Auto Amazon Links: No products found. Blocked by captcha.
output from solar arrays is tough. “It’s really hard to do this accurately,” Donti says. “It depends on so many factors – the time of day, the weather, even the reflections of sunlight off buildings. For a long time, these sites have been treated as a black box, with average output values. But these can be way off.”
At that recent conference, Koivuniemi shared insights from DeepMind’s Solar-Path finder project. “They’re planning to account for lots of different factors – soiling (dirt on the panels), shading from trees and buildings, and other unpredictable issues – to better predict how solar panels will behave. And they believe this will improve the quality of solar forecasts by 25 per cent, reducing the errors in predictions.”
This isn’t just a scientific curiosity. It’s about unlocking greater confidence in solar power, and other renewables, as the backbone of future energy systems. “If we can forecast solar power more accurately,” Donti says, “we can do a better job integrating it into the electrical grid. We can plan for when the power will be generated, move it around the grid, and make sure we’re managing supply and demand in a way that’s more sustainable and cheaper for consumers.”
A big part of the potential lies in AI’s ability to learn from data. Koivuniemi described how historical data can be used to predict how much energy a solar panel might produce, by ‘teaching’ a model with examples of data from similar sites. “So you can imagine in the future having the detailed specifications of your solar panels, and getting really good predictions of how much power they might output,” says Donti.
That’s just one example of the kind of scene-stealing solutions that AI is beginning to offer our transitioning world. For another, look to the AirClimate project, supported by US philanthropic group the Schmidt Futures Foundation, which has just launched a competition offering $20m to teams building new technology to map greenhouse gas emissions. The idea is to use data from atmospheric sensors plus satellite imagery to build a global map of CO2 emissions – gaps in our knowledge that AI can help fill.
“Then you can start monitoring, say, oil and gas operations in near real-time,” says Donti, “and see if they’re the source of local emissions problems. Uptake of this kind of technology could drive a virtuous cycle of accountability.”
And this, too, is just the beginning. The past few years have seen a change in attitude towards AI in dominant tech companies, driven by concerns from both employees and the public. Donti cites Google as a notable example. “The momentum for change started building around 2018,” she says. “Now Google’s AI principles are informed by environmental sustainability.
“It speaks to the fact that employees at these companies are saying, ‘We want to work for a company that really reflects our values. And we don’t want to work for a company that’s taking out large contracts with, say, the oil and gas industry. We’re more willing to work for a company if we know it’s aligned with our values.’ And so, the values of the individual employees start to steer the strategy of these companies. We’re seeing significant shifts.”
That whole world-change thing? It begins with individuals reaching – sometimes tentatively, sometimes boldly – across boundaries into worlds that are not theirs. The worlds of code and climate may indeed be different tribes. But, as Donti and Rolnick prove, it’s at their intersection that some of the most world-changing solutions are emergent.
