Responsible AI: Acknowledging environmental impacts

Artificial intelligence (AI), particularly generative AI (GenAI), holds immense potential to accelerate businesses and drive innovation across various sectors. Companies everywhere are onboarding new AI tools, assistants, and agentic automations with the hopes of cutting costs and improving efficiency. While AI promises numerous benefits for society at large, many individuals, like MIT researchers, are raising concerns about the indirect environmental impacts this will have. Businesses should consider responsible AI practices when implementing these new technologies to mitigate potential risks and maintain customer trust.

What are the environmental impacts of AI?

AI has impacts across the environmental spectrum, including energy, water, carbon emissions, waste, mining, and pollution. Businesses that choose to leverage the cutting-edge technology of AI must decide what to do about the potential impacts it will have on the environment — especially companies publicly working toward net-zero emissions.

Brands that claim to work toward sustainability but don't address the impacts of using large data centers and advanced AI models risk losing brand power due to inconsistent public messaging. While all the potential impacts of AI are still being researched (the technology is advancing quite rapidly), here are some of the common causes for concern when using AI without limit:

Energy consumption

Training large AI models requires substantial electricity, especially for models like ChatGPT-4 or image/video generation tools. Scientists from Google and the University of California at Berkeley estimated that training GPT-3 used 1,287 megawatt-hours (MWh) of electricity. This is equivalent to powering 120 average homes in the US for a year (generating about 552 tons of carbon dioxide).

Energy consumption extends beyond the training of models to actual usage. An average ChatGPT query consumes five times as much energy as a simple web search — and most consumers are unaware of the energy difference.

According to MIT, "Scientists have estimated that the power requirements of data centers in North America increased from 2,688 megawatts at the end of 2022 to 5,341 megawatts at the end of 2023, partly driven by the demands of generative AI."

Data centers that power AI tools also consume enormous amounts of energy for both computation and cooling. And data centers in general disrupt local ecosystems (including humans!) with their powerful lights, massive land footprints, and persistent noise pollution.

Water usage

Most concerning, and perhaps the least discussed, is AI's massive water consumption. Researchers note that there is a "secret footprint" of water usage in AI, coming from cooling these massive data centers. The global AI demand in 2027 is projected to account for 4.2-6.6 billion cubic meters of water withdrawal, which is around half of the United Kingdom’s annual water withdrawal. According to the same source, training GPT-3 evaporated 700,000 liters of clean freshwater.

Data centers require chilled water to cool their equipment, absorbing the heat generated by computing devices. It's estimated that for each kilowatt-hour of energy a data center consumes, it would need two liters of water for cooling.

Carbon emissions

The carbon footprint of AI varies based on:

• The source of electricity (coal vs renewables)
• The number of training and inference runs
• How optimized the data centers are for efficiency

Training large models can emit as much CO2 as five or more cars over their lifetime, according to a 2019 study.

Tech giants are already experiencing a significant surge in emissions due to the use of AI and are having to make immediate adjustments to meet their public net-zero pledges.

Rare mineral use

The hardware needed for AI, like GPUs and TPUs, can accelerate the depletion of our natural resources, especially due to the need for rare earth elements. These elements are energy-intensive to mine and often come from environmentally sensitive regions. Without proper boundaries or legislation, companies could contribute to the exploitation of certain geographic regions and people groups, while also draining precious resources and increasing tech waste.

Infrastructure demands

AI accelerates the turnover of high-performance computing hardware, leading to:

• Increased e-waste
• Shortened device lifespans in data centers and edge devices

AI also increases the demands on existing systems, which means more infrastructure will need to be built, including cloud computing, storage, and data centers. All of these infrastructure demands come with their own pollution, waste, and environmental impacts. The need for proper zoning for data centers and the impacts they can have on chronic illness for those who live nearby are just now being brought into the light.

How can responsible AI help organizations address these concerns?

The environmental impacts of AI are concerning, but researchers and experts are looking for clear ways to address them. It's also up to companies to advocate for and fund these responsible AI practices, so that new technology can continue to help improve our lives without costing us the environment.

Here are some ways businesses can help mitigate the environmental concerns with AI:

• Efficiency improvements: New model architectures, pruning, and distillation help reduce training and inference costs. Companies can advocate for more efficiency in the models they purchase or build to further drive these improvements. Training models at times that lower energy and water usage (like at night when it's cooler or during the winter compared to summer) can help as well.
• Green AI: Individuals are advocating for Green AI, a movement that emphasizes energy-efficient, resource-conscious AI development. Companies can help fund these organizations and elect to use Green AI models over others that are not eco-conscious.
• Climate monitoring and reporting: AI can better monitor climate and sustainability initiatives for companies — improving energy efficiency, reducing waste, and modeling climate scenarios. With advanced, predictive reporting, companies can make decisions that lower emissions across their enterprise and AI usage.
• Renewable-powered data centers: Tech enterprises like Google and Microsoft are committing to clean energy for AI workloads. Other companies can join in and commit to a greener future, thereby increasing the likelihood of success for the entire sector.
• More transparency: Discussing the environmental impacts of AI with end users will help educate everyone on the potential impacts of their usage. Users may opt for simple web searches over GenAI when advanced models are not needed, lowering energy and water usage. Educating the public about current environmental challenges enables them to make informed choices when purchasing new products or adopting new technologies.

Technology experts and academics conclude, "Responsible development of Gen-AI requires a focus on sustainability beyond efficiency improvements' as we need "Gen-AI to develop in ways that support social and environmental sustainability goals alongside economic opportunity."

Visualize a better future, with Spotfire

AI isn't going anywhere — and companies that want to stay on the cutting edge of what's next need to harness it to help reduce costs, optimize operations, and fuel innovation. However, we all have a responsibility to make sure these advancements take into consideration the social, environmental, and security concerns they raise.

Interested in learning more about responsible AI and its implications for your business? Check out this blog on how to keep humans at the center of your AI practices.

You can also learn more about our GenAI assistant, the Spotfire Copilot^TM AI tool, which is agnostic and can leverage LLMs from any source, allowing you to have more control over the kinds of AI used in your business environment.