News archive

How AI Helps Create a Green future?

Artificial intelligence has become a useful tool for many, both in everyday tasks and in professional life. Likewise, for researchers - AI is used both in creating models and forecasting, but also in analyzing and organizing data. Thus, several previously tedious and time-consuming activities have become significantly more convenient and allow resources to be directed to other important activities.

But how does AI affect the environment? First of all, it requires hardware, i.e. powerful computers and server farms, which require various materials and metals, including rare earth metals. Secondly, they use electricity, the cost of which can sometimes be relatively high. In addition, there are also daily administrative costs. However, it is important to take into account what we are using the entire artificial intelligence resource for - use it, for example, to play games, then its benefit is certainly significantly less than in a situation where artificial intelligence is used for research or development. Therefore, the impact of AI on the environment needs to be viewed more broadly than just its energy consumption.

From an environmental scientist's point of view, the positive aspects of artificial intelligence still outweigh the negative ones. For many researchers and specialists, AI simplifies daily work, helps to save human resources, and to improve international cooperation. Automation helps to monitor and analyze data around the clock, which gives more time for those jobs where the human role is still indispensable. At the same time, when assessing environmental impacts, you have to look at the broader picture, i.e. AI may be part of the bigger problem, but next to it, there is also an ever-increasing amount of digital waste. On the other hand, AI has a chance to become a useful helper here as well.

But how big is the impact of one server park on the environment? It is good to use their cost to compare different impacts. For example, it takes hundreds of millions or even billions of dollars to train a single language model like ChatGPT. However, training is a one-time activity and it ends when people can start using it - then it is already an application of AI. Each implementation takes less energy, but this too can add up to a large sum for millions of users. It is reasonable to reach a situation where the same amount of energy has been spent on implementing the language model as on training, as this is a significantly longer process and signals that the language model is fulfilling its purpose. If we look at the energy required to generate the output of one language model, the output generated by ChatGTP, which takes one hour to read, costs about 1 euro or 5 kWh. This is equivalent to the energy consumption of an electric car in 15 minutes. This is considerable energy consumption, but not too high if you look more closely at what is actually done with it. The more targeted the use of artificial intelligence, the bigger the win. During the past year, ChatGPT's energy consumption has decreased fourfold, which is due to a better model, as well as its more intended use.

Artificial intelligence and global climate problems have also brought a lot of work to the field of law. For example, in recent years, there have been several cases in which decisions previously made were reinterpreted due to new norms. This is due to the fact we have received new responsibilities from the environmental point of view. Many questions related to artificial intelligence are also related to ethics, i.e. what's right and what's wrong and what can or cannot be done. We often talk about things and the problems from "our" point of view, but we must define the nature of "us"—whether it is the community, Estonia, the whole of Europe, or the world. In the global picture, the conditions of people and the environment are very different, and it is difficult to even compare them—so the same principles cannot be used. In addition, we also have to take into account the time dimension, that is, we should not rethink the decisions we make today in a few years. For this, a good theory is needed, on the basis of which the necessary assessments are made, which take into account the goodness of the decision to be made, its consequences and responsibility. It must also be taken into account that a decision that is good for one person may be bad for another, i.e. a narrow social group may benefit from a certain decision, but someone else will be harmed. For example, the production of a smartphone is beneficial to its user, but more harmful to a factory worker in a third world country. Therefore, processes and the impact of decisions have to be considered more broadly.

How does AI help environmental researchers? The most important keyword at the moment is forecasting—this includes weather forecasts, flood risk assessment, modeling of forest fires, and other environmental problems. In the case of these examples, there may be little that the human mind can do, and in addition, technology operates 24/7, thus being an irreplaceable tool in certain jobs today. It can also be used very successfully in creating a personalized learning path and learning materials. If AI's accuracy and extensive predictive power have a useful output, it also has a downside - for example, highly personalized ads lead us to consume more and more.

In summary, we can state that despite the advent and use of AI has created a number of new challenges, the opportunities created by it far outweigh the negative side. The future with AI is green, but in order to ensure sustainability, we must also make meaningful and balanced decisions at the individual level and review our consumption habits.