With the expanding utilization of artificial intelligence and the contemplation within societies on how to harness this potent tool for enhancing our lives, boosting productivity, and confronting our most urgent challenges, only a few have contemplated its implications for the environment.
While certain individuals have underscored AI’s potential in addressing environmental issues, others emphasize the necessity of comprehending the carbon footprint attributed to AI itself.
Champions of technological progress, similar to those endorsing cryptocurrency, initially celebrated its potential for curbing carbon emissions, only to be contradicted by wasteful practices like energy-intensive Bitcoin mining.
Nonetheless, experts widely view AI as a positive advancement, with the United Nations Environment Program commending it as a resource capable of enhancing our comprehension of environmental impact and climate change effects.
AI’s potential is vast—it can efficiently process massive volumes of data, such as satellite imagery used by researchers to monitor climate alterations, as articulated by Sasha Luccioni, who specializes in scrutinizing AI models for sustainability. AI enables scientists to refine climate models, recognize trends, and formulate predictions, thereby fostering a clearer grasp of climate change and effective mitigation tactics.
Furthermore, AI’s potential extends to conserving water, combating wildfires, and even identifying and recovering recyclable materials. Luccioni emphasized the manifold applications of AI in diverse climate change sectors, encompassing everything from optimizing electricity distribution networks to tracking biodiversity.
However, a subset of experts focuses on AI’s own carbon footprint. They assert that enterprises aiming to implement AI should exhibit transparency concerning its environmental ramifications and the strategies they are employing to address them.
What is AI’s carbon footprint and why is it worrying some environmental advocates?
The comprehensive environmental impact of AI is intricate to quantify, yet its origins are rooted in the computers it employs. The essential materials required for crafting computer hardware are sourced through mining, a process that can be both resource-intensive and environmentally taxing, as highlighted by Shaolei Ren, an associate professor of electrical and computer engineering at the University of California, Riverside.
Once the requisite hardware is acquired, the process of training an AI model can be exceedingly energy-intensive. While AI firms often withhold precise energy consumption figures, researchers have ventured estimations based on available data. In an unverified study led by Ren and fellow experts, it’s projected that the training of GPT-3, the language model underpinning ChatGPT, might have potentially consumed approximately 700,000 liters of freshwater. The water employed for cooling data centers usually evaporates, rendering it non-recyclable.
There is also the concern of carbon emissions. Researchers at the University of Massachusetts, Amherst, have determined that the training process for a single AI model can release over 626,000 pounds of carbon dioxide. This is approximately equivalent to the greenhouse gas emissions produced by 62.6 gasoline-powered passenger vehicles being driven for a year. Carbon dioxide constitutes the predominant portion of greenhouse gas emissions, contributing to climate change by trapping heat in the atmosphere.
Having consulted these independent assessments, CBS News engaged AI language models in discussions about the environmental impact of the technology.
Bard, developed by Google, noted the challenge of precise estimation. ChatGPT, created by OpenAI, clarified that as an AI language model, it does not possess a direct carbon footprint. However, with an estimated 100 million monthly active users, there exists an associated footprint related to the electricity and computational resources required to operate the servers hosting and powering the model. (OpenAI did not provide comments for this report.)
Microsoft, which has made substantial investments in OpenAI, declined to share assessments regarding the carbon footprint linked to the development of AI tools.
“AI will serve as a potent tool for advancing sustainable solutions, but to power this novel technology with escalating consumption needs, a substantial supply of clean energy on a global scale is necessary,” stated a Microsoft spokesperson. “Microsoft is committed to researching the measurement of energy consumption and carbon impact resulting from AI while concurrently striving to enhance the efficiency of large systems in both training and application.”
Can AI tools be designed in an environmentally-conscious way?
The process of training, deploying, and operating AI systems can be notably energy-intensive, leading companies to consider potential repercussions during the development phase, as highlighted by Junhong Chen, a professor of molecular engineering at the Pritzker School of Molecular Engineering and the principal water strategist at Argonne National Laboratory.
“When designing these systems, it’s essential to be cognizant of the possible adverse effects and strive to minimize them right from the outset through thoughtful design,” Chen remarked.
