Artificial Intelligence for Computational Sustainability: A Lab Companion/Introduction

What is Computational Sustainability?

Gomes (2009) motivates and crystalizes computational sustainability thusly: "it is imperative that computer scientists, information scientists, and experts in operations research, applied mathematics, statistics, and related fields pool their talents and knowledge to help find efficient and effective ways of managing and allocating natural resources. To that end, we must develop critical mass in a new field, computational sustainability, to develop new computational models, methods, and tools to help balance environmental, economic, and societal needs for a sustainable future."(pp. 5-6, ).

The imperative to which Gomes' speaks is the need to understand and grapple with the great complexity and uncertainty of even the simplest of sustainability challenges. For example, designing and setting aside a protected chunk of land for grizzly bears may seem straightforward at first glance, perhaps requiring attention to the monetary costs of various plots of land and the protection agency's available budget, but this problem quickly becomes more complex as decision makers factor in lost opportunity costs by ranchers, residential developers, and other stakeholders who might, at a later date, fight to retract this protection and recoup opportunities; factor in the landscape and the bears' willingness and ability to live exclusively within the protected region; factor in the effects of climate change on the regional characteristics relative to the bears' desiderata and the desiderata to those plant and animal species on which the bears rely, such as cutthroat trout; factor in the size of the gene pool that the protected region can support, and the genetic mixing that can occur; and factor in that a given protected region may be a small part of a larger national plan. The relevant factors can explode, even in problems that appear, at first glance, to have small spatial and temporal scope. A hope for effectively managing this complexity is through computing and information technologies.

Designing effective solutions doesn't simply require that practitioners understand computing principles (e.g., abstraction), methods (e.g., optimization) and models (e.g., of differential equations) and that they be generally adroit at computational thinking so as to advance the state of the computational art; but computational sustainability also requires that practitioners understand those aspects of the sustainability challenges to which computation will be applied -- in reserve design this can include attention to economics (opportunity costs), human behavior (stakeholders), animal behavior, ecological dependency webs, and climatic change. It is not surprising that in the nascent field of computational sustainability, few individuals have all the requisite knowledge for even small problems, and that effective practitioners are typically (members of) highly interdisciplinary research teams, of which computational scientists are a vital part.

Importantly, the scope of computational sustainability goes beyond direct "managing and allocating natural resources," to include the development and management of human-made resources such as synthetic materials and energy production, all of which use the natural world as a source and a sink for these activities and their residue. Even computing itself -- the manufacture, use, and disposal of computing artifacts -- have an increasingly large ecological footprint, and computational methods for developing environmental and society-friendly technology through holistic, cradle-to-cradle design methodologies are a subject for computational sustainability.

The Special Place of AI in Computational Sustainability

Environmental and societal sustainability requires that we humans make choices, the premises behind and consequences of which are accompanied by many uncertainties. Very often, because the context that humans consider is very limited, there are entirely unanticipated consequences of some actions. A classic example of an unanticipated consequence is the so-called rebound effect -- making a process more efficient often results in the process being used more. In computing, for example, the more energy efficient we make computers, the more we use them -- this is at least a historical correlation, if not causal. Thus, as the per-unit(computer) energy efficiency improves, the collective energy footprint across all computers worldwide increases (refs). This kind of rebound effect is only "unanticipated" because a focus on per-unit energy efficiency doesn't factor in human and corporate behavior, at least not as we would have it, though a rebound effect, sometimes known as Jevon's paradox, seems common-sensible and commonplace enough.

Realizing long-term environmental and societal sustainability then is ultimately a challenge for individual and collective human decision making, so that it is informed, rational, timely, and long-sighted. Findings in behavioral economics, however, suggest that human decision making is, at a minimum, too myopic along dimensions important for sustaining a healthy planet. Therein lies the primary utility of AI for sustainability. As the field of computing that aims to model and mechanize thinking, AI can be used to model human thought, good and bad; or better yet, to model different thought entirely -- thought that is more evidence based and strategic, and less ego-centric and myopic.

