Generative AI is currently seen as both a blessing and a curse.

On the upside, it simplifies processes, enables businesses to be more efficient and allows employees to offload some of the more procedural work in favour of the big projects.

On the downside, it has an apparently limitless appetite for energy, producing alarming spikes in GHG consumption at a time when companies cannot avoid a sharp focus on sustainability.

According to IT services and consulting firm Kyndryl, however, there is a way of limiting the cons without destroying the pros.

In a blog, Liz Porter, Global ESG Consult Lead at Kyndryl, and Avintha Moodaly, Director, Environmental Management, argue that gen AI could actually hold the key to business sustainability.

First, the bad news

Before outlining the potential solution, it is worth going over the reasons why we have reached an apparent crossroads.

Liz and Avintha explain that gen AI tools and solutions “bring the promise of creating societal and business value at scale”.

They add: “Surveys indicate that enterprise adoption of this technology has increased substantially over the last year, with most companies testing generative AI in one or more parts of their operations.”

However, with 2024 set to be the hottest year on record, gen AI could turn up the heat dial even more.

Liz and Avintha say: “Running generative AI requires massive computational power, which is increasing the cooling requirements for data centres and upending existing power usage management plans.”

According to an IEEE Spectrum article, current AI technology could consume as much electricity annually as the country of Ireland (29.3 terawatt-hours per year), they add.

To further illustrate the point, they say: “Meanwhile, a single Large Language Model (LLM) interaction may consume as much power as a low-brightness LED lightbulb on for one hour.

“This may not sound like much until one realises that some of these workloads might call for millions of LLM interactions per day.”

The gen AI balancing act

Gen AI and sustainability is a subject that Kyndryl has given particular attention to in recent years.

That is reflected in the fact that Faith Taylor, SVP Sustainability, ESG & Impact, is on the Sustainability LIVE London AI in Sustainability panel on 10 September.

This collective expertise means it is worth taking notice of Liz and Avintha’s thoughts on the gen AI balancing act – and what businesses must do.

They say: “Despite its power demands, generative AI also can play a major role in conserving energy by making business operations more efficient.

“While there is no consensus on the magnitude of its power requirements, organisations must become aware of the impact that it is having on their carbon footprints and adjust their strategies to make generative AI more sustainable.

“To achieve their sustainability goals, companies should begin integrating sustainability data into business decisions and invest in technologies that provide greater visibility into sustainability metrics.”

Four steps to sustainable AI heaven

LIz and Avintha say Kyndryl’s enables its customer to “operate at the intersection of innovation, social impact and environmental stewardship” by focusing on four key protocols:

1 –  Measuring an enterprise’s energy consumption baseline

They say businesses should develop a clear picture of an organisation’s current GHG emissions and energy spend, adding: “Continuous monitoring is required to facilitate more efficient energy optimisation.

“But even before that happens, organisations need to determine whether generative AI is the best choice for their particular business needs.”

2 – Optimising energy usage

Once an organisation determines its energy-usage baseline, it will be better prepared to optimise its generative AI systems.

Liz and Avintha say: “As part of its transition to growth, Kyndryl took a hard look at its asset utilisation – including data centres and real estate.

“We reduced our real estate footprint and shifted operations to more efficient, state-of-the-art data centres. We also consolidated many of our servers by employing virtualisation and reduced energy consumption.

“A critical part of our transformation involved forging partnerships with the world’s major hyperscalers to help increase operational efficiency.”

3 – Designing energy-efficient generative AI

The blog says today’s existing efficiency improvements will not be enough to mitigate the increasing energy demands of generative AI.

It adds: “That’s why we need to harness this technology to help manage its own energy usage.

“Whether through virtualisation, optimised coding practices, cloud migration or more efficient Application Programming Interfaces (API), having visibility into energy usage will equip organisations with the insights they need to monitor and leverage generative AI for sustainability.”

4 – Seeking clean energy sources

LIz and Avintha argue that the future of generative AI will be determined by how well organisations collaborate to leverage cleaner, more renewable energy sources.

They say: “Without this ability to identify ways to conserve more energy than it uses, much of its promise will escape our grasp.

“Designing new types of batteries for backup, running generators on biodiesel and shifting computational workloads to cleaner energy sources will be part of what must be an ‘all of the above’ approach to developing sustainable generative AI.”

******

Make sure you check out the latest edition of Sustainability Magazine and also sign up to our global conference series - Sustainability LIVE 2024

******

Sustainability Magazine is a BizClik brand

******