14 1541-1672/11/$26.

00 2011 IEEE IEEE INTELLIGENT SYSTEMS

Published by the IEEE Computer Society
A I A N D S U S T A I N A B I L I T Y
Editor: Doug Fisher, Vanderbilt University, douglas.h.fsher@vanderbilt.edu
Computing and AI
for a Sustainable Future
Douglas H. Fisher, Vanderbilt University
and sustainability. My search was not exhaustive,
largely based on keywords, but it wasnt trivial ei-
ther. Still, little turned up in the intersection of AI
and sustainability in early 2007, and most of what
did, as I recall, was in environmental science pub-
lications and appeared to be dominated by Euro-
pean researchers using evolutionary computation
for the purposes of optimization.
1
AI and sustainability has grown substantially in
the last few years. To some extent, this tracks with
increasing interest in sustainability and comput-
ing more generally. However, AI is helping to drive
this larger movement, rather than simply riding
along. Indeed, its hard to imagine that AI would
not be central to understanding and managing the
great complexity of maintaining a healthy planet
in the face of pervasive and transformative human
activity.
A visible and scientifcally signifcant landmark
in this growth of AI and sustainability is the es-
tablishment of the Computational Sustainability
Institute,
2
with its focus on AI and many sustain-
ability areas, such as biodiversity and alternative
energy. The institute grew from a 2008 Expedi-
tion in Computing Award from the NSF to Cor-
nell University, Oregon State University, Bowdoin
College, Howard University, and other partners,
quickly attracting other researchers, educators,
government, and industry. The frst conference on
computational sustainability took place in 2009,
followed by a second in 2010 and leading in 2011
to a special track on Computational Sustainability
at the Association for the Advancement of Artif-
cial Intelligence (AAAI) conference.
Coinciding with the institutes founding was a
groundswell of activity to include sustainability
tracks at other AI-related conferences. Machine
learning and data mining have been strong among
these, and in 2010, a second sustainability-focused
Expedition in Computing award was given to the
University of Minnesota and its partners for data-
driven understanding of climate change and re-
lated phenomena.
Forthcoming articles in this new IEEE Intelli-
gent Systems AI and Sustainability Department
will elaborate on AIs deployment in many areas
of sustainability as well as the challenges and op-
portunities that sustainability issues bring to AI
research, education, and practice. This opening
article will touch upon the main themes at the in-
tersection of AI and sustainability, but it will pri-
marily concentrate on the larger contexts of sus-
tainability, and on computing and sustainability,
thereby setting the stage for articles to come.
Sustainability
The United Nations Bruntland report contains a
popular and succinct defnition of sustainability:
Sustainable development is development that
meets the needs of the present without compromis-
ing the ability of future generations to meet their
own needs.
3
To many, the phrase sustainable development
is an oxymoron, but the needs spoken of in the
Bruntland report are not about the luxuries of the
materially wealthy, but rather about the survival
needs of the poor and starving. As contextualized
in the report, development is about bringing all
those on the planet up to a reasonable standard of
living, rather than on those who are already us-
ing plenty. Indeed, there is a nascent AI for De-
velopment Group
4
actively exploring AIs role in
advancing social equity, together with a larger
computing for development group.
More generally, the reference to needs begs
the question as to exactly what these are, both now
and in the future. Achieving and then maintaining
W
hen preparing for a March 2007 talk
at the US National Science Foundation
(NSF), I searched the Web for scholarly work on
AI and climate change, the natural environment,
IS-26-06-Sustain.indd 14 11/10/11 10:42 AM
NOVEMBER/DECEMBER 2011 www.computer.org/intelligent 15
safe and adequate water, food, air,
and other health-related criteria for
all people are high-level goals. When
we work backward from these ulti-
mate human-centric goals, we arrive
at a large number of sustainability
desiderata relating to biodiversity, en-
ergy, toxins, climate change, disease,
community planning, agriculture,
emergency response, transportation,
garbage, materials, economics, pol-
icy, and human behavior, among oth-
ers. An expansion of what is a many-
node and densely connected graph
will defne the purview of the IEEE
Intelligent Systems AI and Sustain-
ability area. Readers are encouraged
to trace out what they imagine this
graph looks like from their own per-
spective and to ask students at all lev-
els to do so as well. What we desire to
be sustained shouldnt be simply enu-
merated for people; it should be a fo-
cus of deep, ongoing conversation.
Sustainability has received mixed
attention from academics, govern-
ments, and industries over the past
few decades. As the Brundtland re-
port indicates, many have sounded
the alarm for a good long time.
