Articles

SUMMER 2014 3Copyright © 2014, Association for the Advancement of Artificial Intelligence. All rights reserved. ISSN 0738-4602

Sustainable development — development that meets theneeds of the present without compromising the abilityof future generations to meet their needs (United

Nations Environment Programme, 1987) — is a critical con-cern to current and future generations. The emerging inter-disciplinary field of computational sustainability (Gomes2009) draws techniques from computer science, informationscience, mathematics, statistics, operations research, andrelated disciplines to help balance environmental and socioe-conomic needs for sustainable development. Artificial intel-ligence (AI) techniques play a key role in computational sus-tainability research, enabling the solution of sustainabilityproblems that involve modeling or decision making indynamic and uncertain environments. In turn, sustainabili-ty problems present unique challenges that further advancethe state of the art of AI.

Since 2011, the main AAAI conference has included a spe-cial track on computational sustainability, encouraging AIresearch in this area and broader participation of sustainabil-ity researchers in the AAAI community. The InternationalJoint Conference on Artificial Intelligence (IJCAI) included

Editorial Introduction

Computational Sustainability

Eric Eaton, Carla Gomes, Brian Williams

n Computational sustainability prob-lems, which exist in dynamic environ-ments with high amounts of uncertain-ty, provide a variety of unique chal -lenges to artificial intelligence researchand the opportunity for significantimpact upon our collective future. Thiseditorial introduction provides anoverview of artificial intelligence forcomputational sustainability, andintroduces the next two special issuearticles that will appear in AI Maga-zine.

Articles

4 AI MAGAZINE

an equivalent track in 2013. Between these confer-ences, the number of publications on sustainabilityand AI has grown from 18 published papers in 2011to 42 papers in 2013. These statistics do not includethe numerous other papers on sustainability and AIpublished in the International Conference on Com-

putational sustainability or in numerous other work-shops and symposia on sustainability.

The next two issues of AI Magazine will highlightrecent AI research in computational sustainability,with an emphasis on projects that have had measur-able impact to practical sustainability problems.

The Three Pillars of SustainabilityThe Three Pillars of SustainabilityThe Three Pillars of SustainabilityThe Three Pillars of Sustainability

Figure 1. The Three Pillars of Sustainability.

Sustainable solutions must balance between environmental, societal, and economic demands (United Nations GeneralAssembly 2005). Together, these interdependent aspects are known as the three pillars of sustainability. These pillars havecomplex interactions at multiple spatiotemporal scales, and so must be analyzed and managed from both local and globalperspectives across many contexts. These three pillars also have a nested relationship, with the life-sustaining environmentsupporting society, which in turn supports the economy (Griggs et al. 2013).

Articles

SUMMER 2014 5

2011 2012 2013 Conference and Track

AAAI-11 Computational

Sustainability and AI

AAAI-12 Computational

Sustainability and AI

AAAI-13 Computational

Sustainability and AI

IJCAI-13 AI and

Computational Sustainability

Number of Accepted Papers

18 21 16 26

Table 1. Publishing Statistics of Special Tracks on Computational Sustainability at Recent Major AI Conferences.

These articles span a large variety of AI techniquesand sustainability problems, from learning large-scale spatiotemporal models for bird migration, tooptimal control for energy storage, to constraint-based interactive social policy planning. Together,these articles provide a snapshot of the field that wehope will encourage further participation in compu-tational sustainability.

Overview of The Issues Problems in sustainability are inherently interdisci-plinary, requiring that any solution balance betweenenvironmental needs, societal demands, and eco-nomic constraints (figure 1). These problems existacross multiple scales in highly dynamic domainswith high levels of uncertainty. AI methods for mod-eling and decision making can enable the creation of

optimal, or near-optimal, policies for sustainabledevelopment, but the multiscale, dynamic, anduncertain aspects can present significant computa-tional challenges. Additionally, sustainability solu-tions must balance between the needs of individualsand groups that interact in highly interconnected,complex manners.

The issues feature articles that discuss problemsacross a variety of key computational sustainabilityresearch areas, including conservation planning,species distribution and ecological modeling, envi-ronmental monitoring and assessment, policy plan-ning, health, agriculture, transportation, and energyand the smart grid.

Additionally, these articles span a wide variety ofAI techniques and topics, from machine learning andoptimization to agent-based modeling and mecha-nism design. Figure 2 depicts the relationship

Active

Info

rmat

ion

Gathe

ring

Sequ

entia

l Dec

ision

Mak

ing

Stoc

hasti

c Opt

imiza

tion

Uncer

taint

y

Prob

abilis

tic G

raph

ical

Mod

els

Ense

mble

Met

hods

Citizen

Scien

ce

Spat

iotem

pora

l

Mod

eling

Rem

ote S

ensin

g

Info

rmat

ion Re

triev

al

Vision

+ Le

arnin

g

Crowds

ource

d Dat

a

Agent

-bas

ed M

odeli

ng

Constr

aint-b

ased

Reas

oning

Game T

heor

y &

Mec

hanis

m D

esign

Krause et al.

