Back to News

DeepMind Institute essay explores paths to artificial superintelligence beyond AGI

#deepmind#asi#agi#ai-research

Google DeepMind Institute published an essay titled 'What Comes After AGI?' arguing that AGI is not the endpoint of AI development but a stage on an intelligence continuum, with potential paths toward artificial superintelligence (ASI) that consistently outperforms groups of human experts. The essay outlines four potential technical routes to ASI, including recursive self-improvement, and highlights challenges in measuring, predicting, and governing progress in the superintelligence era.

Coverage timeline

  1. Techmeme

    Tim Genewein / DeepMind Institute : Google DeepMind researchers argue AI development is on pace to move beyond human-level AGI to ASI, which consistently outperforms groups of human experts — As multiple technological routes toward artificial general superintelligence take shape, the regime beyond AGI is shifting …

  2. 机器之心机器之心

    最近,Google DeepMind Institude 发布了最新博客文章《 AGI 之后是什么? 》,探讨了人工通用智能(AGI)之后的技术演进方向,以及人类可能通过哪些路径迈向人工超级智能(ASI)。 文章链接:https://institute.deepmind.com/essays/what-comes-after-agi/ 文章指出,AGI 并不是人工智能发展的终点,而可能只是智能连续谱上的一个阶段。随着模型能力、计算效率和多智能体协作水平不断提高,AI 有可能进一步超越人类专家的认知能力,甚至加速自身的技术进步。 DeepMind 从智能能力边界、认知工业化、技术发展路径和递归自我改进等角度展开分析,提出了通往 ASI 的四条潜在技术路线,并强调,超级智能时代的关键挑战不仅是如何实现更强大的 AI,还包括如何衡量、预测和治理这种技术进步。 一、AGI 不是终点,ASI 还有多远? 过去几十年,人工智能研究一直围绕一个重要目标展开:实现能够在人类几乎所有认知任务上达到人类水平的 AGI。 但 DeepMind 认为,AGI 并不是人工智能发展的终点。人类水平的智能只是整个智能连续谱中的一个节点,其上方仍存在巨大的能力空间。 文章将人工超级智能(ASI)描述为一种能够在几乎所有人类关注的任务中,持续超越大规模人类专家团队的 AI 系统。 例如,ASI 可能在极短时间内完成一个科研领域需要多年才能取得的进展,或者以更少的资源,实现大型科技公司一年才能完成的芯片设计与技术研发成果。 实际上,当前 AI 已经在国际象棋、围棋、蛋白质结构预测及部分数学证明任务中展现出超越人类的能力。但这些仍然属于特定领域的突破,距离全面的超级智能尚有明显差距。 DeepMind 强调,虽然人类已经能够从数学和物理学角度研究智能的理论极限,但从人类水平的 AGI 到这些极限之间,究竟存在多大的能力空间,目前仍然是一个尚未充分探索的问题。 除了模型本身的能力突破,DeepMind 还指出,计算能力的持续增长和运行成本的快速下降,也可能成为推动 AI 迈向超级智能的重要因素。 文章由此引出了「认知工业化」(Cognitive Industrialization)的概念:随着 AI 计算效率不断提高,越来越多的认知工作可能像工业时代的体力劳动一样,实现标准化、自动化和规模化生产。 文章指

Related stories

  • DeepSeek open-sources Super-agent architecture for autonomous bug-fixing agents

    DeepSeek has publicly released the architectural blueprint for its Super-agent, a research prototype of a personal agent that maintains cross-session context to handle multi-task coordination. Internal tests show the agent can autonomously identify bugs and submit pull requests without being explicitly prompted, with a median task completion time of 12.4 minutes in simulated multi-task environments. DeepSeek positions this as a step toward reducing human attention bottlenecks in multi-task workflows.

    3 reports · 3 sources
  • Google DeepMind launches Genie 4, adds agent and gallery for AI Ultra subscribers

    Google DeepMind announced that Genie 4, its generative interactive world model, is now available. The rollout to Google AI Ultra subscribers also includes new Genie 3 features, a Project Genie agent for building worlds, and a community gallery for sharing them.

    2 reports · 2 sources
  • Google DeepMind unveils AlphaGenome 2, predicting DNA activity per cell type to study disease

    Google DeepMind has unveiled AlphaGenome 2, a successor to its genome-predicting model AlphaGenome, designed to predict how DNA changes affect different cell types and states. The conditional model incorporates biological conditions such as tissue, cell type, or state as inputs at every network stage, aiming to link genetic variants to disease causes and potential treatments.

    2 reports · 2 sources
  • Google launches Gemini 4 Hydrogen, a low-cost agentic model, across products

    Google announced Gemini 4 Hydrogen, a new Flash-level multimodal model with a 1M token window, priced at $1.50/1M input and $7.50/1M output tokens. It is available across AI Studio, the Gemini API, Search, the Gemini app, and Antigravity, and introduces an xhigh reasoning level. Benchmarks show it nearly ties Claude Sonnet 5.5 and GPT-6.1 Sol on the Artificial Analysis Intelligence Index at lower cost, and it powers an upgraded AI Mode in Search.

    9 reports · 6 sources570 points · 352 comments

News updates every hour. Subscribe in your RSS reader to get new stories automatically.

Subscribe