While there have been pessimistic speculations on AI's role in society (e.g., consider HAL from science fiction), there has also been optimism, perhaps most famously represented in Isaac Asimov's Three Laws of Robotics, in which machines are designed to watch out for individual human welfare. But a worldwide artificially intelligent network, suggested in science fiction and in corporate campaigns, that looks out for humanity's collective welfare is a long-term dream at best, and then perhaps only a dream of the most committed technophiles.

Rather, the most realistic promise of AI for promoting sustainability thinking is as a partner in human decision making, with AI tools and agents designed to meet people where people are, "fit" to human limitations, but not confined by these limitations -- indeed, AI designed so that the hybrid human/AI decision maker goes well beyond the capabilities of the human alone or the AI alone. A paradigm of AI as cognitive prosthesis is already alive and well in many of the tools being created by computational sustainability. A second paradigm, also consistent with better, hybrid (collective) decision making, is AI agents serving as positive role models and collaborators in otherwise human collectives.

The purpose of this lab text is to support student exploration of sustainability-motivated AI systems, with the vast majority currently falling into the cognitive prosthesis paradigm, with expectations that numbers in the collaborative agent paradigm will increase, and with any useful distinction between the paradigms eventually disappearing.

A Brief History of Computational Sustainability

Though computational sustainability arguably crystalized around 2008 and 2009, with activity in the area taking off coincident with dedicated conferences and workshops, research into computational models and methods to address problems of environmental and societal sustainability has been going on for much longer. Interested readers are encouraged to dig into the history of computing for sustainability, creating and adding to a detailed bibliography on computational sustainability. Certainly statistical and mathematical modeling has been long used, presumably as long as sustainability itself has been a concern and as long as computing has been available. The fascinating history of climate modeling illustrates this long pairing nicely: "To be sure, the computer at Phillips's disposal was as primitive as the dishpan (its RAM held all of five kilobytes of memory and its magnetic drum storage unit held ten). So his model had to be extremely simple." In the area of social modeling, it is not a stretch to consider computational simulations on the evolution of cooperation to have direct sustainability implications (Axelrod, 1984); in fact, Axelrod's supposition that for cooperation to arise, the future must cast a sufficient "shadow" on the present is an insight that can be realized through mathematics and computation (e.g., policy learning, virtual worlds, visualization) to mitigate the myopia and egocentrism of decision making.

Within AI proper, machine learning in particular, research in the 1990s addressed such issues as ecological modeling and wastewater management. Through a variety of ancestral lines, machine learning projects continued well into the 2000s , with a Machine Learning for the Environment working group of the National Center for Ecological Analysis and Synthesis (NCEAS) begun in 2006. Research on optimization in support of sustainability challenges such as reserve design has also been active, tracing back from the 1980s to the present.

Concern with the environmental impact of computing itself began in earnest by the early 2000s (e.g., Kohler and Erdmann, 2004 ), with Green Information and Communications Technology (ICT) being championed by ICT companies. These investigations into Green ICT went beyond the immediate (or direct or 1st-order) environmental effects of manufacturing, use, and disposal of computing, and sought to categorize the 2nd-order and still higher-order environmental effects of computing on other sectors of society. Investigators would ask, for example, what the promise of ICT was for offsetting the environmental (e.g., CO2) effects of air travel through Web-based video-conferencing and the like; for intelligent scheduling software to reduce travel times and CO2 of delivery vehicles by eliminating delay-laden (at least in the US) left-hand turns from all routes; and the effects on occupant behavior of instrumenting buildings with ICT to inform occupants on energy usage. International policy consultants of the Organization for Economic Cooperation and Development (OECD), among others, focused global attention on a conceptual framework that acknowledged 1st, 2nd and 3rd order effects of ICT in 2008; with related activities continuing to the present, by OECD, the Computing Research Association through its Computing Community Consortium's (CCC's) visioning activities (e.g., on ocean observatories, energy, smart grids, and transportation), and by others.

