Friday Dec 02, 2022

Can AI be green? | THE Campus Learn, Share, Connect – Times Higher Education

Artificial intelligence is seen as an almost magical solution to complex problems in fields as diverse as computer science and art, and from finance, economics, natural and medical sciences to linguistics.

In the education sector, especially engineering and computer sciences, AI is a vibrant and attractive subject for students; AI brings possibilities for developing new intelligent tools and building successful careers. Across disciplines, AI is integrated into systems to improve learning methods and allow better interaction between professors and students. AI can be used to develop innovative teaching and learning practices that raise the quality of education.

As it becomes ubiquitous, why not also consider how to make AI green and sustainable?

Teaching AI alongside its environmental impact

The workshops or seminars that introduce notions of artificial intelligence to our students could also address the risks and challenges that AI poses for the environment.

These are, for the moment, opaque to the public and most students, even those who are fascinated by humanoids, electric cars, androids and other AI-enabled miracles. This is indeed the consequence of courses given by academics and educators who focus only on the advantages and applications of AI and not its limitations, risks and challenges.

Students should be encouraged to ask:

From our experience in the department of engineering of the University of Luxembourg, the teaching of AI and its uses should also highlight its drawbacks and limitations, especially since most users of this advanced technology are replacing, without thinking too much (because they are uninformed about the consequences), conventional algorithms and solutions with AI.

How does AI impact the environment?

Learning outcomes for AI courses should take its costs and risks into consideration.

Big data is not free

AI relies on a large amount of data to learn and reproduce models adapted to the studied system. In other words, the effectiveness of its results depends on the quantity as well as the quality of the data. Complex problems require enormous amounts of data and, of course, exceptionally long and tedious calculations. For example, during the training phase for image recognition software, the algorithm analyses many images; each pixel, including its intensity and colour, is considered to detect patterns. Once the software is trained, many calculations are still needed (pixels, distances, pattern comparisons, normalisations, gradients, etc). The computing power requirements for AI are linked to the data amount and are growing exponentially. Data collection and storage are not free, either; they consume personnel and IT resources. Moreover, often, the data collection needs sensors, generally connected by advanced communication technologies. These technologies represent between 6 and 10 per cent of global electricity consumption, and 4 per cent of our greenhouse gas emissions. This figure increases by 5 to 7 per cent each year. In 2018, OpenAI found that the amount of computational power used to train the largest AI models had doubled every 3.4 months since 2012, which translates into exponential energy consumption over years. Home office equipment accounts for about 9 per cent of the average household electricity bill. Office equipment accounts for about 11 per cent of electricity usage in the tertiary sector. Irresponsible …….


Leave a Reply

Your email address will not be published.

Back to Top