According to ChainCatcher, Google DeepMind research scientist Tim Genewein published an article at DeepMind Institute outlining four potential paths from artificial general intelligence (AGI) to artificial superintelligence (ASI): scaling models and data, a shift in technical paradigm, improved coordination among agents, and recursive self-improvement. The article describes ASI as systems that consistently outperform large groups of human experts across a broad range of tasks relevant to people.
It also examines how hardware investment, algorithmic efficiency, and computing-power growth affect AI system scale and the cost of cognitive tasks. Potential constraints on scaling include limited high-quality data, diminishing returns, and shortages of power and chips. Genewein says assessing AI progress will require new benchmarks and modeling methods, with participation from frontier labs and institutions at all levels. The article is based on the technical report “From AGI to ASI,” co-authored by Genewein and others. DeepMind Institute said the article should not be considered Google’s official view.