Where Does ChatGPT Fall on the 5 Levels of AGI? — Google DeepMind's New AI Rating Scale
Google DeepMind has released a cognitive framework for measuring progress toward AGI and launched a Kaggle hackathon. ChatGPT and Gemini are classified as Level 1, 'Emerging,' out of five levels.
"How far are we from AI being as smart as a human?" Google DeepMind just gave a concrete answer. They broke down the path to AGI (Artificial General Intelligence — AI capable of performing most tasks a human can) into a 5-level rating scale, pinpointing exactly where today's AI systems like ChatGPT, Gemini, and Claude currently stand.
This cognitive framework, announced on March 17, goes far beyond a research paper. Google DeepMind has launched a Kaggle hackathon (a global competition on the world's largest data science platform) and declared its intent to build benchmarks (standardized tests for measuring AI capability) together with researchers worldwide.
The 5 Levels of AGI — Where Does AI Stand Today?
DeepMind's framework evaluates AI along two axes: performance (how well it does things) and generality (how many different things it can do). Think of it like the Level 0–5 scale for self-driving cars — AI now has its own clear grading system.
| Level | Criteria | Narrow AI Examples | General AGI |
|---|---|---|---|
| Level 0 — No AI | Calculators, compilers | Calculator | Amazon Mechanical Turk, etc. |
| Level 1 — Emerging 🟡 | Equal to or slightly better than an unskilled human | Simple rule-based systems | ChatGPT, Gemini, Claude ← Here |
| Level 2 — Competent | Top 50% of skilled adults | Siri, Alexa, spell checkers | None yet |
| Level 3 — Expert | Top 10% of skilled adults | Grammarly, DALL-E 2 | None yet |
| Level 4 — Virtuoso | Top 1% of skilled adults | Deep Blue, AlphaGo | None yet |
| Level 5 — Superhuman | Surpasses 100% of humans | AlphaFold, AlphaZero | None yet (= ASI) |
Why ChatGPT Is Still at Level 1
"ChatGPT is that smart and it's only Level 1?" It might seem surprising. But DeepMind's criteria are clear.
Today's leading AI systems (ChatGPT, Gemini, Claude, etc.) do show 'Competent (Level 2)' performance in specific tasks like short-form writing and simple coding. But in areas like math problems, fact-checking, and complex reasoning, they're still at the 'Emerging (Level 1)' stage.
DeepMind explains it this way: "Even if an AI can write code, if that code isn't reliably accurate, it can't be considered truly general-purpose."
In other words, an AI has to perform well across all domains to advance to the next level. Even if it excels in one area, being a beginner in another keeps it classified as 'Emerging General AI.'
6 Principles — A New Standard for Defining AGI
DeepMind's research team analyzed various existing AGI definitions (the Turing Test, the Coffee Test, OpenAI's economic value criteria, etc.) and distilled 6 principles for defining AGI.
1. Focus on capabilities, not processes — What matters is whether AI can 'accomplish' things the way a human can, not whether it 'thinks' like a human. It doesn't need to have consciousness or emotions.
2. Evaluate both generality and performance — We need to distinguish between AI that only does one thing well (like a chess AI) and AI that handles a variety of tasks.
3. Focus on cognitive abilities — AGI can exist without a robotic body. The core is the 'brain' — the ability to think and make judgments.
4. Potential is enough — Even if an AI hasn't been deployed in the real world, having the capability qualifies it for a given level.
5. Measure with real-world, meaningful tasks — AI should be evaluated on tasks people actually find valuable (economic, social, and artistic value), not just test scores.
6. AGI is a journey, not a destination — Like self-driving cars, AI advances in stages, with each level bringing new risks and opportunities.
Kaggle Hackathon — Building the AGI Report Card Together
DeepMind didn't stop at publishing a paper. They announced a hackathon on Kaggle (the world's largest data science competition platform) to collaborate with researchers and developers globally on creating new benchmarks (standardized evaluation tools) for measuring AGI.
While previous AI benchmarks only tested whether AI could solve specific exam-style questions, this hackathon aims to design evaluation methods based on the full range of human cognitive abilities.
What This Rating Scale Means for You
If you use AI tools at work — Current AI is at the 'Emerging General' stage. It handles writing and simple analysis well, but tasks requiring complex judgment still need human review. When AI reaches Level 2 ('Competent'), this dynamic will change significantly.
If you're an AI developer — This framework is a tool for objectively assessing where your AI stands. You can review the full classification in the original paper (free access).
If you're an AI investor or policymaker — This framework helps you understand the gap between claims that "AGI is just around the corner" and the reality that we're "still at Level 1." DeepMind emphasizes that "each level has its own unique risks and deployment considerations."
Level 5 'Superhuman General AI' (referred to as ASI, or Artificial Superintelligence, in the paper) is a frontier no AI has reached yet. But the leap from Level 1 to Level 2 is something we're likely to witness in our lifetimes.
Related Content — Get Started with AI the Easy Way | Free Learning Guide | More AI News
Sources
Stay updated on AI news
Simple explanations of the latest AI developments