Google DeepMind researchers map four routes from human-level AGI to superintelligence, or ASI
The essay says effective AI compute has grown at least 10x per year, and that scale alone might bridge AGI to weak ASI, though it calls this far from certain.
What happened
Google DeepMind researchers wrote an essay on what could come after AGI, meaning AI that matches humans on nearly all thinking tasks. They say the next step is ASI, meaning AI that consistently beats large groups of human experts on tasks that take them years. They call this far from certain.
Why it matters to you
If AI thinking work gets far cheaper and more plentiful, many office and research tasks could become routine, low-cost services.
The takeaway
DeepMind researchers say the world should start planning for AI beyond human level, though they stress the outcome is not guaranteed.
What it means for you
The essay names data exhaustion, diminishing returns on model scale, and power and chip limits as bottlenecks, plus gaps like continual learning and sample efficiency.
The authors expect many cognitive tasks to become commodities, with standardization and automation of cognitive work across sectors.
The essay reports AI hardware spending growing about 2.5x per year over the past decade, and names physical limits on power plants, grids and chip fabs as constraints.