Computational Sociology computational task of testing theories against

quantitative empirical data.

WILLIAM SIMS BAINBRIDGE An agent is a computational entity that can act,
National Science Foundation, USA somewhat in the manner of an animal or human
being, sensing external events and doing things
that affect the environment. Autonomous soft-
ware agents can be either simple or complex, but
Arguably, sociology was the first science to take
advantage of modern computing. For the 1890 US even the simplest can cause complex effects when
census Herman Hollerith developed information- many of them interact in a multiagent system.
processing technology that was used for decades Agents can be heterogeneous, either following
and led to the formation of the IBM Corporation. different rules of behavior or possessing different
Recently, a very large number of new research resources and memories that cause them to act in
methodologies have proliferated, many of them different ways even when following a shared set
analyzing social data readily available over the of rules.
Internet, achieving rapid scientific advance in Perhaps the most influential early agent-based
computer science but with as yet uncertain sig- simulation for sociologists was actually car-
nificance for sociology, as the discipline was ried out by a political scientist, Robert Axelrod
traditionally defined. Much of the best recent (1984), who explored the conditions under which
research that is clearly relevant for sociology was self-interest could bring people to co-operate
performed by people with training in other fields, with each other. Sociologists in some schools
so this entry will define sociology broadly as the of thought had long argued that shared values,
science that studies the structure and dynam- religion, or stable cultural institutions are essen-
ics of modern societies, without regard to the tial to bring people to act co-operatively in their
theoretical tradition invoked by the researchers. dealings with each other. Axelrod’s simulation
The examples will be drawn from four specific intentionally left out all these factors, to see if
areas, seeking a balance between diversity and they were necessary. He challenged social and
coherence, established and exploratory. computer scientists to write algorithms that
would compete in a great computer simulation
tournament, each representing a strategy that one
Artificial Social Intelligence or more agents would follow in exchanges where
each promised to give the other some benefit.
Computer simulation is not a very new method One successful algorithm was Tit-for-Tat. It had
in sociology, and entering that term into JSTOR two simple rules: (1) on the first turn interacting
searching for sociological journal articles turned with another agent, keep your bargain in the
up 107 of them dated from 1958 to 1989, while exchange; (2) after the first turn, do whatever the
96 sociology articles using the term “artificial other agent did the previous time. In a population
intelligence” were published over the same span of agents following various rules, Tit-for-Tat out-
of years. Yet much of the really sophisticated performed other strategies in terms of allowing
work in what could be called “artificial social the agent to benefit from mutually profitable
intelligence” dates from the most recent quar- exchanges, without being exploited frequently by
ter century (Bainbridge et al., 1994). A review deceitful agents. Axelrod’s study does not prove
article on computational sociology by Michael that human co-operation results from entirely
Macy and Robert Willer (2002), as well as one by self-interested behavior based on a simple strat-
Lars-Erik Cederman (2005), argued that agent- egy, but it does prove that other factors are not
based modeling is the most promising approach logically necessary.
if the goal is to develop sociological theories, As in the real world, interaction in many
rather than to perform the more traditional agent-based social simulation programs is a

The Blackwell Encyclopedia of Sociology. Edited by George Ritzer.

© 2016 John Wiley & Sons, Ltd. Published 2016 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781405165518.wbeosc084.pub2
2 C O M P U TAT I O N A L S O C I O L O G Y

