Dirty intelligence. No, AI will not solve the climate crisis

in August In 2019, a group of mourners gathered in the Icelandic hinterland. They included the country's Prime Minister Katrin Jakobsdottir, Environment Minister Gudmundur Inge Gudbrandsson, and former Irish President Mary Robinson. After short speeches, the funeral procession moved up the slope of the long-extinct volcano. A memorial plaque was uncovered between the ice and rocks:

The Öke is the first Icelandic glacier to lose its glacier status.
Within the next 200 years, all of our glaciers will likely follow this path.
This memorial is to emphasize that we know what is happening and what needs to be done.
Only you know if we do it.

August 2019 415 ppm carbon dioxide

[Autorem treści tablicy jest islandzki pisarz Andri Snær Magnason. Zmianę statusu lodowca wyraża również zmiana jego nazwy w języku islandzkim, z której usunięto cząstkę jökull oznaczającą „lodowiec”, wprowadzając nazwę Ok].

The effects of climate catastrophe in the Far North are painfully real. They are everywhere else Dangerous to life. One of the key elements in the fight against climate change is energy conversion – abandoning the burning of hydrocarbons (in the form of coal, gas or oil and their derivatives), including: by switching to renewable sources.

It's not just about building new wind farms or installing solar panels –

In the context of climate change, launching new sources of “green” energy is only a means to this end.

Data on the percentage of energy generated, which shows the increasing share of renewable sources in energy production, can be misleading. Share of coal and gas power plants in electricity production in Poland It's down about 70 percent. in 2021 to approximately 52 percent today. It seems great! However, the capacity of gas and coal power plants not only decreased at the same time, but actually increased slightly: from about 31.3 GW to about 31.6 GW.

If the only effect of investing in renewables is to increase overall energy consumption, we will not stop climate catastrophe. It is needed Radical reduction The amount of carbon dioxide being pumped into the atmosphere – and therefore a necessity Shutdown Power plants that use fossil fuels.

This means that we must manage the available energy budget wisely. And here we come to “artificial intelligence”, or so-called artificial intelligence.

If you believe the organizations promoting it, AI is indispensable in the fight against climate change.

For example, Forbes text, “How to combat climate change using artificial intelligence” (“How to Fight Climate Change with AI”) from 2 years ago addresses the “opportunities” related to the potential of “AI,” focusing primarily on “improving efficiency.”

The question posed in the title of the Forbes article is an important one: not “if” AI can help in this fight, but “how.” Clearly, the goal is not to find the most effective tools to combat the climate crisis. The same goal seems to be… Investing in AI: “The impacts of climate change are becoming more widespread and severe, so it is extremely important to invest in and support companies that use AI to find solutions,” the author concludes. From this article.

Forbes and the World Economic Forum are two large, well-known and respected organizations. However, both texts present the role of AI in combating climate catastrophe in a one-sided, unambiguously positive and uncritical way.

None of these texts (and many others) mention a single word about how huge the energy and water consumption is in the case of systems based on artificial intelligence algorithms! And what a small percentage of these resources consumed by AI systems go to solutions that could (even in theory) help reduce the effects of climate change.

This is typical AI nonsense, and I've already written about it in Oko. Technology is presented as inevitable, necessary, and almost magical. The message overstates the hypothetical positive effects of their use, while ignoring the real consequences that actually affect us today. The result is an inflating investment bubble.

“Artificial Intelligence” is essentially a marketing term, intentionally vague and ambiguous. That might mean small, specialized machine learning models, for example, those that have been used for years to recognize license plates in city surveillance recordings. But it can also mean massive language models that require massive infrastructure, gigawatts of electricity, and hectoliters of water for cooling (like the one the ChatGPT tool relies on).

Can artificial intelligence help combat climate change? certainly! But that's like asking whether internal combustion vehicles, for example, could help. Of course, there will be specific areas where trucks are very useful – for example, you need to transport elements of a new wind turbine in some way.

