The idea that even if human innovation and technology were to fall into this mess, they could pull themselves out of it offers hope to even the most ardent observers of climate change. And there are few potential tools on the table, sparking exactly the same mixture of optimism and concern, promise and confusion as artificial intelligence.
AI includes advanced computer systems that can mimic parts of human cognition and has great potential to help humans fight climate change and be better prepared to deal with its impacts. Experts are working on ways to use machine learning, for example, to use resources more efficiently and to better predict increasingly common problems. extreme weather phenomenon.
But before AI can be harnessed in such a way, tech companies must put their programs through intensive training sessions and build or expand warehouse-sized data centers to support these systems. The densely packed computer servers in these systems require large amounts of water and energy to keep them cool and running smoothly.
New research from the University of California, Riverside suggests that tech companies have so far not done enough to fairly distribute the environmental impact of this growing AI.
Rather, technology companies appear to be repeating some of the patterns that unfolded over the last century in fossil fuel companies and many other industries, the study says. They are choosing to save costs by letting communities already strained on resources and with other additional strains bear the brunt of the environmental impacts associated with AI.
“Current methods of distributing AI computing based on cost are clearly disproportionately impacting certain regions already stressed by resources such as water and carbon issues,” said Xiaolei Ren, a computer engineering professor at the University of California, Riverside, who helped author the paper. paper.
“When AI wastes resources, AI’s net profit decreases.”
News came from the White House on Friday that the top tech companies had reached a deal. Voluntary safeguards against AIorganizations around the world — united nations To AI Now Laboratory — It also calls for policies that prioritize the development of AI in an environmentally sustainable and equitable manner.
Case Study: Phoenix
An example of what can happen when environmental fairness is not factored into the AI equation is unfolding hundreds of miles east of Ren’s university.
Phoenix and its neighboring communities have become go-to places for technology companies to build data centers. Land and electricity are cheaper in Arizona than in many other areas, Wren said, and Arizona also offers attractive tax incentives for businesses.
Google broke ground on a $1 billion data center earlier this month in Mesa, a suburb of Phoenix. The campus, which will eventually cover 750,000 square feet, will help power Google’s existing tools and “continuous artificial intelligence innovation,” the company said in a statement.
Microsoft will open one data center near Phoenix in 2021 and continues to add to the complex. Facebook’s parent company Meta is also building a data center there. many other tech companies is also trending.
AI requires an amazing level of computational speed, especially during the training phase. The calculation for that level is big energy demand. And since much of the country is still powered by coal and other fossil fuels, the new AI is causing massive carbon emissions and increasing stress on the country’s energy grid.
Training GPT-3, a cousin to the well-known AI system ChatGPT, consumed more than 1,000 megawatt hours of power, Ren said. That’s the same amount of energy it would take him to run over 100 homes for a year.
These data centers also require reliable cooling systems to prevent rows of servers from overheating. Companies typically use liquid cooling systems, which pump water in a closed loop to dissipate heat and keep things running smoothly.
Citing its commitment to transparency, Google announced last fall that its global data centers would be 4.3 billion gallons of water By 2021, this will roughly match the water needed to irrigate and maintain 29 golf courses in the Southwest, the company said. Ren said the company’s center in The Dalles, Oregon, accounts for about a third of the city’s annual water consumption.
Meta’s voluntary report showed 1.3 billion gallons are used to cool 17 data centers in that same year.
Len said Microsoft’s GPT-3 training at its US data center is the most advanced facility available and still requires 700,000 liters of water. This is the same amount of water used by over 2,000 people on average every day. It also doesn’t include the water needed to power data centers, because coal, nuclear and other types of power plants require large amounts of water to operate, Ren said.
Companies such as Intel that make the core chips for these systems are also based in Phoenix, but the process is also very water and energy intensive, Ren said.
But Phoenix, like many other places in the United States, is feeling the strain of global warming. The region has suffered from particularly severe water shortages for many years due to ongoing droughts that threaten to cut off supplies from the Colorado River and other sources. Phoenix is also in the midst of a record heat wave, with temperatures topping 110 degrees Celsius every day this month.