Powell

Krause et al.Conservation &Urban Planning

Environmental Monitoring& Assessment

Health

Agriculture

Transportation

Energy andThe Smart Grid

Species DistributionModeling

Policy Planning

Farnsworth et al.Fink et al.

Quinn et al.

Milano et al.

Figure 2. Article Topic Summary.

Summary of the primary computational sustainability topics (rows) and AI-related topics (columns) discussed in each article, showing thediversity of interconnections between AI and computational sustainability research. Each article in these special issues is listed by theauthors’ names. This figure depicts only the primary topics for each paper; articles may include secondary discussions of topics that are nothighlighted (for example, all articles include some aspect of uncertainty, but may not focus on it).

Articles

6 AI MAGAZINE

between the computational sustainability topics andAI-related topics across all articles, demonstrating thediversity of interconnections between sustainabilityand AI.

Many of these articles also discuss issues of broad-er interest that span different AI techniques andapplications. Some of these broader issues includecompensating for skewed data distributions incrowdsourced data and citizen science (Fink et al.),embracing the unique research opportunities andchallenges associated with problems in developingnations (Quinn, Frias-Martinez, and Subramanian),exploiting structural properties such as submodular-ity to produce efficient solutions with optimalityguarantees (Krause, Golovin, and Converse), andunifying problem formulations for stochastic opti-mization across different subfields, including rein-forcement learning, optimal control, and stochasticprogramming (Powell).

In the following subsections, we briefly describeeach article that will appear in this and the subse-quent issue of AI Magazine.

The EnvironmentSequential decision problems that exhibit adaptivesubmodularity in their structure can be solved effi-ciently using greedy policies with provable near-opti-mality. The article Sequential Decision Making inComputational Sustainability Through Adaptive Sub-modularity by Andreas Krause, Daniel Golovin, andSarah Converse describes the use of adaptive sub-modularity for decision making in uncertain, partial-ly observable environments,. It focuses on two sus-tainability problems: optimally gatheringinformation for decision making in adaptive conser-vation management, and dynamically selecting landpatches for species conservation in WashingtonState’s South Puget Sound region.

In Crowdsourcing Meets Ecology: HemispherewideSpatiotemporal Species Distribution Models, DanielFink, Theodoros Damoulas, Nicholas E. Bruns, FrankA. La Sorte, Wesley M. Hochachka, Carla P. Gomes,and Steve Kelling describe an ensemble method forlearning multiscale spatiotemporal models thatadapts to variations in spatial sampling densities.They applied this approach to learn species distribu-tion models from eBird data, yielding the first hemi-spherewide models of bird migration patterns. Thisarticle also addresses issues of citizen science andlearning from crowdsourced data.

In addition to meteorological monitoring, weath-er radar provides a remote sensing platform fordetecting the movement of birds, bats, and insects.Andrew Farnsworth, Daniel Sheldon, Jeffrey Gee-varghese, Jed Irvine, Benjamin Van Doren, KevinWebb, Thomas G. Dietterich, and Steve Kellingdescribe, in Reconstructing Velocities of MigratingBirds from Weather Radar — A Case Study in Com-putational Sustainability, a Bayesian approach for

learning large-scale models of bird migration fromradar data, advocating the benefits of joint inferenceover stage-based pipelined AI systems. The resultingmodels of large-scale nocturnal bird migration reveala variety of insights at various spatial and temporalscales that are of interest to ornithologists, ecolo-gists, and policy makers.

EnergyThe article Energy and Uncertainty: Models andAlgorithms for Complex Energy Systems by WarrenB. Powell investigates different sources of uncertain-ty across various problems in sustainable energy sys-tems, while describing a perspective on stochasticoptimization that unifies ideas from different fields,including reinforcement learning, optimal control,approximate dynamic programming, and other areasof operations research. These concepts are illustratedthrough their application to modeling and policy-based control of a solar energy storage system.

Policy Making and DevelopmentIn their article Sustainable Policy Making: A StrategicChallenge for Artificial Intelligence, Michela Milano,Barry O’Sullivan, and Marco Gavanelli describeaspects of the e-Policy project, a decision support sys-tem for policy makers that integrates global and indi-vidual perspectives on economic, social, and envi-ronmental impacts of different decisions. The articleposes policy making as a multifaceted domain con-taining a variety of challenging AI-related problems,from integrating and balancing between multiplecompeting objectives, to impact assessment usingagent-based modeling, to the use of game theory andmechanism design for sustainable policy creation.

