CodeCarbon, a tool that tracks how much an algorithm pollutes

FOLLOW

• Follow

If every British adult sent a thank you email less per day, 16 tonnes of CO2 would stop being emitted annually, the same as going to and from Madrid to New York by plane about 22 times in a row.

This estimate from Lancaster University shows how user activity on the Internet affects the environment.

Like an email, every line of code that is written to program an artificial intelligence system results in contamination.

How much exactly?

That is the question that the Institute of Artificial Intelligence (Mila) of Quebec (Canada) has proposed to answer together with a data science team from BCG and the University of Pennsylvania.

Together, they have developed CodeCarbon, an open source tool capable of quantifying how much carbon dioxide algorithm programming generates.

Artificial intelligence has already proven useful in tackling various environmental problems, such as combating deforestation or helping agriculture.

But the exponential growth that it has had in recent years means that more and more energy is needed to support the massive computing behind it.

Training a machine learning algorithm can require running multiple machines over days or weeks.

So the benefits of these systems come hand in hand with a great environmental cost in the form of CO2 emissions.

"Until now, there was no standard way to track carbon expenditure," explains Yoshua Bengio, founder of Mila and winner of a Turing award.

"Putting numbers on pollution can help us become aware of the magnitude of the problem."

The system is simple: it consists of adding one more line of code, which is responsible for calculating the amount of carbon dioxide produced by the devices used to execute the code.

"The tracker records the amount of power used by the infrastructure of the major cloud providers and local data centers," explain the creators.

"Based on publicly available data sources, estimate the amount of emissions produced taking into account the electrical grid to which the

is connected

."

Then it shows the results compared to data such as kilometers driven or hours of television watched in an average home.

So programmers can get a tangible idea of ​​how much their code is polluting.

Its aim is to incentivize developers to optimize the efficiency of their work and advise them on how they can reduce emissions by selecting their cloud infrastructure in regions that use lower carbon energy sources.

"Data scientists can also do a lot to reduce the carbon footprint," agrees Sylvain Duranton, senior partner at the Boston Consulting Group (BCG).

"This shows that changing our behavior can reduce pollution."

But the goal is not just for programmers to be aware of this information, but for it to scale and serve senior positions as a basis on which to make decisions to take better care of the environment.

“Transparency and awareness of our carbon footprint is the most important right now.

When the top managers of the companies have tangible data of how much they are polluting when programming their systems, they will be able to make wiser decisions ”, says Bengio.

"This alone may not solve all problems, but it can help us become more aware," agrees Duranton.

From awareness to action

Once the data on the carbon footprint is clear, steps can be taken to reduce it.

There are some programmers already at it and writing fewer lines of code to combat the climate emergency.

It is a movement known as green programming that seeks to reduce the carbon footprint by designing web pages with the least possible code.

The higher the HTML, the higher the environmental impact.

Aware of this direct relationship, Danny van Kooten, creator

of the WordPress

—a piece of code that extends the functions of a program or tool— that helps page owners use the Mailchimp mass mailing service, decided contribute its bit in the fight against the climate emergency by reducing the lines of code it had created.

In the event that we don't know where to start modifying the code, WebsiteCarbon.com can help us out. Calculate the carbon dioxide footprint of any website and check if the server that hosts it works with renewable energy. “Any element that consumes electricity is a problem for the environment. If a page can consume a smaller amount of energy while maintaining its functionality, it will always be positive ”, explains Van Kooten in this report for EL PAÍS TECNOLOGÍA. This idea coincides with Begio's vision: “Today there are still many

between the human brain and the machine, but they will not be completed by using more data, more algorithms and larger models, but by doing it more efficiently ”.