Artificial intelligence (AI) will transform the way we impact the environment.
AI can contribute positively to sustainability through initiatives like AI4Good, which identifies AI solutions that advance the UN’s Sustainable Development Goals.
It can also contribute negatively, increasing energy and water consumption, and emitting higher carbon dioxide emissions.
To restore the balance between AI’s positive and negative effects, we must change our behaviour.
This means shifting toward sustainable AI, a movement that aims to change every stage of the AI product lifecycle, including training, tuning, implementation, maintenance, and the AI ecosystem itself.
How can we use AI sustainably?
Sustainability addresses the tension between innovation and the distribution of resources across society from one generation to the next.
To make AI sustainable, then, we must strengthen our standards to avoid environmental challenges such as energy consumption and pollution.
The rise of generative AI illustrates its environmental impact. For instance, it has been calculated that for every kilowatt hour of energy a data centre consumes, two litres of water are required for cooling purposes.
Traditionally, energy and water usage are split fairly between model training, data processing, and the use of trained models.
As AI features become standard in applications, it’s expected that the use of AI-trained models will grow exponentially. This will result in increased energy consumption for AI GPUs in data centres and will affect major GPU producers such as NVIDIA, AMD, and Intel.
The increase in energy and water consumption results in direct costs, such as the use of energy and water, as well as indirect costs, including environmental costs (e.g., biodiversity). The European Union regulates sustainability through legislation, taxonomy, and standards.
However, there are still no formal goals or targets to reduce the environmental impact of AI.
Firms recognise the impact of AI, but their economic interests predominate.
Sustainable AI in practice
To understand how firms can use AI sustainably, we can look at an example of a railway maintenance firm in the Netherlands. Last year, this company showcased the importance of AI and how it dealt with negative effects.
The Dutch railway infrastructure is the most intensively used system in Europe and spans approximately 7,000 kilometres. Maintaining the infrastructure is labour-intensive, and delays and derailments are undesirable to users of the system.
The railway maintenance firm was aiming to meet its safety, quality, cost efficiency, and sustainability goals through digital innovations. They said that data and IT can play a crucial role in reducing human involvement in verifying the quality of railway components along the infrastructure.
These innovations can forecast which components need to be replaced or repaired. Consequently, digital innovations improve the quality and safety of railway tracks and may indirectly help achieve sustainability goals.
The firm developed AI solutions based on algorithms comprising various components, including the Microsoft Azure platform, Azure Function Apps, Service Bus, and a SQL database.
For example, the firm developed an AI solution to verify the quality of spring clamps, as seen in the image below.
The image illustrates this AI-driven visual inspection process. On the right is a broken spring clamp beside the track. The AI solution identifies the broken spring clamp on the left and highlights it in green. The firm has conducted use cases for detecting various types of sleepers, spring clamps, and separation welds.
By doing this, AI minimises the labour-intensive work of contractors, including their equipment and machines, creating lower CO2 emissions.
AI helped to detect the spring clamps.
Although AI solutions contribute to more efficient maintenance work, there is a downside. As the firm needs to comply with government regulations on sustainability, IT and data experts calculated that the degree of energy consumption and carbon dioxide emissions of the AI solutions would increase significantly.
Since IT service providers are responsible for managing and maintaining the AI solutions, the maintenance firm argued that these providers must develop a strategy to mitigate the negative effects. Various options were discussed, agreed upon, and listed in contracts.
Examples include green programming, reusable data, carbon-aware computing, carbon-aware software development kits (SDKs), and developing electricity maps. These actions have already resulted in decreased energy consumption and carbon dioxide emissions.
Changing our behaviour
Due to the ease of use of generative AI, multiple providers offer free apps, so users don’t have to pay.
As a result, environmental costs have not materialised. However, application providers who offer professional generative AI solutions take energy and water consumption costs into account as part of their license fees. Importantly, to restore the balance between the value of AI, such as AI4Good, and sustainable AI, we must change our behaviour too.
The examples revealed in the case study already contribute to a decrease in carbon dioxide emissions and energy consumption.
We need to do more and introduce other mechanisms that change our behaviour. What if users had to pay directly for using AI? When AI providers develop new business models and payment methods based on pay-per-use, we, as users, experience the environmental effects. In the virtual world, we are quite accustomed to easily exchanging data with friends and businesses.
Imagine that we had to pay extra to exchange AI data within our own networks. That would affect our behaviour. What if users were incentivised to decrease their environmental impact when using AI solutions? We could start with a virtual wallet of AI credits, and each time we exchange data, our credit balance would decrease.
Alternatively, we could use other options to find and exchange data, such as Google searches or Dropbox. These options would have a lesser environmental impact than AI. What could new AI regulations bring to the table to decrease environmental impact?
Future regulations may require manufacturers of GPUs and AI technologies to design and maintain their products in a more sustainable way. This would affect the entire lifecycle of AI products, as well as the AI ecosystem of suppliers and subcontractors. By doing so, we can benefit from AI while preserving our environment.
It’s time to address the sustainability of AI
Although there is a growing effort to use AI for sustainability (e.g., to achieve the Sustainable Development Goals), it is time to address the sustainability of developing and using AI systems.
By changing the AI lifecycle, ecosystem and our behaviour we are capable to restore the balance between the positive and negative effects of using AI.
