DeepMind Report: Four Paths to ASI and the Economics of Cognitive Industrialization

·2026.11.20 00:00

Key point

A DeepMind Institute technical report identifies four compounding pathways to Artificial Superintelligence (ASI), highlighting that global effective compute is growing at least 10x annually and cognitive work costs are dropping by 13x per year.

Details

  • DeepMind Institute published the technical report 'From AGI to ASI' (arXiv: 2606.12683), analyzing the transition from Artificial General Intelligence to Artificial Superintelligence.
  • Global effective compute is increasing at a minimum of 10x per year, driven by hardware efficiency, algorithmic improvements, and investment, while the cost of fixed benchmark tasks has fallen by 13x annually since 2023.
  • The report outlines four non-exclusive pathways to ASI: model and data scaling, technological paradigm shifts, agent coordination, and recursive self-improvement (RSI).
  • Restricting a single dimension of AI development, such as training compute, is insufficient to halt progress and may create a 'capability overhang' where suppressed capabilities lead to sudden jumps when released.

Pathways to ASI

  • Model scaling faces bottlenecks from data exhaustion, diminishing returns on model size, and physical resource limits like power and chip manufacturing.
  • Recursive self-improvement (RSI) operates through three mechanisms: genetic (code/design), cultural (data), and social (group specialization), with the fastest progress expected in domains allowing automatic verification like mathematical proofs.
  • Agent coordination involves managing millions of AGI instances to achieve collective intelligence, though current research shows diminishing returns due to coordination inefficiencies.

Policy and Implications

  • The four pathways are mutually reinforcing; blocking one often shifts effort to others due to economic and geopolitical competition.
  • Cognitive industrialization alone, even without dramatic frontier model breakthroughs, is expected to restructure knowledge work and the broader economy within the next decade.
  • Addressing the complexity of ASI pathways requires unprecedented interdisciplinary efforts between frontier labs and national or global institutions for modeling, measurement, and prediction.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.