Nicolas Cador, MiM student and ESSEC Business School article competition finalist, navigates through the challenges Information technology faces to reduce its carbon footprint.
Is Information Technology becoming the next Oil and Gas? by Nicolas Cador.
The use of Artificial Intelligence (or AI) is growing. The FAANGs – Facebook, Amazon, Apple, Netflix, and Google – have succeeded in training models based on consumer data to innovate and create value – they are now the most valuable companies on Earth. Volume, velocity, variety, veracity and value are the five keys they have used to make energy-consuming data a huge business.
However, as the climate emergency has just been restated by the 6th IPCC Assessment report, how can we ensure that digital and AI – forecasted to account for 10% of global emissions in 2030 – do not become the most polluting industry by the end of the century?
Monitoring the carbon footprint of Information Technology
AI, thanks to its computing power, enables process optimisation: better logistics, waste reduction, and resource extraction impact computation are indeed made possible with AI. The study “How AI can enable a sustainable future” by PWC indicates that AI could reduce global greenhouse gas emissions by 4% by 2030.
But, on the other hand, the consumption of AI-driven technologies will create a high demand for energy. AI application, to be trained, relies on digital infrastructures – cloud, servers, data centres – and on a large data bandwidth. According to the International Energy Agency, data centres currently consume approximately 200 terawatt-hours (TWh), or nearly 1% of global electricity demand. By 2025, average estimates suggest that energy consumption will increase by a further 25%.
Moreover, the rise of cloud computing, the metaverse and new IoT services reinforce the probability of an energy-intensive evolution scenario. But direct and indirect global digital carbon impact is systematically subject to very wide estimates, given that taxonomy varies between jurisdictions.
While the digital sector is constantly evolving in terms of usage and equipment, it is still difficult to assess the ongoing carbon impact accurately. Models are not refined enough and do not allow us to clearly communicate the energy efficiency of a particular process using AI. This would enable public authorities to adopt measures towards energy optimisation.
Deep Learning vs. Green Coding: A wide gap
Deep Learning, a sub-technique of artificial intelligence-based neural network models, is an even more energy-intensive practice. Its use improves learning speed: DeepMind, owned by Google, recently published an article announcing a breakthrough accelerating fusion nuclear science. (1)
But pure performance gains are achieved through larger data volumes, larger models, and ultimately more computations.
Researchers at MIT have calculated that training a deep learning model for 4 to 7 days emits as much CO2 as a human being for 57 years, or as much as 5 cars during their lifetime (2), only to generate and recognise words very close to human language (NLP) – enabling for instance a voice command to be interpreted by Google Assistant.
Of course, it should be borne in mind that training such complex models is only deployed for a small share of AI solutions. But energy optimisation is often missing from AI development projects.
Software engineers are the first to admit that they lack concrete baselines and indicators to gauge themselves. Using surveys conducted at the 38th International Conference on Software Engineering (464 software engineers), researchers found that where green coding matters most, its practice is rare. (3)
Green coding consists in ensuring that the code uses as few processor instructions and as little memory space as possible to reduce energy consumption. But mobile apps are too often developed in a rush to reach the market quickly. Poorly optimised, some can drain a battery in a day, requiring frequent recharging.
A nascent regulatory framework does not yet encourage developers to update their source codes. Companies are slowly starting to audit their software infrastructure, but competent audit services are still lacking.
Information technology: Educate algorithm programmers and consumers
To create a challenging climate for AI programmers, it is necessary to promote energy-efficient algorithms and finance R&D.
Raising awareness and training developers on these issues during their academic years is essential. As said within the latest World Economic Forum report, AI developers “must incorporate the health of the natural environment as a fundamental dimension.” And in its “Recommendation on the Ethics of Artificial Intelligence”, UNESCO states: “Member States should ensure that AI actors give preference to data-, energy- and resource-efficient AI methods.” (4)
Within the community, some good practices are starting to be shared and dedicated blog sites are emerging. To reduce the energy produced by deep learning models, researchers from the University of Montreal recommend that developers use computationally efficient techniques such as Bayesian optimisation and random optimisation. (5)
It would also be interesting to develop metrics or indexes that consumers could understand – such as Energy Rating Labels used for household appliances or buildings. The energy performance of a given application (website, applied task, API) would be quantified in kWh consumed. The benefits of such a practice would be twofold: 1) educate consumers on their digital carbon impact – pushing for a respectful use of technology (digital sobriety), 2) serve as a recognised benchmark prompting software engineers to develop better designed and less energy consuming software.
Tools do exist today: CodeCarbon seamlessly integrates into Python codebase to estimate the CO2 produced by cloud resources. But the quest for energy-efficient algorithms is still subject to the goodwill of companies. Making regular code audits compulsory under penalties would be a good way to ensure optimisation across the digital world.
While it is difficult to assess AI carbon impact on an ongoing and accurate basis, more refined calculation models and dedicated tools are needed to drive legislation. Enforcing digital solution optimisation is key, as Web 3.0 will become increasingly dominant in the global economy.
Raising awareness today about tomorrow’s AI energy challenges will drive programmers to natively design efficient solutions, while consumers will have to adopt environmentally friendly digital uses. Of course, the energy mix also needs to be further decarbonised: the carbon footprint of the digital sector could be reduced by more than 80% if all electricity consumed came from renewable energy sources.
Sources:
1] DeepMind, “Accelerating fusion science through learned plasma control”; 2] MIT Technology Review, “Training a single AI model can emit as much carbon as five cars in their lifetimes”; 3] Christian Bird, “An empirical study of practitioners’ perspectives on green software engineering”; 4] UNESCO (2021), ‘’Recommandation sur l’éthique de l’intelligence artificielle’’; 5] James Bergstra, “Random Search for Hyper-Parameter Optimization”.
Useful links:
- Link up with Nicolas Cador on LinkedIn
- Read a related article: AI: A paradigm shift in people management
- Discover ESSEC Business School and apply for the GMBA and EMBA programmes.
Learn more about the Council on Business & Society
The Council on Business & Society (The CoBS), visionary in its conception and purpose, was created in 2011, and is dedicated to promoting responsible leadership and tackling issues at the crossroads of business and society including sustainability, diversity, ethical leadership and the place responsible business has to play in contributing to the common good.
Member schools are all “Triple Crown” accredited AACSB, EQUIS and AMBA and leaders in their respective countries.
- ESSEC Business School, France-Singapore-Morocco
- FGV-EAESP, Brazil
- School of Management Fudan University, China
- IE Business School, Spain
- Keio Business School, Japan
- Stellenbosch Business School, South Africa
- Trinity Business School, Trinity College Dublin, Ireland
- Warwick Business School, United Kingdom.