Silent Spring,
5
Rachel Carsons book
on environmental poisoning through
pesticides and the like, was published
in 1962. It is often credited for an en-
vironmental awakening, but one that
has waxed and waned over the years.
The frst scientifc reports on rising
CO
2
levels and the implications for
warming the planet were published
by the early 1960s, reaching levels of
scientifc consensus by the 1980s.
6
Nevertheless, a signifcant (but mi-
nority) proportion of Americans, to
take but one nation, arent simply
skeptics, but are dismissive
7
and eas-
ily shiftedthe materially wealthy
world in general has been slow to re-
act. So we have to wonder whether
the new initiatives will have stay-
ing power, and having staying power
through this and all subsequent gen-
erations is critical, at least if we view
the planet from the perspective of the
human time scales of decades, centu-
ries, and millennia.
In the US, a new NSF initiative
at fully 10 percent of the NSFs pro-
posed budget for the upcoming fscal
yearwill support science, engineer-
ing, and education for sustainability.
The SEES initiative follows a host of
discipline-focused programs, but now
the emphasis is squarely on the need
for strong interdisciplinary partner-
ships leading to a science of sustain-
ability.
8
The Proceedings of the Na-
tional Academy of Sciences launched
a sustainability science section and
academic departments and schools of
sustainability are springing up.
9
It is striking, however, that com-
puting is typically not a component
in these sustainability curricula, per-
haps in part, because computer sci-
entists themselves do not actively
recognize its relevance to sustain-
ability. Yet computing is pervasive
and transformative, potentially af-
fecting human behavior in disruptive
ways, so it seems wise to consider it a
core part of the emerging science of
sustainability.
Computing and
Sustainability
Sustainability science can be reason-
ably viewed as a new and vitally im-
portant discipline, but sustainability
concerns should not be stove-piped.
If we are designing a planet that sus-
tains humanity for millennia (or even
centuries and decades) at anything
like current levels, with wealth ac-
ceptably distributed, sustainability
motivated thought and action must
be at the core of everything we do
and must permeate the societal mi-
lieu. Considering that computing is
already embedded in much of society,
the prescription that sustainability
should be so embedded would result
in a frequent and necessary align-
ment of sustainability and comput-
ing. These realizations have only re-
cently started to take center stage.
In May 2008, the Organization
for Economic Cooperation and De-
velopment (OECD) hosted the Inter-
national Workshop on Information
and Communications Technology
(ICT) and Environmental Challenges
in Copenhagen, following projec-
tions on ICTs growing environmen-
tal footprint.
10,11
The workshop was
convened to share strategies on miti-
gating these footprints, to include en-
ergy, greenhouse gasses (GHG), and
waste. These direct or frst-order ef-
fects of ICTduring the use, manu-
facture, and disposal phasesare
typically detrimental.
Importantly, the speakers and
national delegations also discussed
ICTs higher-order effects, many of
which lead to decreasing ecologi-
cal footprints in other sectors such
as travel and transportation. Exam-
ples of computings second-order ef-
fects include more accurate and rapid
identifcation of species populations
through image and audio recording
and processing, static and dynamic
routing of vehicles to eliminate con-
gestion and the idle time associated
with it, the use of video-conferencing
systems instead of travel for meetings,
and proposed smart grid applications
such as electricity load balancing.
In turn, third-order effects of ICT
alter the ways that people and other
processes operate, in quality and/or
quantity, and these third-order ef-
fects can have profound effects, both
positive and negative, on ecological
footprints. For example, rebound ef-
fects occur when effciency improve-
ments in the per unit costs (such as
energy) of a process result in the in-
creased use of that process so that the
collective costs (notably energy used)
IS-26-06-Sustain.indd 15 11/10/11 10:42 AM
16 www.computer.org/intelligent IEEE INTELLIGENT SYSTEMS
becomes even greater than the collec-
tive costs before the improvements.
These rebound effects are but one ex-
ample of unanticipated (though not
necessarily unforeseeable) effects that
are detrimental to the environment.
Conversely, ICT is responsible for im-
proved data collection and evidence-
based decision making, which itself
might increase these behaviors.
8,12

The Oberlin dorm energy monitor-
ing project, for example, used com-
puting technology to visualize energy
and water usage in an attempt to alter
their human behavior.
13
The OECD 2008 and 2009 meet-
ings resulted in an important con-
ceptual framework for expressing re-
lationships between computing and
the environment.
