Artificial intelligence can increase productivity and advance research in nearly every field, including sustainability, but its environmental impacts should not be ignored.
Over the past few years, artificial intelligence (AI) has risen in popularity across all sectors. AI’s ability to solve complex problems quickly makes it an extremely useful tool. AI can even be useful in combating climate change, including tracking emissions, weather pattern modeling, and informing disaster response. A recent Massachusetts Institute of Technology study used AI to find more sustainable ingredients for concrete, a carbon-intensive construction material. Environmental nonprofit The Ocean Cleanup is using AI to accelerate plastic waste removal from the world’s oceans.
While AI can be used effectively in support of resource conservation and sustainability efforts, the use of this technology comes at a large environmental cost. AI is powered similarly to traditional internet search engines but at a much higher rate. It is estimated that using a generative AI service such as ChatGPT or Copilot for a single query can take seven to 10 times more energy than a simple Google web search.
AI computer infrastructure, which includes data centers, processors, and other specialized computing hardware, can be broken down into a four-stage lifecycle: production, transport, operations, and end-of-life. The production stage of AI includes the mining of natural resources such as cobalt, silicon, gold, and other metals. For example, making an approximately 4-pound computer housed in a data center requires around 1,763 pounds of raw materials. The mining and production of these metals have both environmental and human costs, including soil erosion, pollution, and unjust labor conditions for miners. The transport stage of AI includes the distribution, freight transportation, and handling and storage of the hardware. Information and communications technology accounts for a small part of greenhouse gas emissions within the global transportation sector.
The operations stage produces the most greenhouse gas emissions in AI’s lifecycle and also has a considerable water footprint. Global AI energy demand is projected to increase to at least 10 times the current level and exceed the annual electricity consumption of the country of Belgium by 2026. Data center energy consumption is estimated to be about 6 percent of the United States’ total electricity usage in 2026. In 2022, Google’s data centers consumed about 5 billion gallons of fresh water for cooling, increasing by 20 percent from 2021. During this timeframe of a single year, Microsoft’s water use increased by 34 percent. By 2027, it is estimated that AI’s projected water usage could reach between 4.2 and 6.6 billion cubic meters, equal to four to six times the annual water usage of the entire country of Denmark.
The end-of-life stage of AI technology has negative environmental impacts in the form of electronic waste. Electronic waste from data centers often contains hazardous substances like mercury and lead, which can contaminate soil and water when not disposed of correctly.
Like many environmental issues, there is no silver bullet or one right answer. While the negative environmental impacts of AI are staggering, the potential for its use to advance climate solutions and protect natural resources cannot be ignored. So what can we as individuals do? The University of Maryland, Baltimore (UMB) Office of Sustainability invites you to ask yourself the following questions before you gravitate toward using AI for your task:
- Do I really need AI to find this information, or will a simple search engine get me the result I need?
- Is the information I’m inputting a privacy or security risk for myself or my work if breached? This is a particularly important question for open-source AI tools that utilize user input data to further train themselves.
- Does my use of AI for this task violate UMB’s standards for academic integrity?
- For aspects of AI that automatically show up (like a search engine’s AI summary at the top of the page), are they useful? If not, is there a way to turn them off or opt out? Hint: for Google AI Summary, the answer is yes, like using browser plug-ins or simply typing “-AI” at the end of your search to remove the summary.
- Am I making every inquiry count? Even though it might seem impolite, you can skip saying “thank you” to AI. Even this two-word phrase requires a full response back from AI, using additional electricity and water. Additionally, check out these tips from UMB’s Kris Nicklas on how to craft effective prompts with clear language to get the results you’re looking for more concisely.
Because of AI’s status as a relatively new technology and its rapid expansion and adoption on a global scale, there are still many unknown variables when discussing its impacts, both positive and negative. Estimates related to its current and future electricity and water usage are just that: estimates.
We at the Office of Sustainability encourage the UMB community to uphold the University’s core values — including Well-Being and Sustainability, and Innovation and Discovery — when using AI. Make an effort to strike a good balance using the tips in this article, your best judgment, and UMB’s core values as a guide.