HackerPapers

arXiv:2311.02462cs.AI

Levels of AGI for Operationalizing Progress on the Path to AGI

Meredith Ringel Morris, Jascha Sohl-Dickstein, Noah Fiedel, Tris Warkentin, Allan Dafoe, Aleksandra Faust, Clement Farabet, Shane Legg

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.

arXiv comments
version 5 - We updated the nomenclature of Level 4 to be "Exceptional" instead of "Virtuoso"; due to ICML 2024 position paper titling requirements, the title is now "Levels of AGI for Operationalizing Progress on the Path to AGI" rather than "Levels of AGI: Operationalizing Progress on the Path to AGI"
Journal reference
Proceedings of ICML 2024
DOI
Subjects
cs.AIArtificial Intelligence

On Hacker News

Linked by 14 identifiable commenters in 20 comments across 18 threads, from Nov 2023 to Sep 2026. Found in 4 HN submissions from Nov 2023 to Nov 2023. Open submission discussions.

Comments linking this paper, per year.

Discussed in comments18 threads

  1. Dynamic Abliteration: Non-Destructive Refusal Suppression via Engram Steering 1 comment

    • I think a baseline regime similar to fiduciary duty is a good starting point towards not killing everyone, and in terms of Overton Window, seems very much doable now. Of course, after we stop agents from committing felony hacking crimes. If you believe that capabilities will…

      theptipView on HNParent context

  2. Andrej Karpathy – It will take a decade to work through the issues with agents 1 comment

    • Fully autonomous agents are marketing fluff right now, but there is like $10T of TAM from promoting most knowledge workers to a manager and automating the boring 80% of their work, and this doesn’t require this full autonomy. Karpathy’s definition of “agent” here is really AGI…

      theptipView on HNParent context

  3. Seven replies to the viral Apple reasoning paper and why they fall short 1 comment

    • OpenAI: https://openai.com/index/how-should-ai-systems-behave/#citat... "By AGI, we mean highly autonomous systems that outperform humans at most economically valuable work." AWS: https://aws.amazon.com/what-is/artificial-general-intelligen... "Artificial general intelligence…

      simonwView on HNParent context

  4. How Does Claude 4 Think? – Sholto Douglas and Trenton Bricken 1 comment

    • The ability to extrapolate from facts. I don't think self awareness, emotions, or senses are required. You should give your definition too, so we can debate where computers fall short. See also: Levels of AGI for Operationalizing Progress on the Path to AGI…

      esafakView on HNParent context

  5. John Carmack talk at Upper Bound 2025 2 comments

    • No, it’s never going to be precise but it’s important to have a good rough definition. Can we just use Morris et al and move on with our lives? Position: Levels of AGI for Operationalizing Progress on the Path to AGI: https://arxiv.org/html/2311.02462v4 There are generational…

      theptipView on HNParent context

    • The Wikipedia article on AGI explains it well enough. Researchers at Google have proposed a classification scheme with multiple levels of AGI. There are different opinions in the research community. https://arxiv.org/abs/2311.02462

      johnb231View on HNParent context

  6. Evolving OpenAI's Structure 2 comments

    • Their multimodal models are a rudimentary form of AGI. EDIT: There can be levels of AGI. Google DeepMind have proposed a framework that would classify ChatGPT as "Emerging AGI". https://arxiv.org/abs/2311.02462

      voidsparkView on HNParent context

    • > AGI would mean something which doesn't need direction or guidance to do anything There can be levels of AGI. Google DeepMind have proposed a framework that would classify ChatGPT as "Emerging AGI". ChatGPT can solve problems that it was not explicitly trained to solve, across…

      voidsparkView on HNParent context

  7. Show HN: Business AI Agents Targeting Level 5 AGI 1 comment
  8. Getting 50% (SoTA) on Arc-AGI with GPT-4o 1 comment

    • This seems to be so broad a definition as to no longer mean anything useful. People in general are interested in capabilities or economic impact, and GPT-2 cleared no notable thresholds in those regards. I prefer the exact opposite approach: let’s use a strict definition, and…

      theptipView on HNParent context

  9. Transformers Can Do Arithmetic with the Right Embeddings 1 comment

    • I'm curious about the framing of research like this.. "The poor performance of transformers on arithmetic tasks" (relative to what?) and how that informs the adjacent conversation on progress towards AGI. Some say AGI has already been achieved, others that it's years or decades…

      pmayrgundterView on HNParent context

  10. What is AGI? 1 comment
  11. Elon Musk sues Sam Altman, Greg Brockman, and OpenAI [pdf] 1 comment

    • I think the DeepMind taxonomy is good; recognize that AGI is too broad a term, so add levels to make it specific. https://arxiv.org/abs/2311.02462 “Competent AGI” (or a little earlier) would be my guess for where OpenAI would not hand it over to MS. Honestly if they displaced…

      theptipView on HNParent context

  12. Why are you so sure I'll lose? 1 comment

    • What specific demonstrations of capability by AI would convince you that your proposed timeline for reaching AGI isn't correct, and that it will happen significantly sooner? If it helps to think about it, there's a good DeepMind paper on a relatively concrete scaling system on…

      lucubratoryView on HNParent context

  13. Audiobox: Meta's new foundation research model for audio generation 1 comment

    • What you're looking for is embodiment, and actually this was explicitly left aside in a recent paper that attempts to give measurement criteria for AGI[0]. But I agree with you that the entire lived sensation is critical to approaching any objective involving alignment. [0]…

      kevindammView on HNParent context

  14. OpenAI researchers warned board of AI breakthrough ahead of CEO ouster 1 comment
  15. Emmett Shear becomes interim OpenAI CEO as Altman talks break down 1 comment

    • > Did he really fire Sam over "AI safety" concerns? How is that remotely rational. Not rational iff (and unlike Sustkever, Hinton, Bengio) you are not a "doomer" / "decel". Ilya's very vocal and on record that he suspects there may be "something else" going on with these models.…

      ignoramousView on HNParent context

  16. Ilya Sutskever "at the center" of Altman firing? 1 comment
  17. Sutskever: OpenAI board doing its mission to build AGI that benefits all 1 comment

    • Paper from two weeks ago by Google DeepMind putting together a very good definition that I haven't seen anyone in the field disagree with yet: https://arxiv.org/abs/2311.02462

      lucubratoryView on HNParent context

  18. Levels of AGI: Operationalizing Progress on the Path to AGI 1 comment

Submitted to Hacker News

Connected through shared HN threads, not inferred similarity or a recommendation.