Social Media as Windows on the Social Life of the Mind

Cosma Rohilla Shalizi

Statistics Department, Carnegie Mellon University
5000 Forbes Avenue
Pittsburgh, PA 15213 USA
arXiv:0710.4911v1 [cs.CY] 25 Oct 2007

Abstract of our peculiar bounded rationality and bounded lifes-

pan, is a deep question, fortunately not relevant here.)
This is a programmatic paper, marking out two direc- While individual thinkers invent and discover, it is
tions in which the study of social media can contribute nonetheless true that innovations are typically refined,
to broader problems of social science: understanding
cultural evolution and understanding collective cogni-
extended and perfected by groups, and that it is very
tion. Under the first heading, I discuss some difficulties rare indeed for highly developed concepts and ideas to
with the usual, adaptationist explanations of cultural emerge from a single, isolated thinker, rather than from
phenomena, alternative explanations involving network a process of interaction (Toulmin 1972; Kitcher 1993;
diffusion effects, and some ways these could be tested Collins 1998; Ziman 2000).
using social-media data. Under the second I describe The branches of social science for which these facts
some of the ways in which social media could be used to are common-places have largely developed them philo-
study how the social organization of an epistemic com-
sophically (Toulmin 1972; Turner 2002), or qualita-
munity supports its collective cognitive performance.
tively (Vygotsky 19341986; 1978; Balkin 1998; Mercer
2000) or even ethnographically (Luria 1976; Hutchins
Let me begin by considering two1 senses in which we
1995). (But see (Lupia, McCubbins, & Popkin 2000).)
might speak of human thought as being “social”, and
In part this has been for reasons of cultural and in-
how they might orient the study of social information
tellectual politics, as the relevant scholars have tended
processing and social media.
to fall on the “interpretation” rather than “explana-
The first sense is a common-place of many schools in
tion” side of the divide in the social sciences (Sperber
the social sciences and humanities: our thought relies
1996), so that attention to the social nature of thought
on the cultural transmission of cognitive tools. Every
often goes along with more or less pronounced hostil-
individual thinker, no matter how innovative or even
ity to quantitative and computational modeling (e.g.
lonely they may be, depends crucially on a vast array
Hutchins; Mercer). This supposed opposition is thor-
of cognitive tools (concepts, procedures, languages, as-
oughly mis-guided (Frawley 1997), but it is not likely
sumptions, values, ...) which they did not devise them-
that anyone will be argued out of it any time soon.
selves, and could not have devised for themselves. In-
stead they inherited these cognitive tools from interact- More promisingly, however, one good reason for de-
ing with other people, who for the most part themselves veloping this idea through small-scale qualitative stud-
did not invent them. (Dewey 1927; Vygotsky 19341986; ies has been that it was impossible to gather relevant
Popper 1945; Balkin 1998)2 (Whether this dependence data, suitable for quantitative analysis, on any large
on tradition is a logical necessity, or merely a reflection scale. With the rise of social media, however, many
people are, for their own purposes, generating exactly
Copyright c 2008, Association for the Advancement of Ar- this kind of data for us — traces of their communica-
tificial Intelligence (www.aaai.org). All rights reserved. tive interactions as they work out their thoughts about
1
Of course, people think a lot about their own and others’ matters of common concern. They are doing so on a
social interactions, and a big use of social media is sharing wide range of subjects, under a wide range of different
these thoughts. But in this social media are no different institutional mechanisms which structure their interac-
from any other form of human, or for that matter primate, tions in many different ways, creating natural sources of
association.
2 variation which the social scientist can try to exploit to
“[K]nowledge is a function of association and communi-
learn more about the effects of subject matter, of com-
cation; it depends upon tradition, upon tools and methods
socially transmitted, developed and sanctioned. Faculties municative structure, and of other factors on cultural
of effectual observation, reflection and desire are habits ac- dynamics, and perhaps ultimately even on innovation
quired under the influence of the culture and institutions of and discovery. The next section points out some of
society, not ready-made inherent powers” (Dewey 1927, p. the outstanding problems and methodological pitfalls
158). Cf. (Popper 1945, ch. 23–24). of this area.
 The other important sense in which human thought sition because the former is adapted to the latter, or
can be “social” is that it seems to make sense to regard reflects it, or expresses it.
at least some human social institutions as, themselves, It is natural for us, as beings acutely sensitive to nu-
information-processing systems, engaging in computa- ances of cultural meanings, to try to explain cultural
