Mississippi River floodwaters, July 20, 1951, Gerald R. Massie Photograph, box 54, folder 280, Division of Commerce and Industrial Development, RG104.1; Missouri State Archives, Jefferson City.¹

A Digital and Green Transition Series: Will Artificial Intelligence Foster or Hamper the Green New Deal?²

By Dr. Theodora Dryer

Dr. Theodora Dryer, who leads AI Now’s Climate and Water research, testified before the European Parliament on February 3rd, 2021, covering the complex relationship between AI, climate policy, and possibilities for a just and liveable future. This is in response to two of the European Commission’s key priorities for the upcoming years to “accelerate innovation and digitalisation” while at the same time “reaching climate neutrality and high environmental standards.”

The European Commission posed the overarching question: How can AI be deployed to benefit society, advance research, and accelerate our climate transition without its impact hampering our environmental goals? I argue that before exploring the possibilities of AI and the environment, we should clarify what is meant by AI, which frameworks are used to assess AI, and who primarily experience the harms of AI and climate change. Here I share the key takeaways and open questions that emerged from my testimony:

  • We must ask: Who is developing this technology, for whom and for what, and with what data? Who profits from this technology? And then as a corollary: Who holds sovereignty and ownership rights to the data, and what is at stake, and for whom, in the resulting applications?

Bio: Dr. Theodora Dryer is a historian of computing and technology and STS scholar. She is a research assistant professor at New York University and leads Climate + Water research at the AI Now Institute. She has worked for the past decade on information and algorithmic decision systems as they relate to environmental and economic power. She specializes in weather and climate modeling, water policy, and agriculture.

The original questions offered by the European Commission appear as section headers, organizing my extended written testimony printed in full below.³

{European Parliament Question 1: What are the Possibilities that AI offers for the Environment?}

Before interrogating the relationship between artificial intelligence (AI) and the environment, we should clarify what is meant by AI {The Terms}, which frameworks are used to assess AI {Frameworks}, and who primarily experiences the harms of AI and climate change {For Whom}.

{The Terms}

Artificial intelligence developed in the name of benefiting the environment is not the same thing as establishing environmentally and socially conscious AI systems. It is therefore imperative to center Justice and sovereignty frameworks, rather than economic growth frameworks, in assessments of AI and environmental policy.

Artificial intelligence or AI is a powerful terminology advanced by technological developers that can refer to a broad suite of technologies, data collection and storage infrastructures, and automated decision systems. It often describes predictive statistics-based digital systems such as machine learning. Ascertaining whether or not these technological systems can improve on the climate crisis foremost requires asking precise and situated questions. Assessing the appropriateness of applying AI solutions to environmental issues would involve identifying locally situated technologies and data, examining the environments they extract data from, and understanding AI-backed decision-making procedures. It is crucial to understand how data practices and decision-making procedures impact environmental policy and resource access.

Situating AI in local contexts involves framing questions in the vein of: Who is developing this technology, for whom and for what, and with what data? Who profits from this technology? And then as a corollary: Who holds sovereignty and ownership rights to the data, and what is at stake, and for whom, in the resulting applications? Broadly stated: does this technology reinforce an extractive economic system or do these systems contribute to possibilities for environmental and climatic justice?

Activists and researchers have shown time and again that centering ‘economic growth,’ ‘technological progress,’ ‘optimization,’ and ‘efficiency’ in the development and applications of technological systems is incommensurable with possibilities for climatic and digital justice.⁴ Many have called for postcolonial, decolonial, Indigenous, and anti-capitalist approaches to environmental tech policy as these frameworks provide realistic and justice-centered alternatives to the aforementioned frameworks and cultural commitments — e.g. optimization⁵ — of extractive economic systems.⁶

By clarifying the terms⁷ — of what, exactly, is meant by artificial intelligence — it is evident that promises made that AI will ‘solve’ the climate crisis are in direct opposition to the role AI plays in perpetuating climate injustice. However, hopeful marketing and positivist policy proposals continue to promise AI as a solution in the domains of carbon tracking, agricultural production and control, weather modification, and much more.⁸ For example, Microsoft recently launched a “$1 billion climate innovation fund” to achieve carbon-negativity by 2030.⁹ Concurrent with the promise of carbon-negativity, Microsoft also recently received a $22 billion contract to supply the military with virtual reality (VR) technology.¹⁰ Military production has always been intrinsically murderous to the environment¹¹ and human life¹², and provides the human and technological infrastructures that police access to resources and safety for the populations at the frontline of the climate crisis.¹³

