Article

Greening Data Management for AI

Posted May 23, 2022 | Sustainability | Technology | Amplify
greenAI

AMPLIFY  VOL. 1, NO. 5
  
ABSTRACT

Rohit Nishant and Thompson S.H. Teo discuss how to limit the negative impacts of artificial intelligence (AI) adoption by using concepts from both regenerative and doughnut economics. These two approaches seek to reconstruct economic systems so they operate within the sustainable operating limits of natural systems. Pointing out that AI adoption threatens to exceed sustainable boundaries by increasing aggregate demand for energy and new materials, Nishant and Teo put forth a 3Rs framework with which AI adopters can keep the impacts of AI within sustainable boundaries.

 

With an increased focus on climate change, green IT (environment-friendly and sustainability-related technologies) has gained prominence. This includes green data centers with low carbon footprints and environmental management systems to help firms “green” their operations.

Artificial intelligence (AI) that can help firms measure their environmental footprint or optimize their energy consumption and carbon footprint are a recent green IT addition. Because AI is a disruptive technology based on to its ability to learn, it’s expected to bring impactful sustainability transformation.1 However, achieving such transformation will require a significant change in the system surrounding AI.2

In this article, we focus on the data centers forming the foundational infrastructure for data-intensive AI and the system surrounding AI. Specifically, the proliferation of personal and business AI applications and consumption of data are interacting with each other and fueling data center growth. We discuss that system — primarily an extension of the system surrounding traditional IT — and its emerging challenges. We then discuss how a new system rooted in regenerative economics and doughnut economics can help AI achieve its potential of facilitating sustainability transformation and present a framework aimed at bringing about the necessary systemic changes.

AI’s Data-Centric System

The current discourse on AI emphasizes data as the key driver of the economy. We see glimpses of this discourse in phrases such as “The world’s most valuable resource is no longer oil, but data.”3 Growth in AI is accelerating the growth of data centers, which store massive amounts of data. Hyperscale data center revenue is expected to exceed US $60 billion in the next five years.4 These data centers are the major sources of carbon footprint and are detrimental to environmental sustainability. Consequently, companies that deploy AI are faced with serious environmental challenges.

Recent studies have found that data centers’ carbon footprints are exceptionally high.5 An International Energy Agency estimate pegs their energy consumption at approximately 200 terawatt-hours of electricity. That equates to nearly 1% of global electricity demand or 0.3% of all global CO2 emissions.6

Companies are aware of this and are responding to the issue in a number of ways. Some companies are leveraging green data centers with lower energy use and carbon footprints.7 Some are switching to renewable energy to power their data centers. Companies like Amazon, Google, and Microsoft that are huge corporate purchasers of renewable energy are developing onsite renewable energy sources and forming partnerships with green vendors.8

Of course, these initiatives have limitations. For instance, using energy-efficient data centers might be ineffective if data consumption increases, negating the benefits from energy efficiency. The negation of benefits from energy efficiency improvement due to increased consumption (termed “Jevon’s paradox”) has been observed in contexts like steam engines.9 Increased data consumption can also limit the effectiveness of renewable energy in reducing data centers’ carbon footprint.10

In particular, renewable energy in data centers faces several challenges: managing real-time balancing of energy demand and supply; fluctuation of data center load between peak and normal periods; and properties such as intermittency, variability, and non-dispatchability of renewable power generation.11

It can also involve tradeoffs that pose challenges to sustainability. For instance, using a data center with solar energy can reduce a data center’s footprint, but solar panel manufacturing creates many environmen­tally harmful byproducts.12 Such tradeoffs, if ignored, can make data centers unsustainable. Clearly, increas­ing data consumption can lead to increased energy and carbon footprint both directly and indirectly, create challenges for sustainability, and make green initiatives ineffective or limit their effectiveness.

Our consumption of goods and services keeps our economic engine running, leading to high economic growth levels. IT systems like those that inform firms about their carbon footprint and aid in formulating reduction strategies form the part of this system aimed at regular growth in product and service delivery.13 Even initiatives such as energy analytics, in which energy consumption data is leveraged to formulate ways to improve output relative to energy consumed, aim to achieve growth in product and service delivery, albeit less energy-intensive.14

This calls for a shift toward a greener, more sustainable data-centric system. Specifically, a holistic view of the interactions and interdependencies among various components (rather than a focus on individual com­ponents) is needed because a positive change in one component could negatively affect other components.

Systems Perspective in Managing AI Carbon Footprint

Recent years have brought discussions about new economic systems such as regenerative capitalism and doughnut economics. Regenerative capitalism, a phrase coined by John Fullerton, refers to business practices that restore and build instead of exploiting and destroying.15 It is insufficient to stop exploiting and destroying our habitat — we must regenerate what is lost. In the context of environmental sustainability, this means looking beyond achieving net-zero emissions to having a net-positive impact.

