How do neural networks affect the climate and the environment?

We live in an amazing time.Pandemic, war and the climate crisis go hand in hand with scientific and technological progress. Thus, intelligent computer systems understand language, reason and are constantly learning - in a word, they perform actions that are originally characteristic of a person. It turns out that the world has entered the era of neural networks, because only the lazy have not heard about the achievements of ChatGPT and Bing. But while chatbots generate texts and images, few people think about their impact on the environment, which, meanwhile, is huge, because millions of powerful computers are used to quickly perform operations. According to experts, today neural networks consume from 1% to 2% of global energy consumption, and computing power is doubling every four months. So what is the threat?

The development of high technologies in the midst of a climate crisis is fraught with disaster. Frame from the series "Extrapolations"

The development of artificial intelligence can provideenormous impact on the environment. Thus, the amount of emissions required to train AI models is equivalent to the emissions of gasoline cars over the entire life of their operation.

Content

• 1 Human mind vs artificial intelligence
• 2 Energy consumption of the computer industry
  • 2.1 Impact of neural networks on the environment
• 3 Climate and artificial intelligence
• 4 Crisis-2023
• 5 Technology and climate in the 21st century

Human mind vs artificial intelligence

ChatGPT by OpenAI, Bard by Google and Sydney byMicrosoft is a real marvel of machine learning. These systems take in huge amounts of data, look for patterns in it, and generate statistically probable results that increasingly resemble human behavior. Our mind, however, is not like ChatGPT or other similar systems - their statistical mechanism for processing and comparing data is not comparable to the capabilities of the brain.

Moreover, modern models cannot distinguishthe possible from the impossible and do not always give the right answers: most often, neural networks report a correlation between a set of received data and do not take into account causal relationships. Linguist and philosopher Noam Chomsky believes that AI responses are “nothing more than a probability that changes over time.”

Noam Chomsky is an American linguist, political essayist, philosopher and theorist. Professor of Linguistics at the Massachusetts Institute of Technology

More on the topic: The "dark side" of chatbots: from declarations of love to conversations with the dead

The human mind, on the other hand, isan efficient system that operates on small amounts of information and provides detailed explanations. And although we all make mistakes from time to time and are prone to cognitive distortions (errors of thinking), it is the mistakes and their correction that make our thinking unique. This is why chatbot responses are superficial and, as Chomsky writes in his New York Times essay, likely to remain so.

Power consumption of the computer industry

Chatbots work thanks to millions of powerfulcomputers located in data centers. According to experts, the energy consumption of such centers is from 1% to 2% of the global figure and the figures will continue to grow. So, any generative artificial intelligence (AI) like ChatGPT requires huge computing power.

As the big tech companies(Microsoft and Google) are integrating chatbots into search engines, text editors and email, our interaction with programs is changing dramatically. Until now, we have entered a number of commands to perform certain tasks, but now we literally “talk to the machine”, delegating to it the tasks that we previously performed on our own.

AI, although it doesn't seem like it, is physically shaping the world, says Keith Crawford, director of the AI ​​Now Institute at Microsoft Research.

Modern AI, fortunately, does not look like Skynet from the Terminator. At least for now

As chatbots become a tooldaily use, the demand for AI systems is increasing, and with it, energy consumption. But what impact could these changes have on the environment? No one knows the exact answer to this question, but "global costs" are constantly growing.

Did you know that the Lensa neural network draws portraits from photographs? Try it right now!

The impact of neural networks on the environment

As experts previously calculated, the technology sector in 2040 will account for 14% of global emissions into the atmosphere. However, according to other forecasts, the growth in energy consumption by the IT industry will increase by 15 times by 2030. These estimates are probably not justified, since they appeared long before the release of ChatGPT.

According to the Our World in Data project, in 2022global aviation accounted for 1.9% of greenhouse gas emissions, 2.5% of CO2 emissions and 3.5% of radiative forcing. The carbon footprint of the computer industry has already reached aviation levels.

Chatbots, silently and imperceptibly, add problems to an already difficult climate situation

Today Google and Microsoft have hundredsmillion users. Imagine what would happen if they all started using chatbots. Martin Bouchard, co-founder of Qscale's datacenters, estimates that each search query will require at least four to five times the processing power of today.

Corporations themselves are slow to share forecasts, and their climate policy is likely to be environmentally friendly (for example, by planting trees), says Bouchard.

The problem is modern AI produces more emissionsthan conventional search engines.All because of the complex architecture of neural networks created by analogy with the human brain - there are millions of parameters that need to be trained in the neural network. And this, of course, requires a lot of energy.

