{
  "id": 20848475,
  "title": "What Comes After AGI?",
  "url": "https://urgent.news/2026/11/20/what-comes-after-agi",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-11-20T02:44:13.000Z",
  "source": {
    "name": "Hacker News",
    "slug": "hacker-news",
    "url": "https://institute.deepmind.com/essays/what-comes-after-agi/"
  },
  "original_language": "en",
  "account": "Since decades, AI development has been centered on achieving artificial general intelligence (AGI) - machines exhibiting human-like cognitive abilities. While AGI's realization remains uncertain, the imminent possibility has led to consideration of its potential successors. One such hypothetical advancement is artificial superintelligence (ASI), defined as a system outperforming large groups of human experts on complex tasks within shorter timeframes and with less resource consumption.\n\nTo envision the transition from AGI to ASI, it is helpful to view general intelligence as a continuum. AGI represents a single point, while ASI stands further ahead, able to replicate and surpass the collective output of human experts. Theoretical physics and complexity theory offer insights into the upper boundaries of general intelligence, but the precise path between current AI capabilities and these theoretical limits remains unclear.\n\nOne potential scenario involves an exponential growth in AI efficiency. Over the past decade, advancements in AI have been driven by increasing effective compute, combining three factors that have led to a 10x growth in global effective compute stock every year. Assuming this trend persists, the number of AI agents comparable to today's systems could surpass the global human population within a decade. Moreover, reducing compute requirements for AI tasks by a factor of 4 annually could further amplify this effect, potentially leading to cognitive industrialization akin to the mechanization of physical labor.\n\nWhile this projected growth in AI capabilities raises intriguing possibilities, it is not guaranteed. Practical limitations on compute expansion and algorithmic efficiency could impose constraints. Even if cognitive industrialization unfolds as envisioned, it may not be sufficient to achieve ASI. However, the sheer scale and speed of billions of high-efficiency AI agents could still bridge the gap between AGI and ASI, offering a transformative impact on science, the economy, and society.",
  "summary": null,
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}