11
Coincident with
this, mathematicians
14
and computer
scientists were hosting workshops
on sustainability themes. The NSF
awarded the Computer Science and
Telecommunications Board (CSTB)
of the US National Academy of Sci-
ences a grant to explore, organize,
and report on the opportunities for
computer science research contribu-
tions to sustainability. More recently,
in February 2011, the Computing
Community Consortium (CCC) of
the Computer Research Association
(CRA) convened a similarly intended
meeting, issuing a report on the
broad swath of challenges at the in-
tersection of computing and sustain-
ability, requiring truly interdisciplin-
ary partnerships between computing
researchers and domain scientists.
15
Taken from and nicely abstracting
the CCC report, Figure 1 highlights
the critical role that observational
data and computational models
play in sustainability science. Chal-
lenges in the future of modeling in-
clude downscali ng global mod-
els, say of climate, to better inform
regional planning and policy as well
as the integration of various sources
of information, from social and phys-
ical, to plan for the human burden
on the natural environment. Some
of these challenges might be met by
agent-based modeling approaches,
16

where agents correspond to small
regions and/or particular information
sources.
The CCC report makes recommen-
dations for computing subdisciplines,
a few of which we can highlight here.
For example, social computing is
changing the way that humans com-
municate, collaborate, compete, and
play.
17
Yet, we have not substantially
tapped into the possibilities of social
computing for advancing a sustain-
ability agenda. Encouraging new, sus-
tainable behaviors and growing col-
lective intelligence through social
networking is a goal, though fnd-
ing the incentives that will motivate
people to act is a challenge. Green IT
refers to mitigating the frst-order en-
ergy and material effects of comput-
ing due to its manufacture, use, re-
cycling, and disposal. Advances in
energy effciency and energy harvest-
ing through GHG-neutral means are
relevant. Software is also relevant in
areas such as server virtualization
and all forms of intelligent control.
In addition to research and prac-
tice, the report also stressed the im-
portance of education, in particular
the infusing of computing curricula
with sustainability, and inversely the
infusion of computation into sustain-
ability curricula. Finally, we can ex-
trapolate beyond the US context in
which the CCC report was prepared
and emphasize that government fund-
ing provides incentive for the interdis-
ciplinary and international research
collaborations necessary to advance
sustainability desiderata. These col-
laborations would not simply be be-
tween environmental scientists and
Figure 1. Many subdisciplines of computing will contribute to and will be
challenged by sustainability objectives. Achieving sustainability goals will require
that computer scientists enter into interdisciplinary collaborations with other
scientists, and vice versa, and that researchers across felds integrate their efforts
with education, development, and practice. (Based on a fgure in From Science,
Engineering, and Education of Sustainability: The Role of Information Sciences and
Engineering.
15
Used with permission.)
Big data
A
r
e
a
s

o
f

d
i
s
c
o
v
e
r
y

a
n
d

i
n
n
o
v
a
t
i
o
n
Modeling and
simulation
Human-centered and
social computing
Cyberphysical systems
Optimization
Intelligent systems
Privacy and security
Systems engineering
Collaborative, use-inspired
fundamental research
Energy Transportation Environment and climate
fundamental research
Core
Green IT
Cradle-to-cradle
design
Power-aware
computing
Energy complexity
analysis of
algorithms
Energy harvesting
Sustainability: meeting the needs of present and future generations
Stakeholders
Coordinated federal investment
Researchers
Computer scientists,
systems engineers,
social scientists
Researchers
Computer scientists
Educators
Domain experts
IT manufacturers
IT operators
Domain experts
Electrical engineers, transportation
engineers, environmental scientists,
biologists, climatologists
Stakeholders
IS-26-06-Sustain.indd 16 11/10/11 10:42 AM
NOVEMBER/DECEMBER 2011 www.computer.org/intelligent 17
computer scientists, but because hu-
mans are key to sustainability solu-
tions, there is a great need for socio-
technical sciences that anticipate,
evaluate and design cognizant of re-
bound effects
8
and other infuences
of technology on human behavior.
AI and Sustainability
Finally, we come to the area that this
article inaugurates: AI and sustain-
ability. The computing areas that I
highlighted earlier all invite AI meth-
ods to facilitate progress. In green
IT, for example, there are intelligent
controls during use phases, and plan-
ning and scheduling concerns dur-
ing manufacture, such as shortening
supply chains to reduce ecological
footprints. There is also a nascent
movement toward AI for sustainable
design, including cradle-to-cradle
design,
18
intended to eliminate waste
through low-energy reclamation
processes.
Although Figure 1 labeled intel-
ligent systems as an area distinct
from optimization and cyberphys-
ical systems (CPSs), they are not
mutually exclusive. Optimization has
a rich history both in and outside of
traditional AI boundaries. Exemplar
applications of optimization for sus-
tainability include supply-chain plan-
ning, optimal wind-farm arrange-
ment on small and large scales, and
reserve and corridor design, where
land is purchased for the beneft of
selected species under budget con-
straints.
