Artificial intelligence (AI) is increasingly in our lives, but we find ourselves extremely “hungry” for water and energy. Elitsa Staneva-Britton spoke about this topic before the facts. She is an educational psychologist, earning her bachelor's degree from Sofia University's St. Kliment Ohridski and a master's degree from Ludwig Maximilian University in Munich. Her scientific interest lies in the field of ecology and relates to the effects of climate change on human psychology and the concept of climate justice.
-Britton, when will the “dark side” of artificial intelligence involve us?
– Before we talk about the topic, let's note that artificial intelligence (AI) has great potential to create innovation in the bright side, namely ecology, education, medicine, engineering and finance.
The “dark side” issue of AI comes from not from the technology itself, but from how it is used. In many cases, AI companies focus on profits at a purely organized level, such as automatically deploying applications and saving costs by selecting employees.
When using AI separately, its use is usually reduced to a simple solution. To create content, it often involves unclear reliability, optimizing the process, and even reading a book becomes a short summary. That's precisely because of the frequent use of AI for things that aren't that important, increasing resource consumption and revealing the “dark side” of AI in the long run.
-Why does ChatGpt search use about 10 times more energy than regular searches?
– Traditional Internet search relies on retrieving existing information published online, but the generated intelligence is based on a huge amount of data, through which it is trained and then generates new content.
It is precisely because of this generation of new content that the entire AI artificial neural network is activated and energy consumption is greatly increased.
-The energy consumption of ChatGpt in one day is approximately 564 MWh or the average amount of energy used by a household in a month. How will this consumption grow?
– This depends on how AI is used and whether the company complies with regulations entering the sector.
For now, it is interesting that data on the issue remains hidden from most tech giants. For now, according to the embracing face platform, the Bloom model consumes 433 MWH to train the database, while other models, such as the GPT-3, Gopher, and Open Pre-Traseed Transformer (OPT), train terabytes of data using 1287, 1066, and 324 MWH, respectively. In addition to the training model, energy is also spent on servers and chips, with the giant Nvidia reporting profits of $13.5 billion in the second quarter of 2023 alone. Experts suggest that AI could become more environmentally sustainable. This can be achieved through the right chips that prioritize energy efficiency over the computing power of the AI itself. Such an action will optimize energy consumption 100-1000 times. Another step is developing technologies that reduce energy and water consumption (particularly for cooling) in data centers. As early as 2023, the Institute of Massachusetts Institute of Technology's Lincoln Supercomputer Center is developing technologies to reduce energy consumption.
– By 2026, the AI industry will consume more than 10 times more energy than 2023.
– Some experts say companies are fighting the Cold War over AI. Economic benefits are stronger than social and environmental risks. Meta, Microsoft, Apple and Google have an unabated plan to invest billions in AI infrastructure.
At the individual level, it appears that they forget that behind digital terms such as the cloud space of AI infrastructure, there are energy-consuming physical devices. And we continue to use more and more AI.
– It takes about 8,300 liters of water to make one chip and about ½ liters to write an email from an AI bot. What does this tell us?
– Yes, water consumption of AI models construction, training and use is also a serious waste of resources and risk. Additionally, data centers consume water for cooling.
In a world where climate change is already limiting the availability of freshwater, this data suggests that resource-hungry AI is putting even more pressure on already declining water supplies.
– Bulgaria is one of six European countries where AI factories are developed. The Brain++ project is a great economic opportunity, but also poses challenges in terms of energy and water consumption. And we only have Kozloduy NPP…
– That's right. Bulgaria is also one of six European countries where a new AI factory is soon to be built: Brain++. This is a great economic opportunity. However, such projects not only hide unclear challenges due to energy consumption and water consumption, especially given the lack of water in Bulgaria, but also hide the fact that we are still dependent on energy fossil fuels. According to EU data, our country is one of the highest energy poverty at this time.
– How exactly does AI consume and how does it affect climate change?
– Apart from the data mentioned above on natural resource consumption, we cannot say exactly how much AI consumes. No such data is published, and the scope of such calculations remains overwhelmingly tasked with numerous companies, models and applications (search engines, shopping, social media).
Interestingly, in the EU, AI companies need to document the energy consumption of their models, but at the same time, they need to focus on social risks due to the negative impacts of AI on health, security, equality and climate.
-What are the economic benefits of the sector?
– Economic benefits are undoubtedly stronger than social and environmental risks. Meta, Microsoft, Apple and Google have an unabated plan to invest billions in AI infrastructure. In organizational circles, there is talk of optimizing processes, saving time in creative activities, and developing new capabilities. And whether employees are dedicated to self-development remains questionable. Unfortunately, investments in AI in social activities, education and medicine that can bring extraordinary benefits are lagging behind plans for big tech giants.
– A few months ago, US President Trump – announced a $500 billion Stargate project to create a data center. This business size…
– Data centers are a huge business, and data shows that by 2028 AI will consume more than half of the energy in these data centers.
Unfortunately, the risks are also enormous. For example, in the aforementioned project, the forecast shows that each such center will engulf more energy than the US uses in New Hampshire.
– Water consumption in the development, training and use of AI models is also a serious waste of resources and risk. And the earth is getting dry and warm. What is this trend, and are we paying enough attention to it, or are we just living our lives?
– As we said, this is a really serious problem, but we often distance ourselves from the problem or go to extremes. We need accurate calculations of global energy and water consumption, and we work together to help us consciously reduce electricity consumption and create a sustainable society.
-How can you reduce the energy footprint of AI?
– Against the background of economic benefits – Companies continue to cause individual guilt, but can reduce the ecological footprint of AI companies. AI is not to thank ChatGpt to redirect to a more sustainable model, but not to use it for unnecessary activities in daily life. In fact, Deepseek has proven that AI models don't necessarily have to be very expensive and energy-consuming. Individual steps are not sufficient. Businesses need to take action, one of which is moving towards more sustainable models and technologies. AI has great potential in combating climate change, and in an ideal world, corporate profits will focus on this.
