Urgent.News

What's breaking now, across thousands of outlets.

AI

Google DeepMind researchers argue AI development is on pace to move beyond human-level AGI to ASI, which consistently outperforms groups of human experts (Tim Genewein/DeepMind Institute)

As multiple technological routes toward artificial general superintelligence take shape, the regime beyond AGI is shifting …

In a recent article by DeepMind researchers, AI development is believed to be on the verge of surpassing human-level artificial general intelligence (AGI) and moving towards artificial superintelligence (ASI). This advanced form of AI would consistently outperform groups of human experts on complex tasks, using fewer resources and less time. The researchers argue that understanding the path from AGI to ASI is crucial in order to effectively prepare for the potential impact of this technological shift.

The concept of intelligence is often viewed as existing on a continuum, with AGI representing a single point along the spectrum. ASI, on the other hand, is seen as an advanced system that surpasses large groups of human experts across various tasks, achieving results that would typically take humans years to accomplish. Examples of ASI include a system that can match the scientific progress of an entire research field in a fraction of the time or achieve technological developments similar to those made by large companies within a year.

The researchers highlight the theoretical limits of intelligence as well as the physical, logical, and complexity-theoretic constraints that AI currently faces. However, they emphasize that these upper limits do not necessarily provide a clear roadmap for predicting the emergence of ASI. The gap between human-level AI and the theoretical limits of intelligence is vast, encompassing a range of expert AGI systems and more advanced superintelligent systems.

To illustrate the potential scale of future AI, the researchers present a scenario where the efficiency of AI hardware and software continues to improve. Over the past decade, AI progress has been driven by a compounding growth in effective compute, resulting in a 10x increase in global effective compute stock over just ten years.

If this trend continues, the researchers predict that in a decade, the number of AI agents that could run today would be equal to the number of people on Earth. Furthermore, continued improvements in hardware and software efficiency would reduce the compute required to run AI workloads by approximately 4x per year, leading to a substantial reduction in the cost of cognitive tasks performed by AI.

Written by urgent.news from Techmeme's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at institute.deepmind.com →

More in AI

Emerging AI Agent Protocols & Agent-to-Agent Handoffs: The 2026 Enterprise RevOps & Engineering Playbook

Enterprise AI architecture has moved beyond single-model chatbots and isolated RAG pipelines. Many teams now run networks of agents that query infrastructure, call tools, and execute multi-step…

  • Enterprise AI architecture now uses networks of agents for multi-step workflows.
  • Open standards MCP (Anthropic) and A2A (Google) address inter-agent communication risks.
  • Human escalation required for autonomous actions crossing approval limits.

More from Friday 20 November →