While some environmental benchmarks can serve as meaningful guidelines for assessment during ecological transition, they are often vaguely stated and abstracted from real political and environmental contexts. Companies — like Microsoft — often use these benchmarks to pose themselves as thriving in one singular domain (e.g., carbon), while continuing with economic growth agendas that harm the environment in many other ways (e.g., contracts with oil and gas companies, as well as the military). These statistical benchmarks alone are not adequate for determining a sound environmental justice agenda. And the reality under climate change is that we don’t have 30 years to experiment with the efficacy of these benchmarks.

{Frameworks}

I warn against utilizing frameworks of ‘progress and profit,’ ‘technological solutionism,’ and ‘sustainable development’ that have historically been used by technological developers and extractive economic interests to profit off land and people, and under the premises of climate crisis and resource scarcity.

Assessments of the potential uses of digital technology to foster Green New Deal policy should center frameworks of climate and environmental justice and digital and data justice.¹⁴ These two domains — environmental justice and digital justice — constitute decades long political and social movements and fields of research study. Taken together, they provide us with operable frameworks with which to assess the known impacts of AI technologies as they intersect with environmental and climate policy. Consideration of digital technology as it connects with environmental justice is not an arbitrary conjunction, these two domains are deeply historically linked.¹⁵

Representatives, policy makers, and researchers who are interested in understanding the nexus between digital technology and the environment are therefore tasked with learning the intersections of these two movements and engaging their locally-situated contexts.¹⁶ In this vein, I propose to reframe this question as: Is it possible for the amorphous technological powers of artificial intelligence (AI) to help realize climatic and environmental justice? And: How should we engage the nexus between climate and environmental justice and digital and data justice?

{For Whom?}

In recent decades, activists, lawyers, scholars, and advocacy groups have proven, with consensus, that the same communities at the frontlines of the climate crisis are the same communities most adversely impacted by the development of predictive technologies.¹⁷ In elucidating these realities, a coalition working on global climate justice recently described climate change as an unequally mitigated human-caused crisis:¹⁸

Climate justice perspectives center the fact that the brunt of climate change falls hardest on the most poor and marginal peoples — peoples often trampled by the twin ravages of colonialism and capitalism, who demonstrate resilience despite these depredations. The rampant extraction of resources by imperial powers in colonized lands — and subsequently by local predator elites — left the lands in a state of continuing impoverishment, and with depleted levels of physical and economic resources that make it daunting, if not almost impossible, to withstand the humanitarian and environmental crises caused by climate change.

Given the realities of the climate crisis, the frameworks guiding the terms of engagement and assessment should come from the work and voices of the people most adversely impacted and who are most knowledgeable about digital and environmental injustice.¹⁹

{European Parliament Question 2: What is the Impact AI has on the Environment?}

Artificial Intelligence or AI systems are inherently extractive from and detrimental to the environment. The popular notion of cloud computing as existing in virtual space beyond lived environmental and labor conditions is false. AI is a technological enterprise that is totally energy dependent, and that requires vast amounts of natural resources.²⁰ AI infrastructures utilize vast quantities of anthracite coal and oil materials used to fuel computing and transportation infrastructures.²¹

Digital systems also require vast amounts of water for cooling. In 2019, Google requested more than 2.3 billion gallons of water for data centers across three different U.S. states.²² After cooling usage, water becomes a repository for electronic waste and derivative toxins, making toxicity a permanent feature of surrounding systems and ecologies.²³ Google and other companies are continually expanding, meaning that their consumption of local groundwater will continue to expand with devastating and obfuscated impacts on local communities and ecosystems.

The broader systems of artificial intelligence have generated waste and toxins that have permanently altered the global landscape. These include “silicon” development in the U.S. California region²⁴ — a chemical element used in machine hardware, and namesake to dominant conceptions of Silicon Valley — as well as electronic waste (‘e-waste’) landfills like the one that surrounds Korle Lagoon, in Agbogbloshie Ghana.²⁵ Waste generated by big tech has and will continue to alter ecological and economic systems around the world, entrenching environmental racism and toxic colonialism.²⁶

As emblemized by the e-waste landfills in Agbogbloshie, the global South is carrying the brunt of the environmental impacts of AI. A recent study on the Gulf of Guinea highlights how Western waste-brokers continue to unsafely dispose of toxic e-waste in global South countries despite the existence of prohibitive laws.²⁷ Western waste-brokers’ economic and financial powers override local regulations and enforcement apparatus in places like Côte d’Ivoire, Nigeria and Ghana.