Regenerative capitalism emphasizes the importance of creating a positive impact. Doughnut economics espouses similar ideas but is more encompassing, proposing a framework with a doughnut-shaped area representing a socially just and environmentally safe space for humanity.16 The outer edge of the doughnut depicts the limits of what the planet can endure (e.g., climate change, ocean acidification, chemical pollution, bio­diversity loss, air pollution, ozone layer depletion). The center of the doughnut houses social foundation items like water, food, energy, education, social equity, and health.

Doughnut economics discourages endless GDP growth. It views the economy as connected systems embedded within the larger society such that focusing on individ­ual units or actors is not the best approach. Instead, cooperation and altruism are essential to ensure there is no shortfall in the social foundation and no pressure against the ecological ceiling.

Effectively managing AI’s carbon footprint would require a shift to a system like regenerative capitalism or doughnut economics that does not emphasize continuous growth or increased consumption.

However, the novel opportunities AI offers society make it difficult for many to accept the idea that data consumption related to AI must be managed. For example, data access and sharing create economic opportunities by enabling everyone to produce their own content and leverage a variety of platforms to access new markets.

The framework shown in Figure 1 acknowledges the benefits of data to various sections of society and presents ways to bring about necessary systemic changes. The three pillars of the framework target the consequences of data centers (first R), data production (second R), and data consumption (third R).

Figure 1. The 3R framework for systemic change in managing carbon footprint
Figure 1. The 3R framework for systemic change in managing carbon footprint
 

Regenerate Economic Activities to Productive Use (First R)

The first pillar, regeneration, is grounded in the premise that economic activities should not destroy nature. In the context of data centers for AI, this means complementing existing initiatives like renewable energy or energy-efficient data centers with those focused on regeneration.

For instance, regardless of the type of energy used by data centers or their energy-efficiency level, they generate heat. Companies can capture this heat and use it to, for example, fuel an in-house greenhouse to grow vegetation. Studies by Béla Waldhauser and others provide guidance on these types of projects.17 Lately, companies including Facebook have set up data centers in cold locations, where heat recovery and subsequent use for greenhouse farming can help increase food production, contributing to food security.18, 19

By viewing data center heat generation and its subse­quent use in a greenhouse as interconnected compo­nents, companies can bring a systems perspective to their operations. This approach leads to locating data centers in a natural environment that can benefit from the consequences of their use.

Rationalize Development of Data-Intensive Products (Second R)

The second pillar is rationalization, which includes the idea of “create to regenerate” and “embedded economy.” The latter conceptualizes economy as embedded in the flow of energy and materials.20 In a data context, this translates into rationalizing the development of data-intensive products.

Traditionally, companies are always scouting for new product ideas and data; rarely are their associated carbon footprints explicitly considered. Instead, companies could critically examine the market for their AI products and their intended use. They could consider sacrificing products intended solely for entertainment in favor of data-intensive products that serve a larger social purpose or address a societal problem and complement its development with the use of renewable energy or energy-efficient data centers. Companies can follow the regeneration pillar to compensate for the product’s carbon footprint.

Some might point out that not every product can serve a larger purpose and that entertainment is not void of purpose. This is a valid point, and companies can address this by comparing the purpose served by various products and prioritizing those with the maximum potential (directly or indirectly) to serve a larger purpose. Companies can involve their various stakeholders in making such comparisons.

Companies could, for example, shy away from developing AI products for markets where several products providing a similar level of entertainment already exist. The rationalization pillar thus comple­ments the mindful consumption view increasingly discussed in public discourse with mindful produc­tion.21 Mindful production helps companies minimize adverse environmental consequences from existing product or service delivery processes, such as the manufacturing of solar cells.

Companies can also aim their resources at developing environmentally friendly processes for developing data-intensive products and services. Again, such an approach would connect data centers, data users/consumers, energy generation, and associated manu­facturing, allowing companies to view themselves not as an isolated entity but as constituents of a system. This realization could help them be more environ­mentally responsible while making AI greener and more sustainable.

Responsibility to Manage Data Consumption (Third R)

The third pillar emphasizes that companies must be more socially responsible, finding creative ways to reduce their product-related carbon footprint (other than technical solutions, such as using renewable energy or improving energy efficiency). Companies can leverage their consumers in this endeavor. This approach is based on viewing things from a systems perspective and the idea of viewing humans as social and adaptable.22

As AI and digitization grow more important to lives, we are increasingly consuming and creating data through interactions with various applications and chatbots. However, even as humans continue to consume data, they seek an escape from continuous data consumption. Companies thus have an oppor­tunity to play a role by increasing consumer awareness of how data-intensive products and services influence consumers’ lives.

For instance, gaming companies can make gamers aware of the potential adverse mental and social impacts of excessive online gaming.23 Companies could explore various online features to help curb such behavior. This may sound like a counterintuitive thing for a gaming company to focus on, but positioning a company as socially responsible is a powerful tool. This position incorporates the idea of mindful consumption, thus complementing mindful production (see second R, above).24 The connection between mindful consumption and mindful production allows companies to take a systems perspective, eventually leading to optimization of data consumption. Companies can then leverage the 3Rs to effectively manage their data centers’ carbon footprint.

