Contrary to popular belief, new research finds that the use of artificial intelligence has a minimal effect on global greenhouse gas emissions and may actually benefit the environment and the economy.
For their study, researchers from the University of Waterloo and the Georgia Institute of Technology combined data on the U.S. economy with estimates of AI use across industries to determine the environmental fallout if AI use continues its current trajectory.
According to the U.S. Energy Information Administration, 83 percent of the U.S. economy is powered by petroleum, coal and natural gas, all of which contribute to climate change when burned. The study authors found that while power usage from AI in the U.S. equaled the energy consumption for all of Iceland, the amounts were not noticeable on a global or national scale.
Here is an exclusive Tech Briefs interview, edited for length and clarity, with Dr. Juan Moreno-Cruz, Professor, Faculty of Environment at Waterloo and Canada Research Chair in Energy Transitions.
Tech Briefs: What was the biggest technical challenge you faced while gathering the data?
Moreno-Cruz: At the level we did the analysis, data was mostly available. We had data on the input-output tables, which is the way economists collect information on how industries interact with each other. Then there was some data collected on occupations and tasks. The way we organized the data was there’s a series of tasks that you perform on a daily job, and you combine tasks for your occupation. Some industries have occupations that are required. So, all that data existed.
To talk about AI, then you have to make assumptions about how a particular task is going to be affected by AI. That data came from a science paper that we cited.
They did a series of experiments that where they asked people which of their tasks could be done by AI. For example, if you say, ‘Well, my calendar can be done by AI and I spend five percent of my time dealing with my calendar and schedule,’ then you'd say, ‘OK, so that is five percent of your time that would be saved by AI. So that got aggregated.
And then these economists made a couple of assumptions based on that data that we also borrowed from. One of the assumptions is, ‘OK, so suppose that we said that 70 percent of your tasks can be done by AI, then how much of that 70 percent will really be implemented?’ Because if you said that you can use AI to do your schedule, it doesn't mean that you're going to do it all the time. And the other assumption is about how effective the AI was. So, at the end, the data wasn't too complicated, but there were a lot of assumptions built into the way we collected it.
Having said that, the data that we need right now to do the next step is data tracing actual processes and seeing what their energy use is. Once we had all the economic impact, we said, ‘OK, this particular industry is going to be affected by AI to some extent. Let's say 20 percent.’ Now we take how much is the energy use per industry, and then we multiply those numbers. So, it still is quite aggregate, and there is a lot of uncertainty in the way that we collected that data and the assumptions that we made.
Tech Briefs: Do you have any set plans or further research work, etc.?
Moreno-Cruz: We’re working on two levels. One is the international aspect, because the energy system and the economic structure is different across countries. Then that means that the impacts of AI are going to be different across countries.
We are lucky because there is a dataset that kind of replicates the data that we used for the U.S., but does it for different countries, including Canada and European countries. It's not at the same level of disaggregation, but we can work with that. So, we're working on developing those numbers.
The other area in which we're working is actually the complete opposite — trying to look into a more local level, because we think that's really where the issues are going to show up with AI and resource use.
It's not like what we calculate that is a national or global amount, which is not that big relative to the size of the economy or the size of the energy system. But locally, it's going to be felt more — if you put in a steel plant or something like that in your community, suddenly there is a lot more electricity you need.
Tech Briefs: What would you say to people who are worried about AI's energy use and its effect?
Moreno-Cruz: I will say that if the worry is about climate change, there is no real basis for that worry. The contributions of AI and data centers to climate change are really, really tiny. And they will continue to be tiny, although I'm pretty sure they're going to increase as more models come in and you do training more often. Because training is the biggest part of the energy use.
Then when people start to use it in more tasks, I'm pretty sure it's going to increase. But the possibility of that getting to like 5-10 percent of the energy system, which will be a lot, but I don't think it’s that likely. So in terms of climate change, the impact I don't think is that big.
The impact at the local level is big. Data center use will affect the energy and electricity prices at the local level, and it's already doing it.
And not only that, when you implement or build a data center of the current magnitudes, then you have to redispatch electricity plants unless you come up with your own plant, which is what big companies are doing now. But in that process, there are also changes in air quality because you have different plants being dispatched at different levels. So, the increase in electricity not only increases prices, but also increases emissions associated to that electricity. If we don't have electric plants that are renewables or nuclear, then local communities are going to see an increase in air quality issues.
