Data is the lifeblood of the digital economy, and a rising number of connected devices means that it’s increasing all the time. But data is only one benefit of digitization; another is business intelligence. Processing data to ensure that effective, real-time decisions are made using elements of automation requires power that only artificial intelligence (AI) and machine learning (ML) can bring to modern operations.
The market for AI is growing. As shown in CCS Insight’s Senior Leadership IT Investment Survey, 2022, 57% of organizations are deploying AI in production, up from 45% in 2021. A common theme in our research is the importance of AI for effective automation, which among other things improves problem identification, reduces operational cost and aids faster delivery. These ultimately enable more efficient operations with greater opportunities for revenue growth.
Amazon Web Services (AWS) re:Invent 2022, the company’s global customer and developer conference, highlighted its moves to make AI and ML more accessible for users of all skill levels. The range of existing applications for AI — not to mention all the potential uses being developed — reinforces the prospect of intelligent operations becoming pervasive throughout industry.
But there’s a wrinkle in this positive outlook. AI uses a lot of energy and resources during a life cycle encompassing its design, training, implementation and retuning. If its energy source isn’t carbon-neutral, high emissions are inevitable. The process of training a single natural language processing GPU model has been shown to produce carbon emissions equivalent to those of five cars over their lifetimes. Another study estimates that a single training session for a language model has a carbon footprint similar to a car journey of over 700,000 km.
Once an AI model is trained, it can be reused with less power for inference processing, where projections are made from new data. But this doesn’t negate the initial energy expenditure, nor the fact that models regularly need retuning, further increasing energy consumption.
As the impact of climate change is increasingly felt around the world, governments, organizations and individuals are looking to make positive changes. Processing and analysing data to enable meaningful action is where AI and sustainability meet. Therefore, it’s vital to ensure that the vast energy costs of certain AI methods don’t outweigh the good they can do. Equally important in mitigating carbon emissions for model training is the sustainability of the infrastructure used to power it.
Cloud computing is a crucial part of sustainable AI. The technology offers far more efficient use of resources, with hyperscale providers increasingly using renewable, non-carbon-based energy sources. Infrastructure is shared, and is much more efficient than on-premises data centres. Cloud providers such as AWS are finding new ways to cut energy and water usage alongside creating more efficient computing, storage and memory capabilities. Most major providers have committed to net zero emissions and the use of completely renewable energy sources.
AWS and AI-Driven Sustainability
AWS shows notable strength when it comes to implementing sustainable AI strategies. It does so to eliminate its own emissions, at least in part by enabling customers to achieve their sustainability goals using cloud and AI solutions.
To meet its own sustainability goals, AWS has previously launched green initiatives and made changes to its business and cloud infrastructure. It has a broad portfolio in this area that includes tools, education and training services, community engagement and sustainability programmes.
Optimizing workloads to lower power consumption is an important first step organizations can take, which is why AWS added the Sustainability Pillar to its AWS Well-Architected framework. This adds guidance on the most appropriate practices and services for building sustainable applications. Outlining the potential trade-offs of a specific choice adds depth to the improvement plans that customers can use to meet sustainability targets.
The premise is that by following the strategies suggested and using AWS’ Customer Carbon Footprint Tool, clients can track results and reduce their workloads’ emissions. Practices developed by customers include using serverless computing and reducing unnecessary resource use.
Of note are the AWS Well-Architected framework’s dedicated resources for optimizing sustainable AI and ML workloads. These provide strategies that address the tech’s entire life cycle, including identifying business goals, model development, training, tuning, deployment and monitoring.
Amazon SageMaker: An AI Power Tool
Inside AWS’ portfolio of AI and ML services, Amazon SageMaker offers a model development platform on the cloud that covers data collection, development and training through to monitoring after deployment. AWS sought to create a cloud capability that simplifies things for data scientists, optimizing Amazon SageMaker for those well-versed in ML while making the tech more accessible in general.
A raft of features expanding the capabilities of Amazon SageMaker were announced at AWS re:Invent 2022. Further details can be found in our report on the event; those without a subscription can contact us to learn more. Collectively, the new tools further improve the quality, management, reporting and accessibility of the entire AI and ML development process.
Geospatial capabilities are an important addition to Amazon SageMaker, allowing customers to more easily develop ML models for a range of new uses including climate science, urban planning, disaster response and precision agriculture.
Being Responsible with AI
Compliance with ethical strategies must be considered at all phases of the AI and ML life cycle, from the data used in developing the model to the model itself and how it behaves in production. To help customers, AWS created several features that track the tech’s life cycle and provide dashboards for monitoring.
One example, Amazon SageMaker Clarify, enables customers to detect potential bias in the different stages of developing an ML model, continuing its tracking even once the model is deployed. The service can also demonstrate how different inputs affect results, a critical requirement in many regulated industries.
Paving the Way for Sustainability
According to our Senior Leadership IT Investment Survey, the top barriers to AI adoption in organizations were a lack of trust in the tech, the time taken to build, train and deploy models, and the lack of industry-specific applications. This highlights the value of suppliers with large product portfolios and strategies supported by a choice of sustainability-focussed services and initiatives.
Crucially, AI and ML development has become more accessible, with improved quality outcomes and greater revenue potential. And, as with many of its partners and competitors, AWS has pre-built intelligence solutions that address common industry uses and reduce barriers to entry.
Inside its data centres, AWS is working on efficient water and energy use alongside the reduction of carbon in new builds. For example, Amazon is the world’s largest corporate buyer of renewable energy, with 380 projects, and is on track to power its operations with 100% renewable energy by 2025 — five years ahead of its original 2030 commitment. The company also made a commitment to be water positive by 2030, and at re:Invent 2022 announced its 2021 global water use efficiency metric of 0.25 litres of water per kilowatt-hour.
The company has also pioneered a new generation of AI-optimized hardware, including the AWS Graviton and Inferentia processors that use up to 60% less power than previous generations. The company claims that by migrating to AWS, an organization can lower its carbon footprint by nearly 80% compared with the average on-premises data centre. AWS believes this could rise to 96% as it moves to entirely renewable energy sources.
But moving workloads to clouds fit for sustainability is only part of the path toward effective operations and sustainability programmes driven by AI and ML. The rest requires a clear strategic vision and committed investment and leadership.
AWS has built a broad portfolio in support of making the AI and ML life cycle more sustainable, along with its deployment of operationally proven uses. A lack of guidance is no longer an excuse for not implementing sustainable AI and ML projects when a technology platform like that of AWS provides the assets needed to support an organization’s operations and bolster their green agenda.