By. Michela Taufer, Ph.D. and Chandra Krintz, Ph.D.
From advanced weather forecasting systems to precision medicine tailored to individuals, AI is rapidly transforming virtually every sector of industry and culture. However, as we unlock AI’s full potential, society faces an unsettling paradox: the technology used to solve global problems may exacerbate one of our most pressing challenges – climate change.
Recent projections show that by 2030, AI could consume up to 21% of the world’s electricity supply. This staggering figure is brought into focus by the recent news that Microsoft is considering reactivating a decommissioned nuclear power plant to power a data center. Such developments underscore the urgent need to consider the environmental impact of AI, even as humanity leverages it to push the boundaries of science to solve problems and improve lives. We have to wonder, are we “solving” our way into a disaster?
As part of our work with the Computing Research Association’s Computing Community Consortium (CCC) Task Force on Sustainability and Climate Resilience, we are fortunate to be given a platform at the SC24 conference. We will use this opportunity to gather experts in science and industry to reflect on some of the more important questions that technology leaders should ask as they look to balance the benefits of AI with the threats of climate change. We will challenge the AI and high performance computing (HPC) communities to reflect on AI’s growth trajectory and its environmental implications, and we will pose questions that must be at the forefront of discussions about the future of AI. Here, in preview, we present three of these questions for leaders everywhere to consider in hopes of sparking an industry-wide discussion.
How can AI continue to drive innovation while minimizing environmental harm?
By now, we can all agree that AI has enormous potential, but we must ask ourselves whether the growing environmental cost of scaling AI is undermining its benefits. Large language models like GPT or BERT require vast computational resources and, thus, enormous amounts of energy. A single query to ChatGPT consumes approximately ten times more energy than a traditional Google search. These operations are taking place in expansive data centers consuming increasing energy. As AI continues to evolve, its energy demands and associated carbon footprint threaten to undermine the benefits that it promises to deliver.
There is a tendency for individuals to throw up their hands and decide that problems such as these are too big for them to solve, but a grassroots effort could be a critical place to start. This looming peril calls for a concerted effort from the HPC community to prioritize energy efficiency in AI system design down to the code, where energy-efficient hardware architectures are combined with optimized algorithms that mitigate the impact of AI on carbon output. This means developing a workforce that understands the system-level implications of AI training, including power consumption, data movement, and their associated costs. Instead of focusing solely on advancing AI’s capabilities, it’s important to prioritize these AI systems’ efficiency to achieve a sustainable balance.
The communities engaged in developing AI must be cognizant that while innovation is the aim, achieving balance is critical. They are responsible for ensuring that future AI technologies not only solve complex problems but also do so with minimal environmental impact.
What research gaps must be filled to ensure AI development aligns with sustainability objectives?
While we have some understanding of specific instances of AI energy consumption (i.e., ChatGPT queries), our broader understanding of AI’s overall carbon footprint remains limited. We lack comprehensive knowledge of performance versus efficiency trade-offs across different AI systems. AI’s growing environmental burden could offset its promised benefits without a concerted effort to address these gaps.
Many areas require deeper investigation, including additional research into energy-efficient algorithms and software layers and explorations into alternative architectures, such as neuromorphic computing and quantum computing, as accelerators for energy efficiency. New sustainability practices must be explored, including creating metrics and benchmarks for energy efficiency to measure the impact of these innovations.
We also need to develop educational programs that prepare professionals early in their careers with a foundational understanding of AI’s system-level impacts and implications. This is especially necessary in energy-efficient system design, where optimizing data movement and minimizing power consumption are critical to sustainable AI development. By creating a culture of computing sustainability, we can prepare future generations of AI researchers to address these challenges from the outset.
A comprehensive approach is essential. We must cultivate a workforce that can drive responsible innovation, balancing technological progress with environmental stewardship. Now is the time to lay the foundation for a future where AI’s potential is fully realized without sacrificing the health of our planet.
How can collaboration between technologists and environmental scientists lead to breakthroughs in sustainable AI practices?
Technologists alone cannot develop sustainable AI solutions. The challenges are too complex to be solved by any single discipline. We need robust collaborations that bring together expertise from technologists, environmental scientists, ethicists, and other fields to address these issues.
Our upcoming panel at SC24 aims to be a starting point for this collaborative approach. We have assembled a “dream team” of experts with domain expertise to bring unique perspectives to the discussion and help confront these complex issues. Through this approach, we hope to identify new pathways for reducing AI’s environmental impact while still pushing the boundaries of innovation.
Our message is clear: collaboration is essential to developing next-generation solutions that strike a balance, ensuring we don’t outpace the benefits of innovation at the cost of irreversible climate damage. By leveraging the insights of leaders from different domains, including accelerated computing architectures, advanced cooling technologies, renewable energy integration, improved data center design, and even policy and governance, we can develop more efficient and environmentally responsible AI systems.
This dialogue with diverse perspectives can be a valuable catalyst for change. We encourage the HPC and AI communities to actively engage with experts from adjacent domains and disciplines to identify areas where sustainable AI practices can be co-developed. Through these partnerships, new doors will open to tackle these complicated challenges, and innovations will impact society while protecting the environment.
The Path Forward
The future of AI demands that sustainability be a priority rather than an afterthought. These are not just technical challenges but moral imperatives that require immediate attention. We encourage our colleagues to engage in these critical conversations, participate in relevant forums, and join us in Atlanta this November to ensure these discussions take root in the broader community.
AI’s impacts will span across many disciplines, from public health to agriculture, making the quest for sustainable AI a technical challenge and a societal necessity. By working together, we can ensure that AI’s transformative potential is realized in a way that respects and preserves our planet’s resources. Let us rise to this challenge and shape a future where technological advancement and environmental stewardship walk into the future hand-in-hand.
About the Authors
Dr. Michela Taufer is the Dongarra professor at the University of Tennessee Knoxville and Vice-Chair of ACM SIGHPC, leading research in scientific applications on heterogeneous platforms and AI for cyberinfrastructures.
Dr. Chandra Krintz is a professor at the University of California, Santa Barbara, and co-founder of AppScale Systems, Inc., focusing on the intersection of IoT, cloud computing, and data analytics with applications in farming, ranching, and ecology.
