Across the world, leaders and businesses are looking at how AI networks capable of generating new content or data from reference material can contribute to facing the climate challenge. As generative AI becomes embedded in ways of working and continues to advance and develop, businesses look to understand; what are the benefits and costs to the climate of this emergent technology?
What can generative AI do to help sustainability?
Generative AI can help turn reporting into insights
For many businesses, the hardest step in a sustainability journey is effective baselining – understanding and quantifying their current climate impact. If data infeed’s are automated and understood, generative AI can provide real time reports and create insight, helping to guide sustainable business decisions. Uniquely, if data is not available, generative AI can begin building models and making assumptions to bridge data gaps and fill in the blanks.
Generative AI can help improve planning and design to reduce waste
If it is implemented effectively, generative AI can have a significant impact on planning and forecasting processes. SAP have produced Joule, which can provide contextualised insight and guidance for business operations. By simulating real world scenarios, it can enable better decision making and provide the insight needed to minimise wastage and maximise the use of existing resources. Similarly, Microsoft’s supply chain management system Dynamics 365 uses Copilot’s generative AI to identify problems and suggest alternative solutions. For example, it could help operationalise circularity by making estimates about the most effective route to recover value from a damaged product, thereby increasing its useful life.
In design, whether architectural, mechanical, or industrial, generative AI can produce as many iterations as requested far quicker than would be humanly possible. Coupled with its simulative capability, this provides a compelling use case in design. If parameters are understood, this provides a huge opportunity for building in efficiency and reducing the climate impact of buildings, machines, and businesses.
Generative AI can help guide consumers to better products
Marketplace platforms, such as Amazon or eBay, personalise how they present their offerings to customers based on user data. Google use their Performance Max (P-Max) AI package to go a step further in customising search, individually tailoring product recommendations and automatically creating a summary and a strapline to attract consumers. Coupled with better forecasting, this could reduce product waste by guiding consumers to the most appropriate product. This can also help minimise returns, which are a significant generator of supply chain wastage for marketplace platforms. In fashion, for example, it is becoming common to see return rates of 30% or higher as people select the ‘“wrong’” products and must return them, increasing the climate impact of that product.
Generative AI can help understand our impact on the climate and optimise renewable usage
The climate is inherently complex, with interplaying factors and non-linear cause and effect. This presents a significant challenge for modelling impact. Generative AI provides a tool to analyse the vast quantity of available data on everything from emissions to currents, weather patterns to deforestation, and ocean salinity to soil degradation. These models can provide insight that existing programmes may overlook and can be used by researchers to further understand the ongoing challenge.
Most sources of renewable energy drawing from flows (tidal, solar, or wind) will have variation in their output, while energy demand fluctuates constantly. Generative AI can help predict and model demand and supply of energy and help optimise the utilisation of these resources. It can also feed into investment decisions, guiding the placement and development of new sources, ultimately reducing our reliance on fossil fuels.
But at what cost?
Generative AI is reliant on huge amounts of computing power
While generative AI has enormous potential to help with the required steps to combat climate change, it does require a vast quantity of resources. Machine learning requires a significant amount of processing power, using thousands of graphics processing units in data centres around the world. Data centres have often been built in unsuitable locations, requiring more resources to cool, and putting further strain on limited water resources.
Generative AI requires increasingly large amounts of energy to train and operate
As well as the physical resource cost, each request made of a generative AI has an energy requirement and a resulting carbon impact. To train GPT-3, to understand datasets and provide references required 1.28 gigawatt-hours of energy, equivalent to 552 tons of CO2.GPT-4 is estimated to have required between 51 and 62 gigawatt-hours to train. However, the true energy impact of generative AI is in the use, not in training.
This impact may reduce going forward, but the technology is still in its infancy
The good news is there are potential steps to be made to make this significantly more efficient. Cutting edge modelling techniques require fractions of the power and can produce more with less. Data centres around the world are investing in greening their energy supply, producing renewable energy for their own use and as the world becomes more connected, data storage businesses are incentivised to locate heat producing racks of GPUs (graphics processing unit) and data storage in colder geographies. Any calculation of the costs and benefits of generative AI will change every month and year as the technology leaps forward.
Generative AI will always need oversight
It is becoming ever easier to generate reams of data with the click of a button, and therefore it is becoming ever harder to check outputs. Generative AI could present a challenge to the old maxim of “bad data in, bad data out” by skipping the first step. Without robust checks of what exactly it is generating and firm control over the parameters within which it can operate, there is a risk of organisations mistakenly accepting AI outputs as true. This risk is compounded by the creation of customised AI agents. These use OpenAI models as a backend to provide generative capability, while absorbing new information from business sources. These customised agents have specific knowledge that can only come from company data, but do not have any additional capability to understand what it means. Given that its inputs and outputs are, at surface level, of better quality, users may be more inclined to trust its outputs over other openly accessible tools.
So how sustainable is Generative AI?
Generative AI is a powerful tool that could be used to make a significant difference in tackling climate change and making businesses more sustainable. However, developers and data centre owners need to be aware of their impact on the climate and the steps they can take to mitigate. Businesses should understand and assess what impact generative AI can have to help them become more sustainable, resilient, and profitable
Authors: Tim Oliver, James Byrne, Joe Miller
Share:
