Does the algorithm pollute? The 'footprint' left by artificial intelligence

725-alexa
Date

February 23, 2022

Campus

All Campus

Knowledge Area

Tech & Data

By José Ventura, DataScience Leader - IBM Software and professor of SIBT in partnership with IMB

*This article was originally published in the magazine Ethic on 31st January 2022

 

Sixty years ago, few could have imagined that Alexa would be able to answer all our questions by voice six decades ago or that by scanning our irises, we would be identified with names and surnames. However, 'the science' that would enable these ideas, which at that time seemed like science fiction, was already in the making.

And today, few are aware of some of the effects of this reality. And I am not referring so much to security or privacy issues, but to the environmental impact of asking Siri where in the city the latest Almodovar movie is being shown. Artificial intelligence is not only leaving a significant social and cultural footprint on today's society but also a carbon footprint on the environment. Every time an artificial intelligence algorithm is run, a large amount of energy is consumed and, according to Moore's Law, from 1959 to 2012, the computing power required and the amount of data needed to handle artificial intelligence has doubled every two years. And from 2012 to now, how much has this discipline developed?

Due to the ever-increasing volume of data and increasing computational needs, large data centers have been built that require significant power consumption. For example, in the United States alone, data centers account for 1.8 percent of the country's electricity consumption.

726-green-ai

McKinsey estimates that the IT market will be responsible for between 3 and 4 percent of the world's CO2 emissions in 2020.

Faced with this reality, public and private agents must react with actions aimed at reducing the environmental impact of digitization and artificial intelligence. How? Hand in hand with green algorithms.

Yes, here too, the color of sustainability must sprout. The goal is for any business data to be interpreted by green, sustainable algorithms, whose execution requires less energy consumption to balance the volume of data needed to train the models, the amount of time spent, and the number of iterations aimed at optimizing their parameters. Consequently, although it may sound like science fiction to many, the way an algorithm is designed directly impacts the energy resources required when running that algorithm. It also considers using renewable energy sources in the creation and application of these models and ensuring the efficiency of that energy consumption.

All in all, green algorithms are designed to use artificial intelligence in a more inclusive and planet-friendly way, maximizing the energy efficiency of technological infrastructure. They are algorithms designed so that, when executed, they are more efficient, consume fewer resources, and achieve the same result that a more complex algorithm would obtain, in short, being more sustainable for the environment.

Similarly, the shared use of information technology services hosted in the cloud also contributes greatly to sustainability. This use, through cloud computing, has a very positive impact on the environment compared to traditional computing, since companies or users who do not have their own data centers have information technology services in the cloud or hybrid, so that they can take advantage of the benefits of having a highly scalable technological infrastructure accessible via the Internet. They therefore only use what they need at any given time, allowing them to avoid acquiring excessive resources by anticipating peak usage for each service with the consequent damage to the environment.

At IBM, we work with technologies that allow us to build private and hybrid clouds. In addition, we develop standards-based software that allows the construction of private, public, and hybrid clouds with a management capacity that facilitates the independence of the business process of the cloud implementation and mobility as needed. All this, with high levels of security, data privacy, and scalability, and contributing to making the 'footprint' that artificial intelligence is leaving in our society as sustainable as possible.