The situation has prompted some residents and political leaders to push back against the influx of tech companies looking to build data centers in their backyards.
For example, water problems delayed Microsoft’s plans for building a data center campus. And earlier this summer, Arizona Governor Katie Hobbs said: Restrict new construction around Phoenix After an investigation by state officials found that there was not enough groundwater left to meet the expected demand for the next century.
Categorize the “solution”
Ren said that at some point, water usage restrictions and skyrocketing water bills could make places like Phoenix an unattractive place to build data centers. But judging by the number of ongoing projects, it’s clear that day is not today.
Instead, tech companies have so far largely responded to resource pressures at home and abroad by pledging to draw power from renewable sources such as solar projects, pay for water reclamation projects elsewhere, and switch to air-cooling key systems to keep servers from overheating.
However, Ren pointed out that the air cooling system only works when the outside temperature is below 85 degrees. Over the next few weeks, Phoenix’s lowest temperature is expected not to drop below 87 degrees.
Also, according to Google, air cooling systems require about 10% more energy. Unless companies take steps to ensure that power comes from renewable sources, switching to air-cooling systems will actually increase both water use and carbon emissions outside the facility, while reducing on-site water use, Ren said.
So what can tech companies do to reduce the environmental impact of AI and ensure that impact is more evenly distributed?
First, Ren said, companies can choose to build data centers in locations where water and heat waves are less of a concern. It can also work to build the most advanced system possible, Ren said. This is tied up with renewable energy projects and includes a mechanical upgrade to allow water to be used for cooling several times before being discharged.
Both moves could increase the cost of building data centers. This is where tax incentives and subsidies come into play, which, combined with buying goodwill from consumers while mitigating potential financial risks related to climate change, can help offset price differences.
Meanwhile, parts of Europe are beginning to require companies to factor climate change costs into their projects through mechanisms such as carbon taxes. As such, they may be forced to either mitigate the impact of their projects, relocate them elsewhere, or siphon higher prices and use the money for climate programs.
Such strategies may be useful for future data center projects. But Ren and his co-authors argue that there are operational changes companies can make to existing centers to reduce their environmental impact and distribute them in a smarter way.
“These large companies have many data centers around the world, so they can shift workloads unnoticed or actually shift AI computing from one data center to another without any impact on the user experience,” Ren said. “And a more equitable distribution of environmental costs across different regions requires clear decisions about how to distribute this AI computing across different data centers.”
For example, the same AI that powers these systems can be used to identify data centers that can operate with the lowest carbon and water footprint at any given time, and shift workloads to those locations.
Currently, the system is optimized to shift workloads to where power is cheapest every few minutes, Ren said. But if it’s trained to incorporate climate data into its calculations, it could, for example, move an AI training program from its Phoenix facility to its Washington facility over the summer. Alternatively, daily workloads can be shifted from a center in Virginia that is primarily coal-powered to a center in Texas that relies primarily on solar energy.
Ren said several companies, including Google and Microsoft, are trying to schedule workloads based on real-time availability of renewable energy. But for now, these plans are still experimental and not standard practice.
When asked how climate issues are affecting decisions about where to build data centers and what steps the company is taking to ensure that AI-related advances don’t create new environmental injustices, a Google spokesperson said he withheld the company’s view. autumn statement It advertises a “climate-friendly approach”, including using recycled water for cooling wherever possible. Meta did not respond by the deadline. And a Microsoft spokesperson said, “We have nothing to share at this time.”
Despite his concerns about how companies are building AI systems today, Ren said he remains optimistic about the future of machine learning as a tool to help fight climate change.
His own team at the University of California, Riverside, uses AI to plan energy-intensive AI development work during the hours with the lowest carbon footprint. He also pointed to machine learning’s ability to, for example, significantly reduce water use for farmers by incorporating advanced weather forecasts, and manage building air-conditioning systems more efficiently.
“The use of AI is a concern,” Ren said. “But it has great potential to reduce environmental costs in other sectors.”
https://www.mercurynews.com/2023/07/24/ai-can-help-fight-climate-change-and-injustice-if-it-doesnt-make-them-worse-first/ AI can help fight climate change and injustice – first, if it doesn’t make climate change worse