decentralized or distributed process that occurs text, in terms of a large number of theory-based
locally around individuals and small groups, systems, such as the semantic differential. As of
and that builds from the local level to create 2015, Harvard still maintained a General Inquirer
large-scale social phenomena. Generally, such website, through which researchers could ana-
simulations are not realistic enough to accom- lyze text conveniently, and occasional research
plish empirical tests, notably because real human publications still make good use of this approach.
societies are cacophonies of many complex fac- It has limitations, such as the problem of word
tors; yet they can evaluate the logical consistency sense disambiguation – the challenge of distin-
of well-formulated theories, which can be as guishing different meanings of a word – yet that
important as the theories’ predictive power for is not the primary reason it is little used. Rather,
empirical data. One example was a large-scale until recently, computer analysis of written text
set of multiagent simulations, employing tens of has been developed primarily for military and
thousands of agents, each of which was modeled national intelligence purposes, with very practical
as a set of neural networks, to test a sociological goals that tended to ignore social science theory
theory of religion (Bainbridge, 2006). in preference to algorithms that could provide
Since its first issue in 1998, The Journal of small but measurable improvements in such tasks
Artificial Societies and Social Simulation has as mass translation of written communications,
offered a wide range of exemplary studies, and and data-mining for hints of terrorist threats.
is a good place to gain a general perspective At present, one of the best strategies is to cali-
on the field. Among its most-cited articles is a brate an automatic system with the judgments of
survey of software toolkits that may be used to human beings. For example, a team at the Uni-
build simulations, each possessing strengths and versity of Maryland wanted to develop a neural
limitations (Nikolai and Madey, 2009). Another network system that could automatically classify
compares the decision-making rules followed by politically oriented sentences in terms of the
artificial agents in a range of scientific studies liberal–conservative divide in the United States
(Balke and Gilbert, 2014). Use of existing software (Iyyer et al., 2014). Neural networks and simi-
and decision-making procedures maximizes the lar computerized systems for performing what
clarity and reproducibility of results, yet may specialists in artificial intelligence call “machine
inhibit innovation. For example, one of the most learning” generally require a training data set that
productive long-term projects in computational has already been annotated by human beings.
sociology has been the affect control theory work The training can be done in many different ways,
of David Heise and his many associates, which but one is to input a sentence into the computer,
combines natural language processing with com- let it judge whether the sentence is liberal or
puter simulation, in a variety of research projects conservative, and then tell it whether its guess
ranging from highly theoretical to strictly empir- was correct or incorrect. Through methods like
ical (e.g., Schneider and Heise, 1995). Creative back-propagation of error, the system will adjust
work such as this cannot be done using existing its internal state slightly after being told it gave
software, yet it contributes to the store of reusable a wrong answer, gradually becoming more and
software for the future. more accurate. One potential disadvantage of
this method is that it may be difficult to deduce
what set of cognitive rules the machine develops
Natural Language Processing to achieve mostly correct answers, so unless
transparency is built into a classifier of this kind,
Computer analysis of written text has a long it is more useful as a tool for empirical research
tradition in the social sciences, yet except for than as a tool for developing theory as illustrated
temporary surges in popularity over the years in the previous section of this entry.
it has tended to be quite peripheral to other A sociologically interesting question is how
methods. The best classical example is General the Maryland team annotated text in terms of
Inquirer developed at Harvard by Philip Stone political orientation. One approach was to start
(1966) and his associates. This software catego- with orations from the US Congress, classified
rized the words employed in a long sample of simply in terms of the political party to which the
 C O M P U TAT I O N A L S O C I O L O G Y 3

speaker belonged. But the research employed a Internet offers a vast quantity and wide diversity
much more flexible method, called “crowdsourc- of text, audio, and video for analyses relevant to
ing.” An online system called the Crowdflower many subfields of the discipline.
platform recruited US residents online to classify
phrases as liberal, conservative, neutral, or uncer-
tain. As in much good crowdsourcing research, Virtual Worlds
each research subject did some initial tasks that
served as screening tests, and only those who Over the past decade, a significant fraction of
performed well participated in the annotation. humanity has begun conducting portions of its
A sophisticated aspect of this study was that the social life online, in a great variety of computer-
phrases in a sentence were judged separately supported communication technologies. The
and then judged in progressively more assem- previous section identified methodologies suit-
bled form until the full sentence was presented. able for sociological research on text-based
Crowdsourcing can be expensive, but here it was communications, so this section will use a very
used to prepare training data for a neural network different example. The term “virtual world” gen-
that could then classify vast samples of political erally refers to a computational environment in
text at low cost. which the user experiences sound and vision
In effect, crowdsourcing is a way to tap into the similar to those of television and movies, but,
cultural assumptions shared by large numbers through the perspective of an avatar, is able to
of people, but it is also valid to base automatic move and act spontaneously within the environ-
analysis of large bodies of text on the cognitive ment, and with the expectation that it will persist,
framework belonging to a single individual, such as will the results of the user’s behavior.
as the researcher, or to analyze in parallel from At the present time, most virtual worlds are
the perspectives of a few people, who then can be marketed as multiplayer video games or com-
compared. One research study created a program puter games that facilitate social interaction
incorporating 800 pairs of adjective antonyms over the Internet, displayed through a television
that could describe a person, asking the key set or desktop computer. Some, like Second
respondent to rate each of the 1,600 separately in Life, are not games but systems in which users
terms of two eight-point scales: (1) To what extent may create their own virtual homes, businesses,
do I myself have this characteristic? (2) How good and visually complex environments mirroring
is it for a person to have this characteristic? The either nature or fantasy. For years, a very few
software could then automatically search a num- examples have employed head-mounted displays
ber of text documents, such as novels by various that separate the virtual world from the real
authors, coding each in terms of propensity for one in the user’s perception, and others employ
using adjectives of different kinds to describe mobile devices that link actions in the real and
people, allowing systematic comparison of both virtual environments more closely. It is not clear
the documents and the raters (Bainbridge, 2014: which technical approaches will be popular in
42–49). the future, but already tens of millions of people
The currently available methods for natural invest many hours every month playing massively
language processing do indeed seem to offer new multiplayer online role-playing games (MMOs or
opportunities for sociological research on ideolo- MMORPGs), so they are the standard example
gies, political sociology, and even policy-making today.
within information-intensive organizations Late in 2014 World of Warcraft (WoW) cele-
(Conover et al., 2012; Epstein, Farina, and Heidt, brated its tenth anniversary with fully 10 million
2014). Although many studies limit the data to paying subscribers. It is a successor to a series of
written text, others work from videos of spoken three more limited real-time strategy Warcraft
language, facial expressions, and gestures (Pérez- games that began in 1994, and were explicitly
Rosas and Mihalcea, 2013). We can hope that the inspired by the table-top role-playing game Dun-
accuracy of speech recognition and text inter- geons and Dragons, dating from 20 years earlier
pretation will improve, but such methods can and featuring complex stories set in a consis-
already do useful work for sociologists, and the tent fantasy universe. A WoW player begins by
4 C O M P U TAT I O N A L S O C I O L O G Y

