How to deal with the energy consumption of AI solutions?
Europe’s emission reduction targets are approaching fast. There are great hopes for AI to help industry to reach their sustainability and emission targets. On the other hand, we need to ensure that AI itself is sustainable, in particular with an eye on the significant energy required to train large AI models.
Prof. Ali Hessami, Director of R&D and Innovation, Vega Systems and Chair of IEEE 7000 standard, Valerie Livina, Principal Research Scientist at NPL, Green AI Project Editor at CEN-CENELEC JT21, Anne Mollen, Senior Policy Manager at AlgorithWatch / Project SustAIn, Giacomo Bravaccini, Product & Marketing, Flexidao, Melanie Kubin-Hardewig, VP Group Corporate Responsibility, Deutsche Telekom AG, discussed perspectives on AI development, application and framework as well as standardization with their host Andrea Appel, Sustainability Lead at VDE e. V., at the exhibitor forum of embedded world 2023.
Host Andrea Appel outlined the three waves of the discussion regarding AI. The first wave concentrated on the general application fields. The second wave focused on practical machine learning and trustworthiness and the current third wave addresses sustainability aspects in relation to AI development. The panel discussion also floated on this third wave.
“If we really take in to account the lifecycle costs of AI solutions, the analysis of AI would not be that optimistic, as we see it today”, Anna Mollen pointed out. Melanie Kubin-Hardewig agreed: “AI will drive the energy hunger of the data centers.” Especially when it becomes a mass product that invites users to play around – as we now see with ChatGPT.
The panelists discussed the question if and how this consumption can be steered. Can standardization help to increase sustainability in AI lifecycle? For Giacomo Bravaccini the keyword is transparency: “We know very little about the energy consumption of AI solutions – especially as a mass product.” Transparency in this field can in his opinion be the basis for a good regulation.
Valerie Livina and Prof. Ali Hessami agreed on addressing this issue from a systemic perspective. The unanimous conclusion of the panelists: The goal must be to create awareness for the energy consumption of AI solution. From this starting point, the discussion can be continued.
Watch the whole panel discussion on Sustainability with responsible AI
