“What will count in the end is how we humans use AI”
Christina Hoppenbrock took this occasion to ask physicist Dr. Oliver Kamps how artificial intelligence (AI) can help to conserve resources and, in particular, what challenges it poses for humanity. Kamps is a member of the team managing the “InterAI” teaching programme at the University of Münster and will be one of the guests on the panel discussing the question of “Concerns about AI” on the Sustainability Day.
How can AI help us to live more sustainably?
First of all, we have to distinguish between the sustainability of using AI and using AI for sustainability. In the latter case, AI can help to optimise controlling and planning activities in a wide range of areas and help to conserve resources: from improving traffic flows and controlling the energy network in an efficient way, to process optimisation in industry and the best possible use of water and fertilisers in agriculture. AI can also help in developing new materials for batteries or in improving solar cells – thus making a contribution to sustainable development.
But the use of AI, and especially training the programmes, are extremely energy-intensive, aren’t they?
This concerns sustainability in using AI per se. The simple truth is that the algorithms are performed on computers, and they use up electricity. The manufacture of computers and chips is a complex industrial process which uses up energy and resources. If we use AI on a large scale, the resources needed increase accordingly. According to estimates, data processing centres alone are today responsible for around ten percent of global electricity consumption. When applications such as ChatGPT are used in our everyday lives, electricity consumption will continue to increase. Let me give you one example: In comparison to the traditional use of search engines, electricity consumption rises tenfold if the text used for the search is generated by a speech model. So at the University of Münster for example we have research being undertaken to develop special hardware for AI which needs much less energy.
Do you see any other problems and conflicts which the use of AI can give rise to?
Developing and training large-scale AIs is extremely expensive and, in the case of large-scale speech models, the training can cost between 50 and 100 million euros. An advantage in preparing large quantities of data – for example, annotating images or more training of speech models through interaction – is the availability of cheap human labour. Both aspects throw up ethical and social questions. Systematic distortions in some datasets used for training can also lead to ethical problems if decisions are based on an evaluation carried out by an AI system. Over and above this, the capabilities of today’s AI systems are so highly developed that they can be used for monitoring purposes, for generating fake news or in military applications. These also involve huge problems.
So what is your conclusion? Are sustainability and AI compatible?
If we only look at the question of resources, we can already see two conflicting aspects. Which of the two will gain the upper hand will depend on technical developments. What will count in the end is how we humans use AI.
This article originates from the university newspaper wissen|leben No. 6, 2 October 2024.