The 3Rs framework presents an alternate system grounded in regenerative capitalism and doughnut economics as a way to reduce the carbon footprint of data. Together, the three pillars convey the message that existing initiatives like renewable energy use and improvement in energy efficiency will be inadequate if we continue to emphasize consumption in general and data consumption in particular. In such a system, data infrastructure would not be able to green itself, and goals such as net-zero and net-carbon-positive would be difficult to achieve. It’s clear that continuing our existing system is not a realistic approach and that a system and mindset change is the best way to effectively tackle the environmental sustainability challenges we face.

References

1Vinuesa, Ricardo, et al. “The Role of Artificial Intelligence in Achieving the Sustainable Development Goals.” Nature Communications, Vol. 11, No. 233, 13 January 2020.

2Galaz, Victor, et al. “Artificial Intelligence, Systemic Risks, and Sustainability.” Technology in Society, Vol. 67, November 2021.

3The World’s Most Valuable Resource Is No Longer Oil, But Data.” The Economist, 6 May 2017.

4Global Market Insights, Inc. “Hyperscale Data Center Market Revenue to Cross USD 60 Bn by 2027.” Cision PR Newswire, 15 December 2021.

5Batmunkh, Altanshagai. “Carbon Footprint of the Most Popular Social Media Platforms.” Sustainability, Vol. 14, No. 4, 15 February 2022.

6Ezra, Asaf. “Renewable Energy Alone Can’t Address Data Centers’ Adverse Environmental Impact.” Forbes, 3 May 2021.

7Green Data Centers: 8 Companies Doing Them Right.” InformationWeek, 5 August 2016.

8Ezra (see 6).

9Nishant, Rohit, Thompson S.H., Teo, and Mark Goh. “Energy Efficiency Benefits: Is Technophilic Optimism Justified?IEEE Transactions on Engineering Management, Vol. 61, No. 3, August 2014.

10Ezra (see 6).

11Rostirolla, Gustavo, et al. “A Survey of Challenges and Solutions for the Integration of Renewable Energy in Datacenters.” Renewable and Sustainable Energy Reviews, Vol. 155, March 2022.

12Masili, Alice. “The True Carbon Footprint of Photovoltaic Energy.” Only Natural Energy (ONE), 16 July 2018.

13Nishant, Rohit, Thompson S.H. Teo, and Mark Goh. “Do Shareholders Value Green Information Technology Announcements?Journal of the Association for Information Systems, Vol. 18, No. 8, 31 August 2017.

14Melville, Nigel P. “Information Systems Innovation for Environmental Sustainability.” MIS Quarterly, Vol. 34, No. 1, March 2010.

15Stokel-Walker, Chris. “What Is Regenerative Capitalism and Why Is It Important?” World Economic Forum, 24 January 2022.

16Raworth, Kate. “A Safe and Just Space for Humanity: Can We Live Within the Doughnut?” Oxfam, February 2012.

17Waldhauser, Béla. “Heat Recovery from Data Centers: A Win-Win Situation.” Dot Magazine, November 2019.

18Dawn-Hiscox, Tanwen. “Facebook Plans Third Data Center in Luleå, Sweden.” Data Center Dynamics (DCD), 8 May 2018.

19Ljungqvist, Hampus Markeby, et al. “Data Center Heated Greenhouses, A Matter for Enhanced Food Self-Sufficiency in Sub-Arctic Regions.” Energy, Vol. 215, Part B, 15 January 2021.

20Raworth, Kate. Doughnut Economics: Seven Ways to Think Like a 21st-Century Economist. Chelsea Green Publishing, 2017.

21Milne, George R., Francisco Villarroel Ordenes, and Begum Kaplan. “Mindful Consumption: Three Consumer Segment Views.” Australasian Marketing Journal, Vol. 18, No. 1, 1 February 2020.

22Raworth (see 20).

23Dastoor, Vaspaan. “Mental Health Foundation Calls on Games Industry to Do More.” TheGamer, 28 February 2022.

24Milne, et al. (see 21).

About The Author
Rohit Nishant
Rohit Nishant is Associate Professor in the Department of Management Information Systems at Université Laval, Canada, and a Cutter Expert. His research has been published/accepted in several international journals, including MIS Quarterly, MIS Quarterly Executive, Journal of the Association for Information Systems, Journal of Strategic Information Systems, Information Systems Journal, Decision Sciences, and IEEE Transactions on Engineering… Read More
Thompson S.H. Teo
Thompson S.H. Teo is Professor in the Department of Analytics and Operations at National University of Singapore Business School. His research interests include the adoption of IT, strategic IT planning, electronic government, knowledge management, and sustainability. Dr. Teo has published more than 250 papers in international journals and conferences. He has served as Senior Associate Editor for European Journal of Information Systems, Regional… Read More