How close are we to AGI? That question usually gets vague answers — somewhere between two and twenty years, depending on who you ask. Google DeepMind is now trying to answer it scientifically. And they are asking the community for help.
A Cognitive Map for AI
The DeepMind team developed what they call a cognitive taxonomy — a framework that divides AI capabilities into ten areas and compares them against human abilities. The idea is straightforward: run AI models and humans through the same benchmarks and see where AI keeps up.
The ten areas break down into eight building blocks — perception, generation, attention, learning, memory, metacognition, executive functions, and one additional area — plus two composite faculties: problem solving and social cognition.
Where the Biggest Gaps Are
Here is where it gets honest: for five of those ten areas, there are not even good tests yet. Learning, metacognition, attention, executive functions, and social cognition — these are the capabilities where research knows the least about how well current AI models actually perform.
That is exactly where the hackathon comes in.
$200,000 for Better Benchmarks
DeepMind launched a Kaggle competition where developers and researchers design evaluations for those five gaps. The prize pool is $200,000. Two winners per area at $10,000 each, plus four overall winners at $25,000 each. Results will be announced in June.
Why This Matters
Benchmarks sound boring — but they are not. How we measure AI determines how we build AI. If we only test language skills, we optimize models for language. If we test social cognition or metacognition, entirely different incentives emerge.
DeepMind’s framework is also an implicit critique of existing benchmarks. Most current tests primarily measure knowledge and language comprehension — not whether a system can actually learn, assess its own performance, or respond appropriately in social contexts.
It is a smart move: instead of claiming how close AGI is, DeepMind is building the tool that lets us find out. And they are bringing the smartest minds in the community along for the ride.
Sources: