The singularity may be sustainable after all.
A report released at the Davos meeting of the World Economic Forum outlines ways to make AI not only human-friendly, but “earth-friendly.”
This could be important, what with Google CEO Sundar Pichai saying at Davos, “AI is probably the most important thing humanity has ever worked on. I think of it as something more profound than electricity or fire.”
The big shift will come in the next decade or so as AI develops truly autonomous decision-making, toward presumably agreed-upon goals.
The report, from PwC, cites eight AI “game changers” across autonomous vehicles, distributed energy grids, smart agriculture, livable cities and climate and disaster response. For example, “Substantial greenhouse gas reductions for urban transport can be unlocked through route and traffic optimisation, eco-driving algorithms, programmed ‘platooning’ of cars to traffic, and autonomous ride-sharing services.”
Likewise, “Deep reinforcement learning may one day be integrated into disaster simulations to determine optimal response strategies, similar to the way AI is currently being used to identify the best move in games like AlphaGo.”
Values-aligned
AlphaGo was developed by DeepMind, now a unit of Google’s parent, and is arguably the best Go player ever. It uses “reinforcement learning,” in which the AI agent learns by trial and error with little or not initial data.
The approach, which requires less time and computer processing power, has primarily been used for gaming agents, but is being applied to corporate strategic analysis, process optimization and other areas where the rules are well known. Real-world and natural systems are messier and less bounded, but the report suggests that “deep reinforcement” may be able to solve such challenges by 2030, if not earlier.
A key goal, the report says, is to ensure that AI becomes value-aligned, “that its idea of a good future is aligned with humanity’s values.”
The approach of the fabled singularity, the moment when artificial superintelligence takes off, requires checks and balances to ensure that AI systems remain not only human-friendly, but earth-friendly as well.