AI systems alter economic and ecological contexts in ways that directly impact the health of local communities and hinder workers’ rights.²⁸ In the greater Silicon Valley region, big tech companies including Google, Facebook, Apple, and Amazon have significantly transformed the urban and human environments of Santa Clara, San Jose, Alameda, and San Mateo counties all of which are on Ohlone land. Many activists, scholars, and lawyers have shown that the presence of these companies has had deleterious and deadly impacts on people from local communities’ health and their access to needed resources including healthcare, equitable pay, housing,²⁹ and other survival resources.³⁰

Finally, AI — which is energy dependent itself — is increasingly being used in natural resource extraction and energy production (e.g., for coal, oil, and uranium). With increasing fervor, predictive statistical systems and computer vision technologies are being developed to expand energy mining and oil drilling.³¹ This creates a destructive self-reinforcing feedback loop: mining itself has proved, across centuries, to directly contribute to toxic environments for local communities and produce inequitable impacts.³² Yet this is rarely discussed, even as my colleagues Meredith Whittaker and Roel Dobbe point out that “big AI companies are aggressively marketing their (carbon intensive) AI services to oil & gas companies, offering to help optimize and accelerate oil production and resource extraction.”³³

{European Parliament Question 3: Data is often described as key in the “AI race” and is currently fragmented, missing, and not openly available. Could you elaborate on the data governance and access conditions necessary for a sustainable AI to emerge?}

Data is never apolitical: the availability of and access to data are linked to larger systems of governance and power, and the production of data itself is often linked to extractive environmental policies.

More than a singular technology, we can think of AI as a data-driven and natural resource dependent technological infrastructure that people in power use to govern resource allocation. As such, AI depends on the mass production, extraction, and processing of environmental, climatic, and natural resource information. This data is not magically generated in a cloud. Rather, this data has a history that inextricably encodes the interests and perspectives of the actors extracting the data, and there are significant political and policy contexts behind its production and uses, especially when dealing with environmental and natural resource information.

{Data + Water Rights}

A critically important component of the Indigenous Environmental Justice movement pertains to the right to information and data that belongs to Indigenous communities, data that is critical for decision making and for informing these communities’ water and land ownership rights.³⁴ It matters who creates the data, who owns the data, and who holds power in designing data analysis systems.

In my research, especially on the uses of digital automated decision systems in water allocation and distribution policies in the U.S. southwest, it is clear that data cannot be separated from the natural resource policies and legal apparatus that it functions within. Algorithmic systems function within larger ‘computing landscapes’ that encompass particular power structures, environmental conditions, and policy contexts.³⁵ Often, these political and environmental contexts are hidden by the technical function and opacity of digital systems.

For example, I have traced algorithmic-driven water development projects in the U.S. southwest dating back a century and have uncovered the explicit ways in which algorithmic frameworks contribute to the settler colonial function and environmental racism of water policy in the region. This is to say that the disavowal of Native American water rights is literally encoded in the technical function of U.S. state-run automated decision systems, many of which grew out of resource capture and allocation projects.

Unsurprisingly, in the U.S. southwest, water and climatic data is produced and monopolized by the neocolonial state and federal institutions, especially the Department of Interior. And many of these institutions first ballooned in power during the original New Deal era. Policy and technology function together in shaping the landscapes of data availability, and shaping what questions data answers, and whose interests are served in the course of its being processed through automated decision systems.

{Data + Carbon Tracing}

Questions regarding carbon tracing data are complex and variegated. In terms of the core benchmarks of the Green New Deal — to achieve carbon neutrality and cut down on emissions — this requires a particular ambition to track and map the relevant information. Decisions as to whether to use AI in this endeavor should follow after a process of reflexivity on the political and technical frameworks that define the concept of carbon neutrality, ensuring that these concepts are anchored in the lived reality of the populations most at risk of harm from climate crisis.

Historically, emissions data has been collected for assessing shared and individual responsibility for climate impact across nation states following major establishments such as the Intergovernmental Panel on Climate Change (IPCC) in 1989 and the Kyoto Protocol following that. The mapping, tracking, and data production involved in producing these emissions numbers and information is complicated by much deeper geopolitical issues, and cannot be accepted at face value as an indicator of climate impact or accountability for that impact.

