Artificial intelligence is powering more and more of the things we interact with every day, from our gadgets to our cars. But it’s also playing a growing role in how society’s undergirding resources — energy, food, and water — are sourced, secured, and delivered.

In this three-part series, we’ll consider how AI is being used to make those resources more environmentally and financially sustainable.

AI is not a discrete technology, but rather a school of powerful and widely applicable data science tools, which include machine learning, pattern recognition, and natural language processing. All of these tools can squeeze far more useful information out of data, and more quickly than humans could reasonably do on their own.

Without energy, we wouldn’t be able to grow food or access water, so let’s start at the power outlet.

Supply and Demand

Homes and businesses account for nearly 40 percent of U.S. energy consumption, which is why so much effort has been made to boost the efficiencies of buildings. But swapping in compact fluorescent light bulbs does only so much, especially in places with huge energy demand, such as data centers. Google’s subsidiary DeepMind, based in London, applies AI to control fans and cooling systems for its data centers, reducing the amount of energy needed to manage the indoor temperature (versus the energy needed to run the IT equipment) as much as possible.

Commercial buildings, such as offices, hotels, or factories, can also be major energy hogs. A six-year-old Silicon Valley startup called Verdigris integrates its sensors into gear like heating and cooling systems, manufacturing equipment, and washing machines and uses machine learning to remotely identify faulty, inefficient equipment.

Other energy management companies are focusing on the human occupants in buildings to reduce energy consumption. Startup Comfy uses a smartphone interface through which office workers can communicate: “Hey, I’m freezing,” or the opposite. Based on this, Comfy (over days and weeks) uses AI to learn personal preferences and tweak the temperature in individual zones to make it more comfortable for employees. According to Comfy, the process reduces overall heating and cooling energy use and saves money.

At the macro level of the power grid, a great deal of energy is wasted because of mismatched supply and demand. Energy providers and utilities have for years offered various schemes designed to help bring those factors into balance, such as the often-kludgy demand-response programs, in which consumers opt in to reduce consumption during heat waves or other periods when prices spike and utilities fire up more power plants.

To refine the process by which utilities predict demand (and therefore price), a startup called Drift is turning to data science. The Seattle-based company uses machine learning to analyze a host of unconventional and granular data — such as internet search activity, the condition of energy infrastructure, and even business hours — to improve demand forecasts.

Drift then buys power from producers like hydroelectric plants or from large energy consumers, such as businesses that agree to reduce consumption during low-demand periods. The company sells that power at a rate that allows it compete with energy incumbents. Drift just launched its service in New York this year.

Plugging in to Demand

Drift’s approach, if successful, could disrupt the way electricity is priced and sold. But consumers who can afford to invest in rooftop solar panels have already been doing that for a while (at least in markets where utilities allow them to sell excess energy to the grid through a process called net metering).

The solar industry, now filled with fierce competition, saw a record year in 2016. But there’s a downside to all that upside for solar providers: In the crowded residential solar market, attracting customers is an increasingly expensive process.

An Oakland, California–based startup called PowerScout says it uses machine learning to slash those customer acquisition costs and help solar installers target their marketing to the most likely buyers. “Acquiring a customer costs more than the solar panels themselves. That’s how crazy it is,” explains Attila Toth, founder and CEO of PowerScout.

An antiquated solar sales model also compounds the high costs of signing up new customers. Solar panels are “sold mostly through door-to-door and unsophisticated practices,” Toth says. “The guys in the trenches are good [solar] installers but are stuck in the Yellow Pages era of marketing.”

PowerScout’s platform mines data — 100 billion data points related to 45 million households — from a range of sources to estimate the likelihood that a given household will invest in solar energy. The company then profiles current solar users to extrapolate and build propensity models and uses machine learning to improve the models over time, based in part on data that potential customers provide.

Optimizing Supply

When it comes to producing energy, AI is being used to squeeze efficiencies out of solar plants, as well as wind farms and even oil wells.

General Electric makes many of the parts of our energy infrastructure, from massive wind turbines to gas-powered turbines used in more conventional power plants. The company has spent years developing software, called Predix, that can interpret sensor data from that equipment and uses artificial intelligence to make its machines both operate more efficiently and predict failures before they happen. GE is far from the only vendor of such technology: C3 IoT, founded by tech titan Tom Siebel, is another example, and IBM says its Watson AI platform can be used to track the performance and health of energy infrastructure.

Of course, specific types of energy generation have specific efficiency challenges. A solar panel that follows the arc of the sun can collect 25 percent more energy than a static panel, which is why a company called NEXTracker has seen success.

The company sells software-controlled motors that use networked sensors to tweak solar panel position to capture maximum sun exposure. They’ve been installed at 1,000 solar plants around the world. Last year, NEXTracker acquired predictive modeling startup BrightBox Technologies, and it just released a new product called TrueCapture, which uses BrightBox’s machine learning algorithms to further refine and improve panel tracking.

“Our job is to lower the cost of solar electricity generation,” says Dan Shugar, CEO of NEXTracker, noting that TrueCapture allows solar farms to “make more energy with the same equipment and the same amount of land.” The platform can boost energy production between 2 and 6 percent, the company says.

Conventional (fossil fuel) energy producers are not out of the AI loop. Entrepreneur Dakin Sloss co-founded startup Tachyus to help oil and gas producers squeeze more hydrocarbons out of the ground. But he’s not a petroleum engineer or a hydrologist. He’s a data geek who previously founded government data analysis company OpenGov.

Sloss says he landed on the oil and gas industry because its major players were collecting a great deal of data but not putting that data to best use. Producers use seismic data to build best-guess models of the subsurface showing the composition of underground oil or gas deposits, and they use these models to determine how best to extract it. The problem is that building such a subsurface model can take six months, and by the time it’s completed, the subsurface could have been significantly altered by drilling.

Tachyus uses data science techniques to build models to more quickly and closely resemble the subsurface composition. According to the company, this approach has helped its customers reduce oil and gas production costs by 40 percent and boost production by 20 percent.

Sloss, of course, knows the oil and gas industry has many detractors. He defends it by pointing to the vital role it plays in society. “AI is making energy much more intelligent, in terms of how it’s produced,” he says. “And we should all be very glad about that.”

Whether it’s used to manage temperature in a data center, predict energy demand, keep office workers comfortable, help consumers get the most affordable solar power, or even help oil and gas producers operate more efficiently, AI is being used to boost benefits while conserving resources.