It’s cliche to say that our lives are "data driven." But I can’t think of a more apt description of our modern society.
Around the clock, we are constantly generating data. From smart watches tracking our sleep to the social media algorithms tailoring our digital experience, almost every action we take creates new electronic data, whether we realize it or not.
With the advent of cloud computing in the early 2000s, the amount of data at our disposal seemed limitless. Cloud computing meant that we were no longer constrained by the hard drives on our local devices and could instead tap into the power of data centers, large warehouses storing the IT infrastructure that allows the internet to function.
But nothing is ever really unlimited. Now, the rapid growth of AI is straining data centers and the power infrastructure needed to support them. The data center power demand is expected to double by 2030, with electricity generation unlikely to keep pace.
Our society has entered an AI arms race, a dangerous game that ignores its environmental and social destruction. Like any technology, AI is not inherently evil, but it is being developed in a manner that lacks foresight. As a result, computing power has become the currency of the future, even as we fail to find enough electricity to power servers or sufficient guardrails to mitigate AI’s side effects. AI needs to become a tool for progress, not a product for competition
The data centers, where AI workloads are processed, have significant environmental challenges. Packing so many servers into a tight space makes data centers very hot, which requires massive cooling systems to keep the servers operational. Large data centers require over 100 MW of power, the equivalent of 80,000 homes. Regarding water, water-based cooling systems use 208 million gallons of water a year, enough to fill a swimming pool that’s 10 miles long. The carbon emissions of both Microsoft and Google are heading in the wrong direction, with data centers being the primary culprit in both instances.
A hidden impact of data centers is the waste they generate. Maintaining cutting-edge technology is essential in this industry, which means that many servers rich in critical minerals such as lithium, cobalt and copper, are thrown away as soon as the next model comes out — resulting in a lifespan of 1-2 years. These minerals cannot be wasted as they are valuable inputs to many technologies needed for the energy transition, such as solar panels and electric vehicle batteries. This excessive waste is particularly common in crypto mining, where miners need the newest application-specific integration circuit (ASIC) to remain competitive.
Simply because AI has negative environmental impacts doesn’t mean that the technology can’t help our society. Because big tech companies have instituted ambitious sustainability targets, data centers have served as conduits for adding renewable energy to the electric grid. Data centers account for two-thirds of corporate renewable energy purchases, making the industry a major financier of renewable energy. Given the large energy requirements of data centers, it is essential that they are powered by carbon-free electricity sources such as solar, wind and nuclear instead of by fossil fuels. Additionally, AI can create new insights to better decarbonize industrial processes and promote energy efficiency.
Here at Duke, we are experiencing the benefits of AI and large-scale computing. The Duke Compute Cluster (DCC) is a network of computing resources that community members can use to perform computationally advanced tasks, such as reading DNA sequences and training machine learning algorithms. Without access to the DCC, Duke would not be the dynamic research institution that it is today.
So how do we balance the positive impacts of AI with its negative side effects?
I believe the answer lies in shifting AI development from an arm’s race to a collaborative process. Government agencies and big tech companies leading the industry see AI advancement as a race, where speed is prioritized over safety. The problem with this perspective is that AI development is viewed as a zero-sum game, with discrete winners and losers. Those that amass the most computing resources will gain superiority over everyone else and control the global economic system, or so the thinking goes. An arm’s race approach will result in a misallocation of AI resources towards solving esoteric mathematical equations instead of the world’s most pressing challenges, as we’ve seen with cryptocurrency. AI is only as impactful as the task it is assigned to solve.
Another impact of AI that is seldom considered is the information treadmill it creates. In other words, AI is needed to manage AI. Data centers are a perfect example of this. These facilities which are built to hold AI servers are operated by AI-powered software to manage the ever more complex server networking AI requires. AI generates more data than human could ever interpret unassisted. So, more AI is used to extract key insights. This trend also impacts human workflows. For example, many programmers have noticed that they now spend more time debugging code written by AI than they would if they had written the code de novo. In sum, AI creates the need for more AI, which results in a positive feedback loop that is difficult to break.
First, AI systems should be reconfigured to better match spatiotemporal variations in renewable energy. The big tech companies that dominate AI have global data center portfolios that are synced thanks to the advent of virtualization. Companies should redirect computing loads to data centers in regions that are powered by large amounts of renewable energy. This will optimize wind and solar resources while also lowering the carbon footprint associated with AI.
Second, consumers should rethink their digital footprint and take stock of the ways they maybe be inadvertently generating data. Rejecting cookies, limiting the use of large language models (LLMs) such as ChatGPT and deleting unused apps on your phone are all good places to start. Remember that everything you do online corresponds to server usage in a data center.
Finally, use the devices you have for as long as possible. Our electronic devices are full of critical minerals and require significant amounts of energy to manufacture. Upgrades on new devices are often barely noticeable, as the core components of an iPhone have barely changed since 2007. Avoid buying a new device if your old one is still in working order.
AI has made the fight against climate change even more challenging. AI is a double-edged sword, producing more emissions while helping to reduce emissions throughout the economy. Responsible AI development is needed to mitigate the environmental impacts of computing. Even if they seem negligible to us, our choices as consumers do make a difference.
Aaron Siegle is a Trinity junior. His pieces typically run on alternate Fridays.
Get The Chronicle straight to your inbox
Signup for our weekly newsletter. Cancel at any time.