Here at UC Berkeley, using ChatGPT is nearly interchangeable with a Google search. Ranging anywhere from debugging coding assignments to writing our discussion posts, there is little we assume ChatGPT won’t do to further our understanding of a concept, or help us make a deadline. Given our increasing reliance on AI for completing tasks, many have expressed some apprehension towards the extent of its integration into our daily lives. Some are concerned with the privacy of our information, some with modes of data collection, and our professors are certainly concerned for our work ethic. These questions are raised more frequently due to ChatGPT’s possible impacts on the future of internet use, the workplace, and the broader fabric of human interaction.  Even so,  the relationship between ChatGPT and humanity is only one dimension of its potential influence—what about the rest of the world? An overlooked yet highly consequential variable in the creation of large language models (LLM) like ChatGPT is the massive amount of energy it uses, and the subsequent destruction of our environment.

The Problem

 ChatGPT is a form of Machine Learning AI , which signifies its ability to make predictions in response to language prompts based on a massive dataset of human-produced data, including our conversations.³ Because we are so removed from the realities of yielding and storing this amount of data, it’s difficult to imagine how exactly this negatively impacts the environment. Yet, in training a neural network, a component of machine learning models that determines how the algorithm makes its predictions, the equivalent of around 300 kg of carbon emissions are emitted—this is five times more than the carbon emitted by average cars in their lifetime.¹ This is because heat is the “waste product of computation,” and data centers must be cooled for the successful operation of any data stored on the Cloud.² Even though some environmentally-friendly ideas have been proposed to combat this problem, such as moving data centers to Nordic countries for their naturally cooler climate, no solution has been able to take on the demand posed by our current data needs.² 

Where to Place Blame 

As extreme of an issue as this is, is it fair to place blame on UC Berkeley students? Why is it that we so recklessly use a resource that has a greater carbon footprint than the airline industry, without a second thought?² It’s possible that the carelessness with which we regard ChatGPT is due to how data and AI are framed in our minds. This is the subject of Tim Hwang and Karen Levy January’s work, “‘The Cloud’ and Other Dangerous Metaphors,” which discusses how the terminology used to describe data changes subtly sways our approach to it. Data is so often compared to natural resources, as seen in terms like “data stream,” “data mining,” and “the Cloud.”⁴ Entrenching data in environmentally-oriented vocabulary not only enforces the assumption that collecting data is a natural and bountiful process, but we prime data to be exploited in a manner analogous to the desolation of our environment.⁴ With this in mind, changing how we interpret the concept of data may affect how we use it, and could subtly reduce our contribution to the ongoing climate crisis. 

But is it enough? Is it reasonable to expect UC Berkeley students to reduce or cease the use of a resource they increasingly rely on? While it may be nowhere near enough of an effort, it’s possible that minimizing ChatGPT’s consumer base is a practical measure we can take to minimize the climate crisis, and changing how we think about ChatGPT may be the first step. 

References

1. Coeckelbergh, M. (2021). AI for climate: Freedom, justice, and other ethical and political challenges. AI and Ethics, 1(1), 67–72. https://doi.org/10.1007/s43681-020-00007-2
2. Monserrate, S. G. (2022). The Cloud Is Material: On the Environmental Impacts of Computation and Data Storage. MIT Case Studies in Social and Ethical Responsibilities of Computing, Winter 2022. https://doi.org/10.21428/2c646de5.031d4553
3. Singh, S. K., Kumar, S., & Mehra, P. S. (2023). Chat GPT & Google Bard AI: A Review. 2023 International Conference on IoT, Communication and Automation Technology (ICICAT), 1–6. https://doi.org/10.1109/ICICAT57735.2023.10263706
4. “The Cloud” and Other Dangerous Metaphors—The Atlantic. (n.d.). Retrieved October 16, 2024, from https://www.theatlantic.com/technology/archive/2015/01/the-cloud-and-other-dangerous-metaphors/384518/

Image Reference 

Banner Image: Photo by Sergei Starostin on Pexels.. Pexels. Retrieved October 16, 2024, from https://www.pexels.com/photo/network-servers-on-an-enclosure-6466141/