President Carmen Twillie Ambar recently inaugurated the current academic year as the Year of AI Exploration. The administration will give students access to the enterprise version of ChatGPT and Gemini in the spring and will host AI-related lectures and workshops. While President Ambar’s plan acknowledges, “the concerns that AI raises,” it still assumes that AI can and should have a place at Oberlin. We don’t necessarily disagree with this assumption, but we know plenty of Obies who would, and such an assumption shouldn’t be left unexamined.
With this Year of AI Exploration now begun, we wanted to weigh in on the discourse about AI on campus. We want to encourage Obies to carefully consider the role we want AI to play in our lives and in our education and not allow any preconceptions about this technology to go unquestioned. As we’ll explain, there’s much students don’t understand about the impacts of AI, from its environmental effects to what it means for our cognitive abilities and our education.
Many people are skeptical of the widespread adoption of AI due to environmental concerns. It’s difficult to determine how much AI harms the environment, partially because AI companies are cagey about sharing such data. Different queries also use different amounts of energy: asking ChatGPT to tell a joke uses much less energy than asking Sora to generate a video.
An article in the MIT Technology Review calculates that asking a large language model one question consumes only about 0.3 watt-hours of energy — enough power to run a microwave for one second. But all of this power, multiplied by billions of daily prompts, adds up quickly: ChatGPT, as a whole, may use as much as 144 gigawatt-hours every year. In total, AI-related servers in American data centers may have used between 53 and 76 terawatt-hours of electricity last year. That’s enough to power over seven million U.S. homes for a year — or to power a microwave for more than ten million years. Right now, about half of that power comes from coal and gas combined. As with all statistics involving AI, we should anticipate these numbers getting bigger as AI companies continue to upscale their models.
“That has been the ethos of a lot of these companies, that we can just keep making things bigger and bigger, and that’s gonna get us closer to whatever goal they have,” Associate Professor of Computer Science Adam Eck said. “At the same time, there is a lot of active research about how to make these models more energy efficient.” There is hope that the industry will gradually switch over to renewable energy and find more power-efficient ways to improve performance.
Aside from global environmental impacts, AI can have a serious impact on communities that house data centers. According to the Associated Press, the AI industry is causing a spike in the price of electricity in these communities. It’s also straining the power grid: training large language models requires a variable amount of power, and many central processing units turning on or off all at once causes a dangerous power fluctuation. This is a problem worth addressing, even if AI isn’t the biggest contributor to climate change right now.
Ultimately, the evidence suggests that for most users, reducing AI use will not benefit the climate as much as, say, biking to work will. But the AI industry as a whole is nonetheless on track to seriously impact the climate, at least according to some estimates. Although industry practices may become more environmentally responsible, there’s still reason to be concerned about AI’s environmental impact.
For many, the impact of AI usage on how we think is also cause for concern. The worry is that by saving time and effort through delegating tasks to AI, we’re neglecting the maintenance of our own mental skills. AI is far from the first technology to inspire concerns about such “cognitive offloading.” In the fourth century BCE, Plato described how Socrates opposed the adoption of the written word, partially due to fears that it would weaken our natural capacity to remember things. It’s easy to dismiss this fear as misguided or even quaint. But should we dismiss fears about cognitive offloading to ChatGPT as similarly wrongheaded? Perhaps Socrates’ concerns were well-founded — people in pre-literate cultures were indeed less reliant on external aids for memory than we are today. We accepted a trade-off between our ability to remember things naturally and our ability to preserve information through the written word. In modern times, we face similar trade-offs when we outsource our ability to navigate to Google Maps or our ability to do long division to digital calculators. Even if most of us would agree that these trade-offs are “worth it” for the extra convenience, it’s fair to say that outsourcing our ability to reason to an LLM is different and more concerning than offloading our ability to do mental math to the calculator app on our phone.
As students, we should think about how offloading cognitive tasks might affect our brains and our learning. We are concerned about how easy it is to become reliant on AI. However, we do see positive aspects of AI in academia, regarding accessibility and advanced research.
“I think the important part is finding where the right use cases are,” Professor Eck said.
We agree. Thinking critically about AI means neither rejecting any technology with “AI” in the name nor allowing ourselves to become reliant on it. We call upon the Oberlin community to think deeply about what they do and don’t want AI to do for them, throughout the Year of AI Exploration and beyond.