Sustainability problems in the developing worldhave a variety of unique requirements, stemmingfrom such challenges as scarce resources and lack ofinfrastructure. In their article article ComputationalSustainability and Artificial Intelligence in the Devel-oping World, John Quinn, Vanessa Frias-Martinez,and Lakshminarayan Subramanian examine the useof mobile devices, social computing, and multiple AImethods in various sustainability applications, witha focus on developing-world issues. These applica-tions range from automated microscopy-based diag-nosis of malaria, to cropland monitoring and diseasespread tracking using remote sensing, to preventingtraffic congestion in developing cities.

Making an Impact on Our FutureThese articles are only a sample of current research incomputational sustainability and AI; there are a largenumber of other current projects applying AI to sus-tainability problems, and a wealth of sustainabilityproblems that have yet to be addressed. One hall-mark of sustainability research is the focus on havinga measurable impact on our collective future by

Articles

SUMMER 2014 7

addressing current, important problems. To ensurethat this research has practical significance, manyresearch groups have partnered with regional gov-ernment offices (such as the U.S. Fish and WildlifeService) to deploy the developed technology insmall-scale studies. In other cases, AI technology forsustainability is beginning to reach consumers direct-ly through commercial applications (for example, theGreen Driver project and smartphone app [Apple etal. 2011]). Due to the interdisciplinary nature of sus-tainability problems, computational sustainabilityresearch is also injecting computational thinkinginto other fields and fostering the cross-fertilizationof ideas. Within the AI community, we hope thatcontinued research in this area will help broaden theAAAI community, while providing a rich source ofimportant new problems to further advance the fieldof artificial intelligence.

ReferencesApple, J.; Chang, P.; Clauson, A.; Dixon, H.; Fakhoury, H.;Ginsberg, M.; Keenan, E.; Leighton, A.; Scavezze, K.; andSmith, B. 2011. Green Driver: AI in a Microcosm. In Pro-ceedings of the Twenty-Fifth AAAI Conference on Artificial Intel-ligence. Palo Alto, CA: AAAI Press.

Gomes, C. 2009. Computational Sustainability: Computa-tional Methods for a Sustainable Environment, Economy,and Society. The Bridge 39(4): 5–13.

Griggs, D.; Stafford-Smith, M.; Gaffney, O.; Rockström, J.;Öhman, M.; Shyamsundar, P.; Steffen, W.; Glaser, G.; Kanie,N.; and Noble, I. 2013. Policy: Sustainable DevelopmentGoals for People and Planet. Nature 495: 305–307.dx.doi.org/10.1038/495305a

United Nations Environment Programme. 1987. Our Com-mon Future: Report of the World Commission on Environ-ment and Development. Annex to General Assembly docu-ment A/42/427, Development and InternationalCooperation: Environment. Report of the World Commis-sion on Environment and Development. Nairobi, Kenya.New York: United Nations.

United Nations General Assembly. 2005. Resolution adopt-ed by the General Assembly on 16 September 2005: 2005World Summit Outcome, Resolution A/60/1, adopted by theGeneral Assembly on 24 October 2005. New York: UnitedNations.

Eric Eaton is a lecturer in the Department of Computer andInformation Science and a member of the General Robotics,Automation, Sensing, and Perception (GRASP) laboratory atthe University of Pennsylvania. Prior to joining Penn, hewas a visiting assistant professor in the computer sciencedepartment at Bryn Mawr College and a senior research sci-entist in the Artificial Intelligence Lab at Lockheed MartinAdvanced Technology Laboratories. His research focuses onlifelong machine learning, knowledge transfer, and interac-tive AI.

Carla Gomes is a professor of computer science and thedirector of the Institute for Computational Sustainability atCornell University. Her research themes include constraintreasoning, mathematical programming, and machine learn-ing for large-scale combinatorial problems. Recently, Gomes

has helped found the new field of computationalsustainability, which is her current main researchfocus. Gomes is a fellow of AAAI and a fellow of theAmerican Association for the Advancement of Sci-ence (AAAS).

Brian Williams leads the Model-Based Embeddedand Robotic Systems group within the ComputerScience and Artificial Intelligence Laboratory(CSAIL) at the Massachusetts Institute of Technolo-gy. His research concentrates on model-basedautonomy, model-based programming, and cooper-ative robotics. He is a fellow of AAAI, has served asguest editor of the Artificial Intelligence Journal andhas been on the editorial boards of the Journal ofArtificial Intelligence Research and The MIT Press.

Support AAAI Programs!

Thank you for your ongoing sup-port of AAAI programs throughthe continuation of your AAAImembership. We count on youto help us deliver the latest infor-mation about artificial intelli-gence to the scientific communi-ty, and to nurture new researchand innovation through ourmany conferences, workshops,and symposia. To enable us tocontinue this effort, we invite youto consider an additional gift toAAAI. For information on howyou can contribute to the openaccess initiative, please seewww.aaai.org and click on“Gifts.”