For example, the United States has produced and continues to produce more than its fair share of carbon emissions. And significantly, the U.S. has monopolized the policy frameworks for international agreement and indeed the benchmarks by which climate impact is assessed. This is to say that the data production required for emissions and carbon tracking, and to track the progress of these benchmarks, is itself not a politically neutral endeavor.

{European Parliament Question 4: Relation between AI and “Sustainability”: Why should we not only focus on climate change, but also on the broader ecosystem, such as water and agriculture?}

I have two thoughts on this, the first pertaining to sustainability and the second to water and agriculture. With each of these, I believe it is imperative to center Justice frameworks in considering assessment and impact statements pertaining to the role of digital technology in the Green New Deal.

“Sustainability” is a term with a long history and in the current policy discourses: it holds statistical meanings and functions as a benchmark in data analysis. In recent years, there have been a number of studies conducted on the “role of artificial intelligence in achieving Sustainable Development Goals.³⁶ There is danger in these types of usages that abstract sustainability — as a statistical measure — from consideration of the historical and environmental contexts behind its development and use. This use of ‘sustainability’ is similar to the use of ‘statistical bias’ in algorithmic impact statements that neglect consideration of larger political and social contexts of automated decision systems and the social beliefs baked into these measures.³⁷

Water and agriculture have been central to my work over the past decade, and my research shows just how intricately and historically linked climate injustice and tech development are. Since the eighteenth century, water management and agricultural development have been the central laboratories for the development of data-driven governance on the global stage. Technological advancements in computing and data management have always directly impacted local and Indigenous communities, immigration and agricultural labor, possibilities for food justice, and access to clean water and natural resources.

Over the last decade alone there has been rapid growth in the amorphously defined field of artificial intelligence in these two domains, employing variegated technologies in computer vision, machine learning, and automated decisions systems. While AI proponents in these domains claim that the technology aids in fulfilling sustainability goals, their systems remain unexamined in their impact on food and energy access and rights, ecological systems, and agricultural workers.³⁸

{European Parliament Question 5: How can AI help accelerate the ecological transition and help achieve our green objectives? What is the condition for AI to bring more environmental and social benefits than harm?}

The Green New Deal benchmarks and ambitions are not negotiable, they are a matter of survival. However, it is important to be cautious in how we frame ecological transitions and climate futures in shaping this policy.³⁹ Climate Justice groups are currently confronting “transition” itself as the site for engaging necessary change. The London Mining Network and War on Want coalition, for example, make a compelling and necessary argument that:⁴⁰

Mining corporations are aggressively and cynically marketing their destructive activity as a solution to the climate emergency. It’s critical that we stop extractive industries from greenwashing their crimes and capturing the narrative around the transition to renewable technologies. The climate movement must listen to and learn from frontline communities pushing back the expansion of the extractive economy: communities who are simultaneously advancing solutions that embody social, ecological and climate justice.

Furthermore, the notion of “acceleration” used in the ecological transition to achieve Green New Deal benchmarks through the development and expansion of digital technology is a dangerous framing.⁴¹ Under the urgency of acceleration and rapid economic growth, there is deep historical precedent of deferring to profiteers, technological developers, and private capitalists, and making decisions without equitable and considered consultation. This has supported these actors in going after weather, climate, and natural resources as a site for development in ways that are ultimately destructive to the environment and that perpetuate environmental racism and social inequities.

A necessary condition for AI to bring more environmental and social benefits than harm would be centering a Climate and Data Justice framework in assessments that are led by the interests and needs of Black, Indigenous, and Peoples of Color and those generally at the frontlines of the climate crisis. As discussed in these remarks, there are large networks, collectives, coalitions, and movements working at this nexus whose work and knowledge is critical to these aims.

Endnotes:

  1. Toni Morrison, Charlotte Mecklenburg Library Novello Festival, November 20, 1996: “You know, they straightened out the Mississippi River in places, to make room for houses and livable acreage. Occasionally the river floods these places. “Floods” is the word they use, but in fact it is not flooding; it is remembering. Remembering where it used to be. All water has a perfect memory and is forever trying to get back to where it was.”

Researching the social implications of artificial intelligence now to ensure a more equitable future