A new study produced by researchers at the University of Massachusetts, Amherst, surprises us with regard to the fact that the artificial intelligence industry is a contributor of environmental pollution.
According to the results achieved by the researchers, developing a single software based on artificial intelligence can lead to the emission of more than 100,000 pounds of carbon dioxide equivalent. It is the average material emitted into the air by five cars in the United States during their entire course of use.
What consumes the most, and therefore causes the most emissions in the environment, is the training phase of the software based on machine learning: the latter must in fact overcome a long initial phase during which are inserted in the program large sets of data, increasingly large and massive, to ensure that the same algorithm is “instructed” and then works properly when put into operation.
Obviously this phase needs a lot of hours and days of work, a period during which the energy consumed by the computers reaches the peaks. This is without counting the use of all kinds of hardware, for example by the robotics and automation industry, which involves even more expensive testing phases in this regard.
Naturally, thanks to artificial intelligence, very efficient neural networks have been created, very useful in many contexts and very advantageous also with regard to the context of pollution, but this analysis sensitizes and provides an answer to those who believe that automation through software is unassailable in this sense. And perhaps this data could be used in the future to push researchers to develop increasingly efficient software and algorithms.