NLP’s Cadillac Desert

Water politics affects all of the work we do in natural language processing.

Let me explain.

One of the books that has most strongly shaped my worldview and understanding of American politics is the nonfiction book about water in the American West (and more broadly North America), Cadillac Desert, written by Marc Reiser. The book was first published in 1986 after decades of research on water politics, with extensive interviews and devastating lists that underscore the way the Bureau of Reclamation transformed the hydrogeology of the United States west of the Mississippi. The Bureau of Reclamation and its history feature prominently, with regular mention of the entitlement dam-builders felt to reshape American rivers.

These rivers were dammed on the auspices that they would pay for themselves by generating hydroelectric power, which would be sold to the public. In exchange, farmers would be able to grow improbable crops like alfalfa in some of the hottest, sunniest places on earth such as Yuma, Arizona. Small rivers like the Colorado and massive ones like the Columbia alike were dammed to meet electricity demand. Now, these dams constitute some of the greenest energy in the United States, generating electricity more or less entirely by leveraging gravity. But decades of policy decisions have put tech companies in places that can leverage the massive electrical potential of rivers like the Columbia, which provide a non-trivial proportion of the electricity that is used to power massive generative AI models like Gemini, ChatGPT, and BingChat. Each query from these systems takes so many milliliters of water and effectively boils it to power massive data warehouses; water that could be used to power electricity for other purposes, or be consumed directly. The massive growth in renewable energy in the last few years, it turns out has mostly been diverted to power GenAI systems and data warehouses. This is all in exchange for systems like Google’s newly unleashed, potentially life-threatening “information retrieval” system whose primary skill seems to be creating (as I saw it referred) an endless stream of Tide POD Challenges.

As a child of the desert, I feel a personal twinge of pain every time I think about these water statistics. It’s seared into my brain the image of a little blue droplet on roadside billboards encouraging us to save water out in West Texas. El Paso and Las Cruces receive about 8 or 9 inches of rain each year, squarely high desert climates. I remember very hot summers and limited water breaks at school, sometimes going inside during recess for water and, hit by air conditioning, briefly losing the ability to see and hear. The idea of throwing away bottles of water just to answer a question that isn’t even as accurate as Wikipedia turns my stomach.

But water politics isn’t just present in the models we run in natural language processing. It’s also a driving influence in the way we choose where to host our conferences, and the assumptions we make about resource availability in places we put scientist-tourists. As organizers of conferences, where we host partly reflects our willingness to avert our gaze from the (international) politics of conference destinations in exchange for our other priorities (fun location, visa opportunities, diversity).

Consider, for example, that the Association for Computational Linguistics has been holding its international conferences in places that are increasingly under anthropogenic climate change-induced water stress. At least two major venues demonstrate the field’s disregard for water.

First, let us examine the alfalfa-growers and the deep pockets of Saudi and Emirati money by considering the 2022 meeting of Empirical Methods in Natural Language Processing (EMNLP), which took place in Abu Dhabi. On purely scientific grounds, NLP needs the linguistic, cultural, and methodological diversity of the entire world to understand how to build systems that can can process and produce language. We are severely limited as a science if we do not not keep the field open to researchers who would otherwise not be able to participate. Few places can easily host folks from across the globe like Abu Dhabi — the United States and Canada make it basically impossible for researchers in China, India, Africa, and the Middle East to visit to present their research. This makes it an attractive conference destination.

But, Abu Dhabi receives basically no rain — not even two inches a year on average. In the summer, the average high air temperature is well above 100 degrees, at a relative humidity of 61%. According to a wet bulb calculator, Abu Dhabi is uninhabitable outdoors in August. But, the average relative humidity is high enough that not only is it possible to stay cool with air conditioning, but much of the public’s drinking water is obtained by collecting condensation from air conditioning systems, as long as one burns fossil fuels. But there is still not enough water. If you have ever flown over Las Vegas or Phoenix, you’ve seen giant green crop circles in the desert. It turns out that Emirati nations obtain alfalfa by exporting American water from one of the driest places in the United States, sucking it out of aquifers and emptying rivers. All of that feeds animals on the other side of the world, shipped with the speed and ease provided by fossil fuels. We buy oil, they buy water.

EMNLP took place in 2021 in Punta Cana, Dominican Republic, at a resort, for similar reasons to Abu Dhabi. The Dominican Republic is not trivial to fly to — it is also not a place where many researchers in NLP are attending from. Nearly every attendee came from across the globe to attend, requiring a minimum of several multi-hour flights. But, Punta Cana is gorgeous as a destination, and cheap at the tail end of hurricane season, making it attractive to scientist-tourists. The carbon cost of attending EMNLP in 2021 and 2022 makes you recoil at least a bit, right?

I’m watching the earth roast from impossible wet bulb temperatures in India and Bangladesh. Just a couple of years ago, Chennai ran out of water and it had to be delivered by train. Water from glaciers has dried up, gets absorbed into the ground before it makes it down the mountains to communities in Bolivia and Colombia, Argentina and Peru. My mind turns to news of Mexico’s severe drought and the continual fall of record high temperatures. Mexico is seeing temperatures as high as 50ºC/122ºF in May. Mexico City is projected to reach its Day Zero (when water levels will be too low in the aquifer to provide water for residents) just five days after NAACL 2024 ends.

NAACL is doing an excellent job as far as I can see of engaging and lifting up the Latin-American NLP community and has picked a much more welcoming destination for the world’s scientists. But there are consequences to the tourism of conference travel. How many of the scientist-tourists who are flying to Mexico City for NAACL are aware of the water crisis that is mounting? Heck, growing up in El Paso along the Mexico-Texas border, water was always a concern (and the way the Rio Grande has been leveraged as a tool of violence against immigrants to the United States is no coincidence), but our water troubles were restricted to boil orders when the aquifer ran too low, or when monsoons flooded the streets. And sometimes the water just isn’t safe to drink. But traveling somewhere without water never seemed like a plausible thing that could happen.

Day Zero is mostly unthinkable to tourists. I would bet that if you are a scientist-tourist going to NAACL, you probably feel like there should be enough bottled water and tap water at the conference venue. You probably expect lots of delightful Mexican coffee, pristine flush toilets. If you grew up outside the desert, you will probably let the faucet run the entire time you wash your hands, and run the shower until the water is scalding hot. The sunny skies in Ciudad Mexico will great you as a sign of a lovely summer after a scorching heat wave has killed untold numbers of residents of Mexico. You will probably not wonder why you have access to water and others do not. But you should. Water is a public resource, and everyone is entitled to safe, clean drinking water — not just people who can pay enough for it.

I’ll end this with a simple request. If you absolutely must go to NAACL, please conserve as much water as you can. Your decision to attend the conference in person during a water crisis directly impacts the lives of other people. Drink responsibly.