AI is changing the grid. Could it help more than it harms?

The increasing popularity of AI leads to an increase in electricity requirements, which is so significant that it can redesign our network. The energy consumption by data centers rose by 80% from 2020 to 2025 and will probably continue to grow. Electricity prices are already increasing, especially in places where the data centers are most concentrated. However, many people, especially in Big Tech, argue that AI will be a positive force for the network. They claim that the technology could help to become online faster, to carry out our electricity supply system more efficiently and to provide and prevent errors that cause power failures. This story is part of the "Power Hungry: Ai and our Energy Future" series by with Technology Review to the energy requirements and carbon costs of the artificial intelligence revolution. There are early examples in which AI already helps, including AI tools that use supply companies to forecast the supply and demand. The question is whether these big promises are realized quickly enough to outweigh the negative effects of AI on local networks and communities. A sensitive balance in an area in which AI is already used for the network is forecast, says Utkarsha Agwan, member of the non -profit group climate change AI. Running the grid is a balancing act: the operators have to understand how much electricity there is, and turn on the correct combination of power plants in order to meet them. You optimize on the way to the economy and select the sources that keep prices for the entire system lowest. This makes it necessary to look ahead for hours and in some cases. The operators take into account factors such as historical data (holidays often see a higher demand) and the weather (a hot day means that more climate systems suck power). These predictions also take into account which level of care is expected of intermittent sources such as solar panels. In the case of the forecast, there is little risk when using AI tools. It is often not as sensitive to time as with other applications that can require reactions within seconds or even millisecond reactions. A network operator can use a forecast to determine which plants have to switch on. Other groups may also carry out their own forecasts and use AI tools, for example to decide a system how a system can be occupied. The tools cannot control anything physically. Rather, you can be used in addition to more conventional methods to provide more data. Nowadays, the network operators make many approaches to model the network, since the system is so incredibly complex that it is impossible to really know what is going on at every place. There are not only a whole series of power plants and consumers that you can think about, but there are also considerations of how to ensure that power lines are not overloaded. Working with these estimates can lead to some inefficiencies, says Kyri Baker, professor at the University of Colorado Boulder. The operators tend to create a little more electricity, for example, than the system. The use of AI to create a better model can reduce some of these losses and enable the operators to make decisions about how the infrastructure can be controlled in real time in order to achieve a closer range of supply and demand. It gives the example of a trip to the airport. Imagine there is a route you know that you will take you there in about 45 minutes. There could be a different, more complicated route that you could do in ideal conditions - but you are not sure whether it is better on a certain day. What the grid is doing now corresponds to the reliable route. "So this is the gap that AI can close. We can solve this more complex problem quickly enough and reliably enough so that we may be able to use it and switch off emissions," says Baker. Theoretically, AI could be used to operate the network completely without human intervention. However, this work is largely in the research phase. Network operators operate one of the most critical infrastructure in this country, and the industry is hesitating to deal with something that already works, says Baker. If this type of technology is ever used in network operation, there will still be people in the loop to make decisions, at least when it is used for the first time. The predecessor of another fertile area for AI is planning future updates of the network. Construction of a power plant can take a long time - the typical time of a first request for commercial operation in the USA is about four years. One reason for the lengthy waiting is that new power plants have to prove how they could affect the rest of the network before they can connect. In a connection study, it examines whether adding a new power plant of a certain type require upgrade to the network at a certain location to prevent problems. After the regulatory authorities and supply companies have determined which upgrades are needed, they appreciate the costs and the energy developers generally falls into account. Today these studies can take months. They contain try to understand an incredibly complicated system, and since they rely on estimates of other existing and proposed power plants, only a few can occur in one area at a certain point in time. This has contributed to creating the years of connecting queue, a long series of plants that are waiting for them to connect to the network in markets such as the USA and Europe. The vast majority of the projects in today's queue are renewable energies, which means that clean electricity is only waiting to come online. AI could help accelerate this process and create these reports faster. The independent system operator of Midcontinent, a grid operator who covers 15 countries in the central USA, is currently working with a company called Pearl Street to automate these reports. AI will not be a remedy for network planning. There are other steps to eliminate the connecting queue, including securing the necessary permits. But the technology could help take things with them. "The earlier we can accelerate the connection, the better we get," says Rob Gramlich, President of Grid Strategies, a consultation specializing in transmission and electricity markets. There is a growing list of other potential uses for AI online and in electricity generation. In advance, the technology could be monitored and planning of failures in devices that range from power lines to gears. Computer vision could help recognize everything from forest fires to faulty lines. AI could also help compensate for supply and demand in virtual power plants, systems of distributed resources such as EV chargers or intelligent hot water. While there are early examples of research and pilot programs for AI from Grid planning to the company, some experts are skeptical that the technology at the level that some hope for. "It is not that the AI ​​had no transformation into power supply systems," says Agwan from AI from Climate Change. "It is the case that the promise was increasing and hope was increasingly bigger." In some places, higher electricity prices are already recorded due to the electricity requirements of data centers. The situation will probably worsen. The electricity requirement from data centers will be twice as high by the end of the decade as the 945 terawatt hours, approximately the annual demand from all over the country Japan. The infrastructure growth required to support AI load growth has exceeded the promises of technology "from quite a bit", says Panayiotis Moutis, assistant professor of electrical engineering at the City College of New York. Higher invoices caused by the increasing energy requirements of AI are not justified by existing options for using the technology for the network, he says. "At the moment I am very hesitant to lean on the side of the AI ​​that is a silver ball," says Moutis. Correction: This story has been updated to correct the Moutis belonging.