind isn't just publishing theory—they're crowdsourcing solutions. On [March 17, 2026](https://www.techbuzz.ai/articles/google-deepmind-unveils-cognitive-framework-to-track-agi-progress), they launched a Kaggle hackathon targeting five abilities with the largest evaluation gaps: learning, metacognition, attention, executive functions, and social cognition.

The prize structure is substantial:

  • $10,000 for the top two submissions in each of the five tracks
  • $25,000 grand prizes for the four best overall submissions
  • $200,000 total prize pool

Submissions are open from March 17 to April 16, 2026, with winners announced June 1, 2026. The hackathon uses Kaggle's Community Benchmarks platform to test submissions against top AI models.

Reality Check: What This Framework Actually Tells Us

Let's be clear about what this framework does and doesn't do. It doesn't predict when AGI will arrive. It doesn't define AGI itself. What it provides is a standardized measurement infrastructure for tracking progress.

The choice of the five hackathon tracks reveals where current AI evaluation is weakest. [Industry observers note](https://www.theregister.com/2026/03/18/google_deepmind_agi_hackathon/) that metacognition and social cognition—abilities requiring self-awareness and understanding of others—are notoriously difficult to benchmark. Learning and executive functions present similar challenges: how do you measure a system's ability to acquire new knowledge independently or plan across extended time horizons?

The framework's strength lies in its scientific grounding. By using human baselines from representative populations, it avoids the trap of comparing AI to cherry-picked expert performance or outdated datasets.

Implications for Developers and Researchers

For the AI community, this framework offers several actionable opportunities:

Benchmark Development: The hackathon invites researchers to create novel evaluation methods for undermeasured abilities. This is applied research with immediate impact and significant financial incentive.

Model Evaluation: Once benchmarks exist, developers can map their models' cognitive profiles, identifying specific weaknesses rather than broad performance gaps.

Progress Tracking: The framework enables systematic comparison across model generations, architectures, and training approaches using consistent metrics.

Research Prioritization: The identified gaps—particularly in metacognition, social cognition, and executive functions—highlight where fundamental research is most needed.

Resources

  • [DeepMind's Official Blog Post](https://blog.google/innovation-and-ai/models-and-research/google-deepmind/measuring-agi-cognitive-framework/)
  • [Full Paper: "Measuring Progress Toward AGI: A Cognitive Framework" (PDF)](https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/measuring-progress-toward-agi/measuring-progress-toward-agi-a-cognitive-framework.pdf)
  • [Kaggle Hackathon: Measuring Progress Toward AGI](https://www.kaggle.com/competitions/measuring-progress-toward-agi-cognitive-abilities)
  • [DeepMind Research Page](https://deepmind.google/research/)

DeepMind has given the AI community something we've desperately needed: a shared vocabulary and methodology for discussing AGI progress. Whether you're building benchmarks or building models, this framework deserves your attention.