Google DeepMind researchers argue AI development is on pace to move beyond human-level AGI to ASI, which consistently outperforms groups of human experts
First reported by Institute.deepmind ·
The cost to perform cognitive tasks with AI is projected to fall by 1000x within a decade.
Google DeepMind researchers propose that artificial intelligence development is not only on track for Artificial General Intelligence (AGI) but could rapidly advance to Artificial Superintelligence (ASI). ASI, defined as a system that consistently outperforms groups of human experts across virtually all tasks, is seen as the next frontier beyond AGI. The researchers outline four potential pathways from AGI to ASI: scaling models and data, technological paradigm shifts, improving agent coordination, and recursive self-improvement. They emphasize that current trends in compute efficiency, hardware improvements, and AI hardware investment suggest a significant increase in effective compute power and a dramatic decrease in the cost of AI cognitive tasks over the next decade. This could lead to a 'cognitive industrialization' where many cognitive tasks become commoditized, analogous to physical labor during the industrial revolution. The research also highlights that the vast space between human-level AI and theoretical intelligence limits remains largely uncharted, with potential for AI to surpass human capabilities in ways that are not yet fully understood.
The rapid advancement in AI development, driven by compounding factors in hardware, algorithmic efficiency, and investment, suggests a future where cognitive tasks become increasingly commoditized. This 'cognitive industrialization' implies a standardization and automation of intellectual work, potentially reshaping economies and societies on a scale comparable to the industrial revolution. The researchers highlight that while AGI matching human cognitive abilities is approaching, the subsequent leap to ASI, systems far exceeding human expert performance, is a more tangible and nearer-term prospect than previously assumed.
This progression towards ASI suggests a potential paradigm shift in scientific discovery, technological innovation, and problem-solving capabilities. The ability of ASI to outperform human experts across diverse domains could accelerate breakthroughs at an unprecedented rate. However, this also raises critical governance and safety questions as AI capabilities outpace human understanding and control, necessitating proactive research into tracking progress and managing the transition beyond AGI.
AI-written summary. May contain errors.