When Google DeepMind published its GraphCast weather forecasting model in late 2023, meteorologists were sceptical. A machine learning system predicting ten-day global weather forecasts, using a laptop-sized model, in under a minute? It seemed implausible. Then they tested it.
In independent evaluations, GraphCast outperformed the European Centre for Medium-Range Weather Forecasts on 90 percent of metrics. The era of AI-powered weather forecasting had arrived, and it had arrived faster than almost anyone expected.
The energy grid applications may ultimately be the most consequential. One of the fundamental challenges with renewable energy is that wind and solar are intermittent. Better forecasting allows grid operators to schedule backup generation more precisely, reducing the amount of fossil-fuel capacity that needs to be kept on standby.
Perhaps most ambitiously, AI is being applied to materials discovery for clean energy. Google DeepMind’s GNoME system discovered 2.2 million new crystal structures in 2023, including 380,000 that are predicted to be stable — a tenfold increase in the number of known stable materials. Among them are candidates for better batteries, more efficient solar cells, and novel hydrogen fuel cell catalysts.
Leave a Reply