Just preceding the OECD's 2008 conference, mathematicians forwarded plans to use their specialized expertise to address climate change, which was an important publically-visible thrust into sustainability by a research community not usually acknowledged as sustainability-science focused. Along these same lines and with an Expeditions in Computing award from the National Science Foundation (NSF), the Institute for Computational Sustainability (ICS) was established in 2008 to unite computer scientists and applied mathematicians in pursuit of solutions to difficult, long-term sustainability challenges. In 2009, special sustainability-focussed computing symposia sprang up, and the International Conference on Computational Sustainability (ICCS) was founded. The inaugural ICCS in 2009 was at Cornell University, the second ICCS in 2010 at MIT, and the third ICCS in 2012 at the University of Copenhagen. While there was no ICCS in 2011, two of conference series leaders, Carla Gomes and Brian Williams, established a special track on Computational Sustainability at the flagship conference of the Association for the Advancement of Artificial Intelligence (AAAI). Inclusion of sustainability into a mainstream and top-rated conference was an important step towards infusing sustainability into AI, computing, and the thinking of computer scientists generally. This move reflected the belief that sustainability can and should be deeply infused throughout scientific disciplines, rather than a separate discipline; this is also a motivation for this lab text.

Activity in computational sustainability continues to grow. A second Sustainability-focused Expeditions in Computing award was made to the University of Minnesota (Vipin Kumar, Principal Investigator) and its collaborators in 2010, to develop data mining and computer modeling/visualization approaches to better understand climate and other Earth dynamics. Sustainability-related tracks and awards in other conferences continually sprout up, with organizations such as the CRA/CCC supporting this growth through special awards to authors of outstanding papers and the like. In this time period too, NSF initiated large-scale,coordinated sustainability funding under the banner of Science, Engineering, and Education for Sustainability (SEES), at about 10% of NSF's budget, to include vital roles for computing. NSF has also supported smaller, but still notable efforts, such as research into artificial intelligence applications for environmentally sustainable, energy efficient, and ultimately cradle-to-cradle design.

Its an exciting time to be studying computational sustainability. Following a number of early pioneers in computational sustainability research and a swell of support in policy, corporate, and military circles for "green" ICT, the founding of an institute, conferences and conference tracks in 2008-2009 appear to have been seminal in growing computational sustainability. Nonetheless, in this still nascent field there are uncountable sustainability-related applications to be explored, with computing and mathematical theory yet to be developed. In the conceptual framework of 1st, 2nd, 3rd order effects of the OECD and others, most of the research opportunities will relate to the higher order effects of AI, for example, the 2nd order effects of intelligent planning/routing/scheduling software that reduces travel time and idling in a smart street system, all the way to the highest order effects of changing the way that humans, in conjunction with smart tools, think -- in ways that are evidence-based, strategic, long-term, and for the collective good.

How to Use this Book

This book is intended as a laboratory text, with sustainability-relevant AI exercises and projects, to accompany a standard or personalized course of study in AI. This text is not an AI textbook. The contributions of AI projects and exercises will include sustainability-relevant background as appropriate, and perhaps rudimentary AI summary material at the discretion of authors, and by editors who worry about transitions and readability. Contributions will, however, almost certainly contain pointers to the necessary AI prerequisite concepts in other sources, except perhaps in cases where part of the exercise is to determine the AI relevancy to a sustainability problem.

As you make progress through an AI course of study, look for the relevant AI topics in the table of contents and/or index, then look at the sustainability assignments that accompany this material. Inversely, should you read about a sustainability topic in another setting, likewise look it up, and see what in the AI world is relevant to it or approximations of the topic.

As a Wikibook, this text is open to change and addition by anyone, and as such adheres to most of the eleven tenets of a Sustainability Code advocated by McElroy, to include acceptance of fallibility, transparency, inclusiveness, questioning, and the growth of knowledge. The interested reader might ask, however, whether the tenet of internalization is achieved, and if so, how? Readers who feel so drawn are encouraged to read the guide for contributors and add to the text. While polished text is nice and desirable, it is ok to let others do some of the polishing -- in fact, imperfect text gives others an easy entry point for contributing to the project. This is not to suggest that a contributor deliberately make mistakes, but only that its perfectly fine at some point to let go, allowing others to help with the editing and content.

This book is expected to grow for a good, long time, possibly moving beyond a laboratory text for AI, through both evolution or reproduction.