2
We can imagine that in each
of these examples, climate change
(whether the reader believes it is hu-
man caused or not) will alter what is
optimal, and thus characterizing so-
lution robustness and adapting solu-
tions in the face of change are impor-
tant challenges.
Machine learning is another im-
portant methodology for sustainabil-
ity. Machine learning methods are
used for such varying applications
as learning to identify and count in-
dividuals, or otherwise estimate dis-
tributions of a particular species;
19
learning patterns of use for different
appliances from simple household
sensors;
20
and learning to predict fail-
ures in aging civil infrastructure.
21
Learning is also integral in real-
izing a great promise of computing
for customization, where nuanced
characterizations of individuals
are possible that are much richer than
binary-valued opinion-poll labels such
as liberal, conservative, waste-
ful, or thrifty. With these richer
characterizations, we can ft sustain-
ability-relevant actions to individuals,
resulting in large savings in energy
and waste, for example, in comfort-
driven services. Consider hotel air
conditioners that are left running
so that a new guest will not experi-
ence a few minutes of uncomfort-
able warmth. In an integrated cyber-
physical-social network that has
learned my preferences, air condition-
ers in a room reserved for me will be
shut down, at least until my arrival.
Intelligent CPSs, as the last illustra-
tion suggests, are yet another class of
systems that will receive considerable
attention in this new AI and Sustain-
ability department. CPSs are at the
intersection of computing and the
physical world. They include static and
dynamic sensor networks and smart
appliances, buildings, cars, highways,
and cities. Through monitoring and
action in the physical world, CPSs will
have second-order effects relative to
sustainability concerns, and these ef-
fects might be environmentally harm-
ful or benefcial. Robotics is another
highly relevant CPS class, particularly
for monitoring the environment.
Considerable work is underway on au-
tonomous underwater vehicles (AUV)
for monitoring ocean and fresh-
water ecosystems and autonomous
aerial and ground vehicles for moni-
toring in emergencies ranging from
nuclear accidents to wildfres.
Looking Ahead
This article has barely touched on
the vast possibilities for intelligent
systems to address sustainability
concerns. Work in this area will of-
ten boil down to augmenting human
decision-making capabilities in the
face of uncertainty and other complex-
ities. In some cases, such as emergency
response, intelligent systems will re-
duce the latency of response while in-
creasing its quality. In other settings
requiring and allowing for delibera-
tion, intelligent systems can facilitate
better-informed and better-reasoned
decisions. We can hope that reli-
ance on intelligent systems will have
positive second- and third-order ef-
fects on the manner in which humans
reasona machine learning system,
for example, typically requires data
and decisions stemming from their
recommendations will be informed
by evidence, perhaps serving as ex-
emplars of reasoning. Pedagogical
goals and strategies can be designed
into these systems from their incep-
tion, thus not simply offoading work
Work in this area will
often boil down to
augmenting human
decision-making
capabilities in the face
of uncertainty and other
complexities.
IS-26-06-Sustain.indd 17 11/10/11 10:42 AM
18 www.computer.org/intelligent IEEE INTELLIGENT SYSTEMS
and/or providing recommendations,
but helping humans to become better
problem solvers at the same time.
Clearly, research, education, and
application in sustainability will chal-
lenge AI along many trajectories, tak-
ing us outside our usual boxes, as ap-
plication inspired and use-driven basic
research often does.
22
Its an impor-
tant time for AI as we grapple with the
complexities of designing a sustainable
and equitable society. I am excited to
see what emerges and hope that much
of it will be reported in these pages.
Acknowledgments
I thank Erwin Gianchandani and Mary
Lou Maher for helpful comments on earlier
drafts of this article. I acknowledge support
from the US National Science Foundation,
under the auspices of the Intergovernmen-
tal Personnel Act, where I served as a pro-
gram director from 2007 to 2010, during
which much of my experience in computing
and sustainability was amassed and ideas
formulated. The opinions expressed herein
are not necessarily those of NSF or the col-
leagues who I have acknowledged.
References
1. S. Dzeroski et al., Equation Discovery
with Ecological Applications, Machine
Learning Methods for Ecological Ap-
plications, A.H. Fielding, ed., Kluwer,
1999.
2. C.P. Gomes, Computational Sustain-
ability: Computational Methods for a
Sustainable Environment, Economy,
and Society, The Bridge, vol. 39,
no. 4, 2009, pp. 513; www.nae.edu/
File.aspx?id=17673.