downloading gigabytes of program and data onto may have thousands of members. The two guilds
a desktop or laptop computer and creating an that co-operated in staging the 2008 conference
avatar, belonging to one of 13 races: Blood Elf, were still active six years later. The software
Draenei, Dwarf, Gnome, Goblin, Human, Night systems for managing guilds and markets in
Elf, Orc, Pandaren, Tauren, Troll, Undead, or these gameworlds, and some online databases
Worgen. Each race has a different appearance, concerning avatars and guilds, are useful tools
and a different history or backstory, and enters the for quantitative research. For example, the freely
vast, multicontinent virtual world at a different available online database of WoW avatars offers
location. All the races except the Pandarans have hundreds of variables for millions of cases.
formally combined into two competing factions, Several researchers have proposed what they
the Horde and the Alliance, each with a number call the Proteus Effect, a tendency for people to
of capital cities, its own economy, and under var- treat their avatars as somewhat real, and to be
ious conditions bloody conflict against the other influenced in real life by experiences their avatars
faction. In creating an avatar, the player must also had in virtual worlds (Yee, 2014). Others have
select a class, which determines some key abilities employed sociological concepts such as “values”
and the kind of weapons and armor used: death in suggesting design principles, for example when
knight, druid, hunter, mage, monk, paladin, creating a gameworld for educational purposes
priest, rogue, shaman, warlock, or warrior. Over (Flanagan and Nissenbaum, 2014). There is even
time, an avatar ascends several status ladders, of a small group of researchers who have explored
which general experience from level 1 to level 100 virtual religions, for example comparing their
is most central, often taking hundreds of hours manifestations in a game like World of Warcraft
to do so. Dedicated WoW players typically have with those in a nongame virtual world such as
several different avatars, experiencing different Second Life (Geraci, 2014). One widely discussed
stories within races, and playing different func- theory holds that the research on computer game-
tional roles in teams because the class system is a worlds can illuminate the meaning of religion
division of labor. in the material world. In one view, religion is a
Many researchers have employed a variety of serious art form that should be respected but not
methods to study WoW, often a combination believed, and it imagines noncomputer virtual
of ethnography and role-playing (Bainbridge, worlds with names like Heaven and Hell, thus
2010b; Nardi, 2010). The use of role-playing suggesting that online gameworlds represent a
has been rare in sociology recently, but social- major cultural transition to a fully secular society
network pioneer Jacob Moreno, who founded the (Bainbridge, 2013).
American Sociological Association’s social psy- Research in virtual worlds is quite active at
chology journal, developed an approach he called the present time, and may provide models for
“psychodrama,” which seems to be coming back comparable research in other fields of social
into fashion (Moreno, 1946). Because everybody computing. It is worth noting that essentially
in an MMO is playing a role – their real-world all public wikis allow contributors to register
identities are usually concealed – and MMOs are under pseudonyms, similar to creating an avatar.
public spaces, ethical issues concerning human Individuals may create alternative identities for
subjects are generally considered minor, whereas blogs, forums, and video services, and many of
the use by researchers of false identities in real- these online communities may be studied using
world settings is usually considered unethical. the same methods as virtual worlds.
A scientific conference held inside World of
Warcraft in May 2008, with as many as 120 aca-
demics and students in virtual attendance for the Citizen Science
main meetings, published a conventional book
of proceedings that contrasts many theories and Across many fields of natural science, a major
methods (Bainbridge, 2010a). Many MMOs pro- social movement is well advanced to include
vide very effective software systems for creating volunteer enthusiasts in teams of researchers,
short-term small teams, large groups of perhaps sometimes thousands of them, under the general
40 members for raids, and enduring “guilds” that term “citizen science.” For example, the website
 C O M P U TAT I O N A L S O C I O L O G Y 5

