Green algorithms: the road to more sustainable artificial intelligence

AI is fascinating, as evidenced by ChatGPT going viral and attracting so much interest. But in this enthusiasm, the high energy cost of these services is forgotten.

Source: Colpisa

We don't see it, but it is everywhere. Artificial intelligence (AI) has become ubiquitous in our daily lives. As technology has crept into everyday life and we have become more dependent on it for more things, so has the presence of AI.

Technological advances in recent decades have allowed it to be perfected and become much more efficient and effective. And while terms like algorithms, automation, 'bots' and the like are mostly left in the technology sections of the news—and in analyses of where the business world is going—the average person comes into contact with them far more often than they think.

"If you stand up and unlock your phone with the image of your face, that's artificial intelligence," explained Verónica Bolón, a researcher at the Center for Research in Information and Communication Technologies (CITIC) at the University of A Coruña, where she leads a study on green algorithms.

"In mobile devices, everything is pretty much AI," she added. Likewise, more and more things are becoming equipped with intelligence, which means they are generating data streams to respond to the services requested of them and expending energy. "A smart fridge, or whatever, may be sending that data to a supercomputer that consumes a lot," the expert explained.

How much does AI consume?

There are no exact measurements of how much energy artificial intelligence consumes each year globally—which business consultants continuing noting that it grows year after year—but estimates on specific aspects give an idea. According to researchers at the University of Massachusetts, Amherst, training a single AI model has emissions similar to those of five cars over its lifetime.

And the viral ChatGPT probably needed considerable resources to become what it is. "You don't think about it when you use it, but all it has had to learn has been many, many hours of computing on computers that obviously consume energy," stated Bolon, who points out that some estimates on an earlier model of the service speak of consumption equivalent to that of 136 Danish homes in a year.

The expert recalls that the final energy balance does not only include computing hours, but also other elements such as the cooling systems of the data centers supporting this service.

As artificial intelligence becomes more accurate, the amount of resources it needs to operate also increases. "I've read articles that say there are new algorithms with huge power consumption that are only 0.01% better than their predecessor," Bolon asserted, noting that sometimes it pays to avoid that "huge" power consumption because the gain in accuracy is neither as high nor really, in some cases, as necessary.

But, even so, artificial intelligence is there. It is set to become an increasingly important part of our economy and will become more and more sophisticated, more usable and more widespread. So should we change what it does and how it does it? This is the starting point for the so-called "green algorithms," which will be used to create green artificial intelligence. It is "more efficiency in terms of consumption," summarized Bolón.

A different way of 'cooking'

"An algorithm is the way of making the software applications that devices use," explained Coral Calero, director of the Green Algorithms area at OdiseIA and professor of computer languages and systems at the University of Castilla La Mancha. "It's the recipe that we write so that the computer understands it and makes the dish," she explained in a way to help all audiences understand. As recipes tell you what to do before and after to get an appetizing and tasty dish, algorithms tell you the order of things to get to a final result. "There are many ways to make the same dish, and it's the same with algorithms," she noted.

"There are many ways to program a functionality," she stated, pointing out that to make a green algorithm you have to "take the one that makes the best use of resources." The best way to understand it is with a mixed sandwich, as Calero explained it to us. You can go to the fridge to grab each ingredient separately, also taking a tour of the bread cabinet, or you can directly grab everything you need on your first visit to the fridge, saving you a lot of work. A green algorithm would not wait until the slices of bread and ham to go to the other side of the kitchen to get the cheese.

Green algorithms could be the future of a scenario in which, for now, there is nothing defined and there are only recommendations. "This problem of energy consumption is not regulated," Bolon pointed out. "For now there are good intentions, but it is expected that in the near future there will start to be laws and regulations [about AI] and that they will also consider energy consumption," she added.

Government lays the foundation stone

Spain just launched, with the Ministry of Economic Affairs and Digital Transformation, the National Plan for Green Algorithms, which aims to promote "an artificial intelligence that respects the environment and provides intelligent solutions to ecological challenges." The plan has been endowed with 257.7 million from European funds. "I think it's both a good idea and one that should go further," Calero said, noting that creating policies is fine but that "the problem" is knowing whether it will have continuity. It is "a very important first step", she reassured, with "things that can be improved."

In Europe, there is also interest in this issue and in legislating, in general, AI, not only on an ethical and legal level but also on its impact on the environment. This can have a direct effect on the global market, if we look at what happened with the European data protection regulations a few years ago.

The European Union then adopted a framework that was much more protective of citizens' rights over their online data than was customary. The standard was not only the mirror in which subsequent legislation elsewhere was looked at, but also became a default standard since it does not pay to make one service for Europe and then one for elsewhere.

This could be an important endorsement because, right now, green algorithms are more theory than practice. It is something that is being worked on and researched to bring change in the near future. Bolon is one of those researchers who are working in this field, to make, first, "algorithms green by design, optimized in their energy consumption" and, second, to "apply artificial intelligence algorithms to problems related to climate change and sustainable development".

And, as the experts noted, there is no way to measure the consumption that is being made in AI from the outside. "It is not easy to prove it," stated Calero, "but the fact that people are now talking about it is a first step. After all, as Bolón acknowledged, "even the researchers themselves were not aware of this a few years ago."