Research map / Large language models / AI and society
Position: Levels of AGI for Operationalizing Progress on the Path to AGI
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a leveled, two-dimensional ontology of AGI (performance depth × generality breadth) plus six autonomy levels for human-AI interaction, with principles for defining and benchmarking AGI.
Problem
Existing definitions of AGI are vague and not operationalizable, which hinders comparing models, assessing risks, setting policy, and measuring progress toward AGI.
Approach
Derives six principles from nine prior AGI definitions and proposes a matrixed taxonomy crossing performance levels (Emerging to Superhuman) with narrow vs. general scope, and separately a six-level autonomy scale that is unlocked by, but not determined by, AGI capability level.
Key results
- Proposes a matrix of five performance levels (Emerging, Competent, Expert, Exceptional, Superhuman) crossed with narrow and general breadth, and estimates frontier LLMs as Level 1 (Emerging AGI) as of September 2023.
- Defines six Levels of Autonomy (No AI to AI as an Agent) and argues that autonomy is unlocked by, but not determined by, AGI level.
- Explicitly does not propose a benchmark; instead outlines properties a living, ecologically valid AGI benchmark should have.
Surprising findings
- Argues that current frontier LLMs, often called AGI, should be classed as Emerging AGI rather than general-purpose AGI because they are not reliably performant across most tasks.
- Suggests that a system capable of Expert-level performance may in practice be only Competent because prompting interfaces are too complex for end-users (DALL-E 2 example).
Limitations
- No benchmark is proposed; operational task sets and generality thresholds remain open questions.
- Example system level assignments are approximate and based on the authors' judgment.
Experimental setup
Builds on: Levels of Driving Automation
Closest papers by meaning
- AGI-Elo: How Far Are We From Mastering A Task? (NeurIPS 2025 · similarity 0.88)
- Position: Stop Making Unscientific AGI Performance Claims (ICML 2024 · similarity 0.85)
- Position: Stop treating `AGI' as the north-star goal of AI research (ICML 2025 · similarity 0.84)
- The AI Race: Why Current Neural Network-based Architectures are a Poor Basis for Artificial General Intelligence (AAAI 2025 · similarity 0.84)
- The Alignment Problem from a Deep Learning Perspective (ICLR 2024 · similarity 0.84)
- Position: AI Scaling: From Up to Down and Out (ICML 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.