Levels of AGI for Operationalizing Progress on the Path to AGI

2023/11/04 by Meredith Ringel Morris, Jascha Sohl-Dickstein, Jascha Sohl‐Dickstein +15 · 16 voices · 52 citations
Business, Management and Accounting · Computer Science · Decision Sciences · Psychology · #Artificial intelligence #Autonomy #Big Data and Business Intelligence #Computer science #Data Quality and Management #Epistemology #Explainable Artificial Intelligence (XAI) #Generality #Ontology #Operationalization #Path (computing) #Programming language #Psychology #Software deployment #Software engineering #cs.AI

paper · pdf · doi:10.48550/arxiv.2311.02462

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2023/11/04 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

We propose a framework for classifying the capabilities and behavior of Artificial General Intelligence (AGI) models and their precursors. This framework introduces levels of AGI performance, generality, and autonomy, providing a common language to compare models, assess risks, and measure progress along the path to AGI. To develop our framework, we analyze existing definitions of AGI, and distill six principles that a useful ontology for AGI should satisfy. With these principles in mind, we propose "Levels of AGI" based on depth (performance) and breadth (generality) of capabilities, and reflect on how current systems fit into this ontology. We discuss the challenging requirements for future benchmarks that quantify the behavior and capabilities of AGI models against these levels. Finally, we discuss how these levels of AGI interact with deployment considerations such as autonomy and risk, and emphasize the importance of carefully selecting Human-AI Interaction paradigms for responsible and safe deployment of highly capable AI systems.

Cited by

• Defining AI-Native Systems: Autonomy as Revision Authority
• The atomic structure of work: a micro-action instrument reveals two-pole AI occupational exposure and its decade-scale polar inversion
• AI Must Embrace Specialization via Superhuman Adaptable Intelligence
• Legal Alignment for Safe and Ethical AI
• LLM Social Simulations Are a Promising Research Method
• Stop treating `AGI' as the north-star goal of AI research
• SANet: A Semantic-aware Agentic AI Networking Framework for Cross-layer Optimization in 6G
• The Possibility of Artificial Intelligence Becoming a Subject and the Alignment Problem
• I2-NeRF: Learning Neural Radiance Fields Under Physically-Grounded Media Interactions
• Experience-Driven Exploration for Efficient API-Free AI Agents
• Improving AGI Evaluation: A Data Science Perspective
• Limitations on Accurate, Trusted, Human-level Reasoning
• From Mimicry to True Intelligence (TI) -- A New Paradigm for Artificial General Intelligence
• Comparative Analysis of Large Language Models for the Machine-Assisted Resolution of User Intentions
• AI Compute Architecture and Evolution Trends
• Making AI Inevitable: Historical Perspective and the Problems of Predicting Long-Term Technological Change
• FinMMR: Make Financial Numerical Reasoning More Multimodal, Comprehensive, and Challenging
• Lessons from complex systems science for AI governance
• OS-MAP: How Far Can Computer-Using Agents Go in Breadth and Depth?
• MultiVox: A Benchmark for Evaluating Voice Assistants for Multimodal Interactions
• UrbanMind: Towards Urban General Intelligence via Tool-Enhanced Retrieval-Augmented Generation and Multilevel Optimization
• VoyagerVision: Investigating the Role of Multi-modal Information for Open-ended Learning Systems
• Levels of Autonomy for AI Agents
• Augmenting the Generality and Performance of Large Language Models for Software Engineering
• SANNet: A Semantic-Aware Agentic AI Networking Framework for Multi-Agent Cross-Layer Coordination
• Gauging Growth: AGI Mathematical Metrics for Economic Progress
• A New Theory of Value for Post-AGI Economics
• Grand challenges for human factors and ergonomics
• Towards Machine-Generated Code for the Resolution of User Intentions
• Multimedia and Visual Analytics in the Agentic Era

Discussions

• Levels of AGI: Operationalizing Progress on the Path to AGI [hn, 89 points, 74 comments]
• (Inglês) Google DeepMind publica estudo com níveis para categorizar o potencial de uma I.A. O potencial é medido de acordo com a performance em relação a uma porcentagem da população. [lemmy, 4 points, 0 comments]
• Quick self-test: At what level are you being overpowered by Artificial General Intelligence (AGI)? And here is the full article from researchers at Google DeepMind: arxiv.org/pdf/2311.024... #AI #KI # [bsky, 4 points, 0 comments]
• Levels of AGI: Operationalizing Progress on the Path to AGI [hn, 3 points, 0 comments]
• DeepMind on the Definitions and Levels of AGI [hn, 3 points, 0 comments]
• Levels of AGI: Operationalizing Progress on the Path to AGI [hn, 3 points, 0 comments]
• #Ai #AGI #Ki #KünstlicheIntelligenz #IoT _ arxiv.org/abs/2311.02462 [bsky, 1 points, 0 comments]
• framework for classifying the capabilities and behavior of Artificial General Intelligence (AGI) models and their precursors. arxiv.org/abs/2311.02462 [bsky, 1 points, 0 comments]
• According to Google DeepMind, we are only on the first of five stages to super AI Google DeepMind has proposed a new framework for classifying the capabilities and behavior of artificial general int [bsky, 1 points, 0 comments]
• How should we assess AGI (Artificial General Intelligence) here Meredith R. Morris, et. al. highlights aspects of these models we should be looking with LLM (Large Language Models) leading the way. a [bsky, 1 points, 0 comments]
• Paper defining relevant terms: arxiv.org/abs/2311.02462 Paper outlining a technical case for safety (PDF warning): storage.googleapis.com/deepmind-med... [bsky, 0 points, 1 comments]
• It's not just scaling. We're not at the singularity but you can start to see it from here. And we'd better think hard about it because we're not ready and it will break everything about our social and [bsky, 0 points, 0 comments]
• Meanwhile, OpenAI and Microsoft also have contractual AGI definitions and verification processes. That is a different definition altogether. The definition matters less than the capability. When it wo [bsky, 0 points, 1 comments]
• Paper here: arxiv.org/abs/2311.02462 (Imagen is a Level 3 "Expert" Narrow AI) [bsky, 0 points, 0 comments]
• WE JUST ENTERED THE AGE OF #AGI. This is not another tech cycle. This is the #IndustrialRevolution compressed into years instead of decades. The #steam engine automated muscle. AGI automates #intellig [bsky, 0 points, 0 comments]
• Google DeepMind vypustil framework pro klasifikaci schopnosti AGI. Podle něj by mělo být možné porovnávat současné (a budoucí) systémy a řadit je do úrovní, podle jejich schopností. Podle něj je Chat [bsky, 0 points, 0 comments]

Related

• How Far Are We From AGI: Are LLMs All We Need?
• Toward the Starting Line: A Systems Engineering Approach to Strong AI
• Towards best practices in AGI safety and governance: A survey of expert opinion
• Creating Scalable AGI: the Open General Intelligence Framework
• AGI-Elo: How Far Are We From Mastering A Task?