of the popular magazine Scientific American citizen science is reduced cost, which should be
advertises dozens of such projects so that readers especially appealing to sociologists who often
may volunteer. Among the most famous and have difficulty obtaining funding from the federal
best-established large citizen science projects government, or, depending upon the specific
are Galaxy Zoo, in which volunteers classify topic of the research, from private foundations.
photographs of galaxies, and eBird, in which they A question that arises when contemplating the
collect field data about sightings of bird varieties potential future of citizen sociology is how objec-
by time and place. Unfortunately, hardly any of tivity can be achieved when so many sociological
the projects publicly advertised or studied by topics are controversial. The answer may be that
information scientists are sociological, yet the this question is moot, because all social science,
potential for citizen sociology is immense. perhaps all science, is conducted within the
A logical starting point would be the many context of a particular culture, economic interest,
local oral-history projects, such as the extensive and political ideology. Thus, citizen sociology
one carried out for many years by the public may generally take the form of responsible but
library in Greenwich, Connecticut, which could reflexive research that explicitly states its concep-
be expanded and transformed into urban, com- tual framework so that anyone outside the team
munity, or industrial sociology, depending on the can evaluate the relevance of its findings, in terms
nature of the local society (Bainbridge, 2014: 145). of their own needs and perspectives.
We can imagine a time in the future when every
student who graduates college with a sociology
degree is considered a lifelong sociologist, work- Conclusion
ing as a volunteer on a series of research projects,
in collaboration with former professors or PhD The four areas discussed in this entry are by no
sociologists employed at local institutions. means the only points of convergence between
The computational aspects of citizen science sociology and computer science, yet they provide
are many. Highly innovative projects often need a diverse set of examples, and highlight areas
to create new software for collection and man- where many sociologists will be able to complete
agement of data, but then that software can be high-quality research projects in the near future.
employed by other projects, but only if it has Among other areas are some that would merely
been designed from the beginning for modifi- apply traditional analytical techniques to online
cation and adaptation to new tasks (Prestopnik data, in many cases cheap to obtain but good for
and Crowston, 2012). Typically, communica- testing general hypotheses. For example, prefer-
tions between team members take place over ence ratings for movies or books are comparable
the Internet, whether simply in the form of to opinion questionnaires, in cases when many
data submission, via e-mail and voice chat, or individuals rated many of the same items. Popular
in classes or committee meetings held in more search engines seek Web sites that are similar, and
elaborate virtual environments. Each step in the could be used to map relations among religious
work may require specialized computing and or political groups, for example in terms of how
communication resources, for example collect- many other Web sites link to both sites in each
ing, managing, analyzing, and interpreting data, pair from the set of interest. Online genealogy
and disseminating results (Newman et al., 2012). Web sites offer historical census data, and other
A rather extensive literature already exists that kinds of information such as birth and mar-
resembles sociology, reporting research on the riage records, which could be used in historical
organization, management, and motivations of demography or urban studies.
citizen science projects (Shirk et al., 2012; Wig- Given that all college students these days are
gins and Crowston, 2011). Thus, one way sociol- very familiar with the World Wide Web, projects
ogists could gain expertise in this new area would in many of these areas could be designed to
be to attach themselves to citizen science teams in serve educational goals. If a relatively simple
well-established fields, such as astronomy, zool- research project could employ many brief ses-
ogy, ecology, and health science. One obvious sions online, each student in an introductory
reason for casting a research project in terms of class could do one of them, submit the data as
6 C O M P U TAT I O N A L S O C I O L O G Y