See also: Can a neural network replace artists, writers and programmers?

Climate and artificial intelligence

As El País columnist Manuel Pascual writes,training a natural language processing model is equivalent to the emissions of five gasoline cars over their lifetime, including the manufacturing process, or 125 round trips between Beijing and New York.

The results of a 2021 study published in the journal Nature showed that consumption of water resources for cooling systems Google Corporation amounted to 15.8 billion litersper year, and Microsoft - 3.6 billion liters, respectively. At the same time, the use of rare metals for the manufacture of electronic components makes AI more efficient.

Electricity consumption in January 2023 by OpenAI, the developer of ChatGPT, is equivalent to the annual consumption of 175,000 Danish families

According to OpenAI, training the GPT3 model on which ChatGPT is based, costs 500 tons of carbon, which is equivalent to flying to the moon in a car back and forth. The numbers are relatively small, but the models are regularly retrained and new ones are created, for example, GPT4, which we talked about earlier.

Note also that, unlike mostlarge technology companies, OpenAI provide data on electricity consumption. Thus, the maintenance of a chatbot in the first month of 2023 is equivalent to the annual consumption of 175,000 Danish families.

Carbon dioxide emissions from useAI systems can be divided into three factors: the power of the equipment used, its carbon intensity and the energy used during the entire time the model is trained, writes Alex Hernandez, professor at the Institute of Artificial Intelligence of Quebec (MILA).

The climate on the planet is rapidly changing, putting the future of all mankind at risk

To train AI models, most data centers use advanced processors, and the necessary computing power doubling every three to four months. Back in 2018, the results of an internal OpenAI study showed that smart systems need much more energy and should be prepared for this moment.

Read even more interesting articles about scientific discoveries in the field of high technologies on our channel in Yandex.Zen - articles that are not on the site are regularly published there!

Crisis 2023

At the end of March, the media reported on an open letterwhich was signed by more than 1000 technical specialists, including Elon Musk and Steve Woznik, demanding that the training of AIs more powerful than GPT-4 be suspended for at least six months. According to the signatories, at this stage, neural networks are beyond the control of even their creators, and therefore the regulation of AI should be dealt with, including in the government.

And this is just the tip of the iceberg. As a previously published study by the investment bank Goldman Sachs showed, in the near future about 300 million people could lose their jobs due to AI.

In the future, forest fires will become more intense and prolonged.

Don't Miss: What happens when Artificial Intelligence reaches its peak?

At the same time, the climatic situation inworld, according to the Intergovernmental Panel on Climate Change (IPCC), is rapidly deteriorating. “A climate bomb is about to explode,” write the authors of the sixth IPCC report. At the same time, the actions of world leaders to reduce greenhouse gas emissions are insufficient, and climate change is already affecting human life and well-being: over the past decade, mortality from floods, droughts and storms has increased 15 times.

Human influence has warmed the atmosphere, ocean and land, endangering life and well-being at least 3.6 billion inhabitants of the Earth. Today's global temperatureexceeded pre-industrial levels by 1.1°C. And although a tenth of a degree may seem insignificant, this change is of great importance for the planet, the authors of the report write.

The price of technology development is too high

Combining all the available data, you can geta depressing picture: the scale of extraction and combustion of fossil fuels does not change, and further development of AI will increase the amount of emissions. Moreover, if smart computer systems put hundreds of millions of people out of work, the economic situation (which is already unstable) could turn into another Great Depression on a planetary scale.

Read also: What artificial intelligence is better than humans in 2023

Technology and climate in the 21st century

As temperatures rise, weather events willare becoming more extreme, and the “point of no return” outlined in the Paris Agreement may soon be passed. Recall that the agreement, signed by world leaders in 2015, aims to limit the "increase in the average global temperature above 2 ° C compared to pre-industrial levels." The IPCC pays special attention to measures to mitigate the effects of climate change, emphasizing the exchange of experience and technologies.

The current decade is decisive forof humanity – major industries need to change, including energy and food production. As the American comedian Josh Carley said in one of his speeches, in the end everything will be fine with the planet, but people will be finished.

Extrapolations is an American anthology drama television series created by Scott Burns for Apple TV+.

You can read more about what the world will be like by 2070 here, don't miss it!

Alas, but inaction in relation to the fight againstclimate change turns us into characters in the sci-fi series Extrapolations, whose characters survive in a high-tech world engulfed by floods, landslides and fires. And if we continue in the same spirit, creating more and more powerful AI and technologies, we will have to pay for them at the cost of our own lives and well-being. So is the game worth the candle, do you think? We will wait for the answer here, as well as in the comments to this article.