In Summary
- AI can harm or help the planet, depending on how it's used.
- Organisations sometimes struggle to measure AI's impact, but awareness and expectations are growing.
- The STAR Framework offers a clear process to make AI more sustainable.
- Using the right tools and actions, companies can reduce AI's environmental footprint.
In 'Why AI poses a threat to Net Zero goals', we discussed the sustainability challenges associated with AI as well as the barriers for organisations to overcome them. Since then, AI has been in the spotlight more and more for its impact on the environment, and public consciousness of the sustainability of AI has risen significantly.
In April, the Government Digital Service (GDS) added sustainability to the Government Design Principles, this marked public acknowledgment of the large amount of energy, water and materials required to run digital services. GDS advises practitioners to consider both short- and long-term environmental impacts when designing and delivering services.
However, barriers for organisations to measure and reduce the environmental impact of their use of AI remain. In this article, our focus now turns from the “why” to the “how” – how we can do something about these environmental impacts to ensure the sustainability of AI and the implementation of AI for sustainability outcomes.
What’s the definition of Sustainable AI?
Sustainable AI contains two related topics:
- Sustainability of AI: The practice of reducing the negative environmental impact of the use of any form of AI.
- AI for sustainability: The use of AI technologies to address environmental, ecological, and social challenges and promote sustainable development.
Measuring, reducing and prioritising impact using the STAR Framework
We’ve looked at the landscape surrounding the development of AI – the large and growing issues around AI’s environmental impact – but organisations are struggling to measure and act to reduce it. Alongside this, AI has the potential to be used for sustainability outcomes, but businesses need help to prioritise and implement them.
To solve these problems, we have developed the STAR Framework (shown above). The Framework guides organisations through a four-step process to identify the scale of risk and opportunity around Sustainable AI, supporting the achievement of business objectives with this at the forefront.
Here, we break down the stages of the process and provide some examples of what can be done in each to create a holistic approach to Sustainability in the context of AI:
Strategy
Firstly, it is crucial to understand what the organisation in question is trying to achieve or what the business objectives are that AI may be able to solve? What, if any, AI strategy is in place on AI currently?
As part of this, we engage stakeholders to align with the overall business strategy and goals to ensure AI use cases are identified and validated against a business need. We would identify what business and environmental factors are important to the organisation in question – is AI’s impact on water usage, carbon, hardware or others the most relevant.
For example, we can help understand the energy requirements, associated emissions and sustainability of the infrastructure used for AI development and deployment, considering whether the system is optimised for energy efficiency. We can also set a strategy to realise sustainability benefits (direct and indirect) considering AI’s potential to address environmental issues. We also get clarity around the operating environment, associated regulation (present and future), government commitments where relevant and provide confidence that our plan aligns to these influencing factors.
Tally
The next phase is all about quantifying and measuring the impact of AI in the particular context of the organisation. We understand the macro numbers from our previous piece but it’s critical to get a view on the individual business in question to be able to produce a targeted action plan in the later steps. What are the impacts on the ground and how can we reduce them? Also, considering the correct tooling and solution to make a tangible impact on the situation, helping the organisation track progress moving forward.
We’ve created a measurement approach aligned to the organisation’s current and future use of AI by leveraging a tailored combination of 30 open-source tools developed by Sopra Steria. There are a number of open-source calculators out there to help with measurement activity, with AI Energy Score, EcoLogits and ML CO2 Impact notable as particularly useful. Implementing Sustainable AI measurement tooling allows us to quantify current risks and opportunities of using AI, setting a solid foundation for future implementations. Implementing Sustainable AI measurement tooling allows us to quantify current risks and opportunities of using AI, setting a solid foundation for future implementations.
Action
Once we have a baseline on our actual usage from Steps 1 and 2, and a view on the measured impact of current and predicted future AI programmes, what are the best practice methods we can employ to put a plan in place to define a tangible reduction in environmental impact of AI?
We tailor action from 31 different best practices that have been developed alongside government, industry and academia. These best practices have been ranked and categorised based on the reduction in environmental impact achieved, the implementation effort and the lifecycle stages they are relevant to. Best practice spans improvements to the design, data collection, training, governance and user awareness.
Review
In the final step of the STAR Framework, we re-baseline and re-measure to analyse the benefits of the work that’s been done in the previous steps. Out of this, a roadmap and options for continual improvement continue the journey. This may be a simulated, scenario-based step depending on the scale/profile of the organisation or if it is too soon to see a tangible difference. We re-measure the impact of the action taken and review against strategy to identify and reprioritise further use cases for development.
What’s next? Let’s explore it together
If the above has resonated with you or sounds like we can help with your challenges around implementing AI sustainably, please do reach out directly to us. We’d love to discuss your thoughts on the topic, different perspectives from your own experience on what we have discussed, or problems we might be able to help you with.
Look out for more content from the Sustainable AI team within Sopra Steria Next over the coming months!