However, this does not mean that in the context of climate change “it is very important to invest in and support companies that use combustion vehicles.” This does not mean that every type of combustion vehicle is absolutely indispensable in the fight against climate catastrophe. And the question “how to combat climate change using combustion compounds” seems completely ridiculous.

“There are many areas where machine learning can be useful. “But at least in the short term, it looks less like magic, more like statistics,” explains Smari McCarthy, a former member of Alingi (the Icelandic Parliament), who hosts Today Climate change modeling company.

“Most of the data we use is generated on the basis of physical simulation models, which are created to accurately and verifiably represent the physical processes to which our planet is exposed,” he points out. So, not any “AI”, just good old mathematical models.

“It's important to stress that what people call AI is basically a very good system for detecting patterns. There's a lot of it in the climate,” McCarthy continues. “However, it's hard to get around this because global weather and climate systems are messy.” Essentially, it is sensitive to a large number of variables that are difficult to understand. So AI-based approaches are not very effective yet, except in very specific areas.

And it is appreciatedNearly 50 GWh of electricity was used when training the GPT-4 model (on which ChatGPT is based). This is equivalent to all the electricity generated from all energy sources (dirty and renewable energy) in Poland for about an hour! However, this does not take into account the energy used to collect the terabytes of data that the model was trained on.

In addition, there is the energy used to produce and transport thousands of dedicated servers, and the non-stop energy used to run the entire system after it is released and available to users.

Sam Altman, President of OpenAI He said recentlyThat further development of the so-called artificial intelligence will require a “breakthrough” in energy production. Bill Gates, who has been building an aura of environmentalism around himself for many years and at the same time tirelessly promoting large language models, admitted that the energy consumption of so-called artificial intelligence systems is “unbelievable“.

With the spread of so-called artificial intelligence, the energy consumption required to use popular applications and services is increasing. In addition, larger data centers increasingly consume huge amounts of water,” says Dr. Zuza Warsaw, Research Director at Open Future Foundation.

Water is often overlooked, but it is just as important as energy consumption. This is very evident in Latin America, where data centers are popping up like mushrooms after… well. The Mexican state of Queretaro is absolutely at risk of drought. However, the ten data centers currently operating there will soon be joined by dozens more. Meanwhile, local farmers' crops dry up.

“We must think about how to direct the development of artificial intelligence in such a way that this technology, which is very expensive in many respects, actually benefits humanity and the planet, and the costs we incur are proportional to the profits,” Dr. Hans concludes. Warsaw.

The costs are increasingly real and measurable: generating a single image using some generative model It consumes a similar amount of electricitywhich fully charges the smartphone. According to some researchers By 2027, the same AI infrastructure will annually use an amount of water several times higher than the annual water consumption of the whole of Denmark. The planned use of coal-fired power plants is being expanded, For example in Kansaswhere Meta is building a data center dedicated to artificial intelligence.

The gains – at least when it comes to combating climate catastrophe – remain largely hypothetical.

One might think that awareness of the massive consumption of energy and water by AI systems would prompt Sam Altman or Bill Gates to issue a plea to manage these limited resources more rationally. For example, by suspending work on large, more resource-intensive language models and focusing on smaller, more specialized models – something that has been done for years They call him Researchers like Dr. Tamnit Gebru. However, it is in vain to search for such an appeal.

We know what's happening and we know what to do. Artificial Intelligence systems are neither the only nor the best tool at our disposal. But the money must be correct.

So, I'll end with a quote usually attributed to Kurt Vonnegut: “We will go down in history as the first society that decided not to save itself because it was not profitable.”

March 2024, 421.79 ppm carbon dioxide.

turn “Saturday will tell you the truth.” This is the OKO.press offer for the first day of the weekend. Here you will find a check of facts (using the OKO lie scale) both from Polish politics and from around the world, because it is not only in our country that politicians lie, fabricate and fabricate. We will also return to the past, because lies have accompanied all of history. We disarm popular myths and delusions circulating on the Internet and in people's minds. We write about the cognitive errors that make us defenseless against lies. But are we really there? We will look into this as well.