a homework assignment, and then the teacher Flanagan, M. and Nissenbaum, H. (2014) Values at Play
could assemble the data and present findings to in Digital Games, MIT Press, Cambridge, MA.
the class in a lecture. Given good guidance, and Geraci, R.M. (2014) Virtually Sacred: Myth and Meaning
a proper balance of imagination and rigor, some in World of Warcraft and Second Life, Oxford Univer-
sity Press, Oxford.
graduate students may be able to complete highly
Iyyer, M., Enns, P., Boyd-Graber, J., and Resnik, P.
publishable dissertation research in one of the
(2014) Political ideology detection using recursive
computational sociology areas, at low cost except neural networks, in Proceedings of the 52nd Annual
for the student’s own labor. Meeting of the Association for Computational Lin-
guistics, Association for Computational Linguistics,
Stroudsburg, PA, pp. 1113–1122.
Disclaimer Macy, M.W. and Willer, R. (2002) From factors to
actors: computational sociology and agent-based
The views expressed in this essay do not neces- modeling. Annual Review of Sociology, 28, 143–166.
sarily represent the views of the National Science Moreno, J.L. (1946) Psychodrama and group psy-
chotherapy. Sociometry, 9, 249–253.
Foundation or the United States.
Nardi, B. (2010) My Life as a Night Elf Priest: An Anthro-
SEE ALSO: Complexity and Emergence; Inter- pological Account of World of Warcraft, University of
Michigan Press, Ann Arbor, MI.
net; Qualitative Methods; Quantitative Methods;
Newman, G., Wiggins, A., Crall, A., et al. (2012) The
Science; Theory Construction future of citizen science: emerging technologies and
shifting paradigms. Frontiers in Ecology and the Envi-
ronment, 10 (6), 298–304.
References Nikolai, C. and Madey, G. (2009) Tools of the trade: a
survey of various agent based modeling platforms.
Journal of Artificial Societies and Social Simulation, 12
Axelrod, R. (1984) The Evolution of Cooperation, Basic
(2), 2.
Books, New York.
Pérez-Rosas, V. and Mihalcea, R. (2013) Utterance-level
Bainbridge, W.S. (2006) God from the Machine: Artificial
multimodal sentiment analysis, in Proceedings of the
Intelligence Models of Religious Cognition, AltaMira
51st Annual Meeting of the Association for Compu-
Press, Walnut Grove, CA.
tational Linguistics, Association for Computational
Bainbridge, W.S. (ed.) (2010a) Online Worlds: Conver-
Linguistics, Stroudsburg, PA, pp. 973–982.
gence of the Real and the Virtual, Springer, London.
Prestopnik, N.R. and Crowston, K. (2012) Citizen sci-
Bainbridge, W.S. (2010b) The Warcraft Civilization:
ence system assemblages: understanding the tech-
Social Science in a Virtual World, MIT Press, Cam-
nologies that support crowdsourced science, in Pro-
bridge, MA.
ceedings of the 2012 iConference, ACM, New York, pp.
Bainbridge, W.S. (2013) eGods: Faith Versus Fantasy
168–176.
in Computer Gaming, Oxford University Press, New
Schneider, A. and Heise, D.R. (1995) Simulating sym-
York.
bolic interaction. Journal of Mathematical Sociology,
Bainbridge, W.S. (2014) Personality Capture and Emu-
20, 271–287.
lation, Springer, London.
Shirk, J.L., Ballard, H.L., Wilderman, C.C., et al. (2012)
Bainbridge, W.S., Brent, E.E., Carley, K., et al. (1994)
Public participation in scientific research: a frame-
Artificial social intelligence. Annual Review of Soci-
work for deliberate design. Ecology and Society, 17
ology, 20, 407–436.
(2), article 29.
Balke, T. and Gilbert, N. (2014) How do agents make
Stone, P.J. (1966) The General Inquirer: A Computer
decisions? A survey. Journal of Artificial Societies and
Approach to Content Analysis, MIT Press, Cambridge,
Social Simulation, 17 (4), 13.
MA.
Cederman, L.-E. (2005) Computational models of
Wiggins, A. and Crowston, K. (2011) From conserva-
social forms: advancing generative process theory.
tion to crowdsourcing: a typology of citizen science,
American Journal of Sociology, 110 (4), 864–893.
in Proceedings of the Forty-Fourth Hawai’i Interna-
Conover, M.D., Gonçalves, B., Flammini, A., and
tional Conference on System Science, IEEE, New York.
Menczer, F. (2012) Partisan asymmetries in online
Yee, N. (2014) The Proteus Paradox, Yale University
political activity. EPJ Data Science, 1 (6), 1–17.
Press, New Haven, CT.
Epstein, D., Farina, C., and Heidt, J. (2014) The value of
words: narrative as evidence in policy making. Evi-
dence and Policy, 10, 243–258.