NoticeThis site demonstrates one possible use of this domain. For acquisition, partnership, or investment inquiries, please use our contact form.

AGI Fundamentals

The Five Levels of Machine Intelligence

Google DeepMind's 2024 Levels of AGI framework classifies systems from narrow specialists to artificial superintelligence. Here is what each level means and where today's systems sit.

fig / five levels of machine intelligence// field plate
Diagram comparing human and machine capability
Plate / A capability ladder from narrow tools to superintelligence.

Executive summary

The Morris et al. (2024) Levels of AGI framework defines six performance tiers (No AI, Emerging, Competent, Expert, Virtuoso, Superhuman) across two capability axes (narrow vs general). It gives the field a shared vocabulary for comparing systems and tracking progress without overstating it.

Key concepts

  • Performance tiers
  • Narrow vs general axis
  • Emerging AGI
  • Competent AGI
  • Superhuman AI

The six tiers

Morris et al. define six tiers: No AI, Emerging (equal to or somewhat better than an unskilled human), Competent (50th percentile of skilled adults), Expert (90th percentile), Virtuoso (99th percentile), and Superhuman (outperforming all humans).

Two axes

Each tier applies separately to narrow systems (a single task family) and general systems (most cognitive tasks). A chess engine like Stockfish is Superhuman Narrow. AlphaFold is Superhuman Narrow at protein structure. ChatGPT and Gemini are Emerging General. Competent or Expert General AGI does not yet exist.

Why this framework matters

Most public debate confuses narrow superhumanity (a chess engine) with general superhumanity (superintelligence). The framework keeps the axes separate. It also forces precision: claims about AGI should specify which tier on which axis.

Key takeaways

  • 01There are six tiers, applied separately to narrow and general capability.
  • 02Today's frontier chatbots are Emerging General.
  • 03Superhuman performance on a single task does not imply general superintelligence.
  • 04The framework is the closest thing the field has to a shared yardstick.

Frequently asked questions

Is AlphaFold AGI?

No. AlphaFold is Superhuman Narrow at one task (protein structure prediction). It cannot do unrelated cognitive work.

What level is GPT-4 or Claude 3.5?

Both sit at Emerging General — broadly competent across many tasks but not yet matching the 50th-percentile skilled adult reliably across all of them.

When will we reach Competent AGI?

Unknown. Forecasts vary from later this decade to several decades out.