3. G.H. Brundtland, ed., Report of the
World Commission on Environment
and Development: Our Common
Future, United Nations, 1987.
4. N. Eagle and E. Horvitz, eds., Papers
from the AAAI Spring Symposium on
Artifcial Intelligence and Develop-
ment, tech. report SS-10-01, AAAI,
2010; www.aaai.org/Press/Reports/
Symposia/Spring/ss-10-01.php.
5. R. Carson, Silent Spring, Houghton
Miffin, 1962.
6. S. Weart, The Discovery of Global
Warming: Revised and Expanded
Edition, Harvard Univ. Press, 2008.
7. A. Leiserowitz et al., Global Warmings
Six Americas, Yale Project on Climate
Change Communication, Yale Univ.
and George Mason Univ., 2011; http://
environment.yale.edu/climate/fles/
SixAmericasMay2011.pdf.
8. D.H. Fisher, Sustainability, Leader-
ship in Science and Technology: A
Reference Handbook, W.S. Bainbridge,
ed., Sage Publications, 2011.
9. E. Redden, Schools of Sustainabil-
ity, Colleges of the Environment,
Inside Higher Education, 2009; www.
insidehighered.com/news/2009/07/23/
sustainability.
10. A. Kohler and L. Erdmann, Expected
Environmental Impacts of Pervasive
Computing, Human and Ecological
Risk Assessment, vol. 10, no. 5, 2004,
pp. 831852.
11. S. Roberts, Measuring the Relationship
between ICT and the Environment, Or-
ganization for Economic Co-operation
and Development (OECD), 2009; www.
oecd.org/dataoecd/32/50/43539507.pdf.
12. B. Tomlinson, Greening through IT,
MIT Press, 2010.
13. J.E. Petersen et al., Dormitory Resi-
dents Reduce Electricity Consumption
when Exposed to Real-Time Visual
Feedback and Incentives, Intl J. Sus-
tainability in Higher Education, vol. 8,
no. 1, 2007, pp. 1633.
14. D. Mackenzie, Mathematics of Climate
Change: A New Discipline for an Un-
certain Century, Mathematical Sciences
Research Inst., 2007; www.msri.org/
at t achment s / workshops / 462/
MathClimate.pdf.
15. R. Bryant et al., Science, Engineer-
ing, and Education of Sustainability:
The Role of Information Sciences and
Engineering, version 18, Computing
Community Consortium, 2011; http://
cra.org/ccc/docs/RISES_Workshop_
Final_Report-5-10-2011.pdf.
16. E. Bonabeau, Agent-Based Modeling:
Methods and Techniques for Simulating
Human Systems Proc. Natl Academy
of Sciences, vol. 99, suppl. 3, 2002,
pp. 72807287; www.pnas.org/
content/99/suppl.3/7280.full.
17. D. Zeng, K. Carley, and F.-Y. Wang,
Social Computing, IEEE Intelli-
gent Systems, vol. 22, no. 5, 2007,
pp. 2022.
18. D.H. Fisher and M.L. Maher, eds.,
Papers from the AAAI Spring Sym-
posium on Artifcial Intelligence and
Sustainable Design, tech. report SS-11-
02, AAAI, 2011; www.aaai.org/Press/
Reports/Symposia/Spring/ss-11-02.php.
19. T. Dietterich, Machine Learning in
Ecosystem Informatics and Sustainabil-
ity, Proc. 21st Intl Joint Conf. Arti-
fcial Intelligence, Morgan Kaufmann,
2009, pp. 813.
20. S. Gupta, M.S. Reynolds, and S.N. Patel,
ElectriSense: Single-Point Sensing Using
EMI for Electrical Event Detection and
Classifcation in the Home, Proc. Conf.
Ubiquitous Computing (UbiComp),
ACM Press, 2010, pp. 139148.
21. P. Gross et al., Predicting Electricity
Distribution Feeder Failures Using Ma-
chine Learning Susceptibility Analysis,
Proc. 18th Conf. Innovative Applica-
tions of Artifcial Intelligence (IAAI-
06), AAAI Press, 2006, pp. 17051711.
22. D. Stokes, Pasteurs Quadrant: Basic
Science and Technological Innovation,
Brookings Inst. Press, 1997.
Douglas H. Fisher is an associate profes-
sor of Computer Science at Vanderbilt Uni-
versity. Contact him at douglas.h.fsher@
vanderbilt.edu.
Selected CS articles and columns
are also available for free at
http://ComputingNow.computer.org.
IS-26-06-Sustain.indd 18 11/10/11 10:42 AM