3 The sustainability of AI
Sustainable AI covers two broad areas: first, sustainability in terms of the impact on society, and second, sustainability in terms of the impact of AI on the natural world (not only the resource being consumed, but also the effects of this consumption). Work in this area involves both adapting AI to make it more sustainable, based on insights drawn from the field of sustainability, as well as working out ways in which AI can contribute to sustainability.
Read theby Sam Gould on Deloitte’s website (Gould, 2020), which discusses several important aspects of how to go about achieving sustainable AI. The following activities draw these out – go through these in order, and write your responses in the free text box provided for each.
From Gould’s post, make a list of the sustainability challenges that could be addressed by AI. Try to turn these into brief discussion points, adding details where necessary; these points will be useful for later activities. List these points in the follow free response box.
Some possible sustainability challenges that AI could address include:
- Enhancing renewable energy sources
- Energy modelling for infrastructure optimisation and urban planning
- Energy network management
- Utilising diverse data sources for environmental monitoring and targeted sustainability
- Acceleration of climate science (physics emulators, climate forecasting, materials science, etc)
- Reduction of global food waste
- Algorithms to classify images of illegal animal products
- Green AI within Smart Cities
Note Gould’s mention of the so-called ‘Circular Economy’ and try to find out more about this (e.g. on the internet). Summarise the ways in which AI could be involved in this. Enter your summary in the following free response box. Note: Try to go beyond Gould’s text, by finding examples from the internet.
If you are stuck on this part, note that from Gould’s post, there are three main concepts:
- Keeping products and materials in use
- Regenerating natural systems
- Designing out waste and pollution
Look on the internet to see if you can find examples of each of these, and add a brief description in the free response box.
In the following free response box, list some suggestions from this article on how to make AI more sustainable.
As Gould observes, one very interesting recent approach to making AI more sustainable is the method of ‘compressing’ models. This method acknowledges that the very large models built from huge amounts of data, often from the internet, frequently include a high-level of redundancy, so that shrinking such models in clever ways (i.e. removing redundancy while maintaining performance), can make the use of such models far more sustainable.
Reflecting back on the above activity, Gould’s post presents an example of AI being deployed in a new area, which has its own ways of addressing the key challenge, in this case sustainability, and that AI is being adapted to this area.