AI enlarging its carbon footprint
The artificial intelligence revolution is heating up. So is the climate. Unfortunately, AI plays a role in climate change.
Since at least 2019, researchers have been warning that the training and use of artificial intelligence models have a severe environmental impact, requiring large quantities of energy. As an article in MIT Technology Review put it, back then machine learning already had “a terrible carbon footprint.”
And that was before companies deployed dozens of chatbots and other generative tools, prompting millions of people to engage extensively with AI that produces text, images or sounds.
The water consumption used to cool data centers during the training and running of AI models is also huge.
Recently, a research team found that ChatGPT uses the equivalent of a 500milliliter bottle of water for each “conversation” of 20-50 prompts and replies.
“Given ChatGPT’s huge user base,” the researchers noted, “the total water footprint for inference can be enormous.”
Add to that, of course, the user bases of all the other successful artificial intelligence tools that have come along in just the past few months, with no sense that the pace of deployment is about to slow.
In fact, chatbots and other tools that incorporate machine learning are now being rapidly integrated into many areas of our lives.
Whether their benefits outweigh related harms depends on the context and the purpose of each product.
But in assessing their usefulness, as well as their risks, we all need to consider their environmental cost, too.
Unfortunately, most environmentally conscious consumers and lawmakers are not focused on this particular aspect of AI.
Even discussions specifically devoted to ethical concerns related to artificial intelligence (highlighting issues like bias, privacy violations, indiscriminate scraping to build training data sets, etc.) often fail to include environmental issues.
When the White House announced last month that seven of the key companies developing AI had agreed to a list of voluntary commitments focused on protecting Americans’ “rights and safety,” the announcement said nothing about sustainability.
It did, however, mention “transparent development of AI technology,” as well as “broader societal effects.”
Perhaps within those general terms we can locate some additional commitments to be required beyond those spelled out in the initial fact sheet: that the companies commit to disclosing the energy and water consumption involved in the development and deployment of their products, that they commit to educating the broader public and their direct clients about the environmental costs and that they commit to large investments in research focused on what some are calling “green AI,” which stresses energy efficiency as one of the criteria in evaluating models.
The National Artificial Intelligence Research and Development Strategic Plan update released by the White House in May did include a paragraph titled “Embracing Sustainable AI and Computing Systems,” which referenced the growing environmental impact of AI, but that level of attention is grossly insufficient.
Strangely, too, that paragraph was not part of the section titled “Understand and Address the Ethical, Legal, and Societal Implications of AI” — as if environmental impact is not, itself, an ethical (and societal) issue.
In the meantime, Phoenix just experienced the hottest month of any U.S. city, and dozens of other cities are set to break their own records.
Fires are burning in many states. The heat is depleting sources of water, too.
And, of course, climate change doesn’t impact only Americans.
The development and deployment of AI, it turns out, affects the rights and safety of people all around the world — and the impacts are not equitably distributed.
Researchers have shown that AI’s environmental footprint varies from region to region and have called for “equity-aware geographical load balancing to explicitly address AI’s environmental impacts on the most disadvantaged regions.”
In our particular part of the world, those driving the artificial intelligence revolution have a greater duty to address that impact.
Many Silicon Valley technology companies tout their efforts in developing “responsible AI.”
They also present themselves as incubators of innovation.
Given their outsize power and profits, these companies must play a commensurate role in addressing this important challenge.
At a minimum, the public can insist that they not make things worse by encouraging the integration of these — as it turns out — environmentally costly tools into everything. One of the questions of the day should be, “Do you really need a chatbot interface for that?”
The images we see each day in the news, of forest fires and of people struggling to survive heat waves, might not be rendered with generative artificial intelligence — but it turns out that AI will play an increasing role in generating them unless we address this issue at least as fast as new AI models come to market.