Findings from Google reveal that water-cooled data centers emit approximately 10% fewer carbon emissions than their air-cooled counterparts. According to the Energy Department, data centers rank among the most energy-intensive types of structures in the U.S., consuming 10 to 50 times more energy per unit of floor space compared to standard commercial office buildings. Collectively, these centers account for approximately 2% of total electricity consumption in the U.S.
For Google’s selection of new data center sites—largely determined by proximity to users—the company assesses the availability of reclaimed and nonpotable water resources in the vicinity, stated Ben Townsend, Google’s head of data center sustainability.
“Data centers share many similarities with personal computers. They necessitate space, energy, and cooling,” Townsend explained.
Regarding energy grids, a delicate equilibrium must be struck, noted Ram Rajagopal, who leads the Stanford Sustainable Systems Lab. With an emphasis on decarbonization and resilience, AI can contribute to optimizing electricity systems, lowering costs, expanding deployments, and devising efficient strategies for reducing greenhouse gas emissions.
Yet, as the adoption of AI becomes more widespread, the existing data centers managing AI tasks might prove inadequate.
“As the scale of AI usage expands, it leads to a bottleneck in terms of data centers, necessitating their expansion and subsequently increasing power consumption,” Rajagopal cautioned.
How can AI help?
Scientists are currently harnessing AI for a multitude of beneficial applications. AI models have proven instrumental in aiding researchers to uncover methods for water recycling and reuse. By identifying contaminants within water and devising optimal extraction techniques, these models contribute to advancing the field, as explained by Professor Junhong Chen, an expert in molecular engineering. Furthermore, AI holds the potential to explore ways of repurposing these contaminants for alternative uses.
In a recent collaborative initiative involving Google, American Airlines, and Breakthrough Energy, AI was employed to analyze and sift through diverse datasets, including satellite imagery, weather data, and flight paths. The objective was to create comprehensive maps capable of predicting contrails—those faint, white streaks often visible behind airplanes. This research aids pilots in optimizing flight routes to minimize contrail formation. Given that contrails are responsible for approximately 35% of the aviation sector’s overall global warming impact, these AI-derived insights have the potential to significantly mitigate the industry’s environmental footprint.
Artificial intelligence also finds practical application in the realm of battery research, offering optimization opportunities for lithium batteries that serve as the core power source for most electric vehicles, according to experts.
Numerous enterprises, including AMP Robotics and MachineX, have developed AI-driven tools to enable robotic identification and retrieval of recyclable materials. AMP Robotics, boasting over 300 deployed AI systems worldwide, reports that these robots can outperform humans in collecting recycled materials by up to twice the speed and with heightened consistency.
Through these innovations, AMP Robotics claims to have prevented nearly 1.8 million metric tons of greenhouse gas emissions—equivalent to removing around 375,000 cars from the roads—by enhancing recycling efficiency.
In California, scientists employ AI to combat wildfires. Cameras linked to AI can swiftly identify wildfires and detect smoke before they propagate extensively, marking a significant advancement in wildfire prediction, as affirmed by Cal Fire Battalion Chief David Krussow.
At the National Oceanic and Atmospheric Administration, AI assists in refining climate, weather, and earth system models. Meanwhile, the United Nations Environment Program leverages AI to scrutinize and forecast atmospheric carbon dioxide concentration, glacier mass changes, and rising sea levels. This tool aims to function as a “mission control” for the planet, as explained by David Jensen, a coordinator within the team. Additionally, the International Methane Emissions Observatory (IMEO), developed by the U.N., utilizes AI to monitor and mitigate methane emissions, a potent greenhouse gas influencing global temperature.
According to Jensen, utilizing data-driven strategies to curb methane emissions from the energy sector presents a swift, feasible, and cost-effective approach to mitigating climate change impacts.
However, despite the potential, both companies and AI language models acknowledge the uncertainties surrounding the technology’s environmental consequences.
ChatGPT responded to CBS News, stating, “My purpose is to assist users in accessing information and knowledge on various topics, including environmental matters, to enable informed decisions and actions aligned with a better and sustainable world.”
Google’s Chief Sustainability Officer, Kate Brandt, acknowledged the challenge in predicting AI’s future energy usage and emissions growth. She emphasized that the environmental effect of AI models would hinge on their application, underscoring the positive impact through solving environmental issues versus the potential negative impact by creating new problems.