tions which cannot be localized to representations in differences by trying to explain the content of widely-
the mind of any one of their members. On large scales, shared, cultural representations. It is natural to sup-
market economies, corporations and other bureaucra- pose that, say, one news story rises to the top of a
cies, scientific disciplines, and democratic polities all social aggregation system because it is more interesting
have something of this collective information-processing than other stories which did not. Such explanations are
character. Knowing how they accomplish this would even valid a lot of the time. It is nonetheless important,
be deeply rewarding, and, if that understanding can as a point of methodological hygiene, to develop ways
be used to make them work better, of profound eco- of telling when some bit of culture succeeds in propa-
nomic and political importance. A frontal assault on gating because its content fits its circumstances, if only
this problem, as represented by one of those grand insti- because, being creatures acutely sensitive to nuances of
tutions, is unlikely to succeed (though it may be a mag- cultural meanings, it is far too easy for us to spin such
nificent failure). Fortunately, social information pro- stories no matter what the truth might be. Lieberson
cessing also occurs in much humbler institutions, such (2000) points out that many widely-accepted explana-
as tagging systems and collaborative filtering, where is- tions of trends in fashions, children’s names, etc., can-
sues of data collection and even experimental manipu- not possibly be right (because, e.g., the trend pre-dates
lation are much more manageable, and where we might or is more widely spread than the supposed cause), and
hope to learn more, before tackling the fundamental that these are instead better explained by purely in-
problems of social science. I will lay out some of what ternal mechanisms of the respective fields. In biology,
should be on the agenda of the study of social infor- adaptive and non-adaptive evolution are demarcated by
mation processing, in particular points of contact with means of neutral models. These are models of the ge-
machine learning. netic changes which would be expected due to reproduc-
tive mechanisms and chance alone, all genetic variants
Cultural Evolution being assumed to be “adaptively neutral”, i.e., of equal
“Culture is the precipitate of cognition and communi- fitness. Only when actual populations depart markedly
cation in a human population” (Sperber 1996). That from the predictions of neutral models can adapta-
is, cultural traits — beliefs, practices, habits, conven- tion be (reliably) inferred (Nitecki & Hoffman 1987;
tions, expressions, norms — are not just ones which are Harvey & Pagel 1991). Before the student of social
common across a population, but ones which are spread media, or other cultural media, can start explaining
across a population because its members communicate phenomena by reference to content, they need to check
with one another. (Knowing that it’s painful to look at that there actually is something to be explained.
the sun directly is not cultural; knowing that the direc- A highly simplistic model may make this point more
tion in which the sun rises is called “east” is cultural.) concrete. Consider a network in which people have two
Cultural phenomena are thus emergent, the result of binary traits, one of which is stable (we may think
the communicative interaction of cognitive agents. If of this as “class” or “race” or some similar status),
we are to understand how cultures work, we need to and the other is changeable (think of fashions, or po-
understand something about both parts, the internal litical opinions). Assume that the network is assor-
cognitive mechanisms and the effects of different pat- tative on the stable, social-type trait, so that people
terns of interaction. Social media offer a window into are more likely to be linked to others of the same
the communicative part of the problem of unrivaled type than those of a different type. Such “assorta-
clarity and breadth. This is extremely exciting, but tivity” or “homophily” is observed in many, perhaps
in looking through this window we should bear in mind most social networks, often on such stable social-status
some methodological difficulties to interpreting the view type variables (McPherson, Smith-Lovin, & Cook 2001;
through this window. Newman 2003). Now assign the cultural trait to peo-
It is a common-place observation that there are ple uniformly and independently of their social trait (or
strong relationships between cultural traits and so- anything else). Initially, then, there will be no correla-
cial attributes; that different social groups accept and tion between social and cultural traits, and no assorta-
transmit different bits of culture. Most attempts to tivity based on culture.
explain this from within the social sciences (emphat- We might expect such correlations to appear if the
ically including historical materialism (Elster 1985; process of cultural transmission and retention is biased
Cohen 2000) and its variants) argue that this is due — if, say, certain cultural values only make sense for
to some causal influence of social organization on the those in certain social positions. In that case, we would
content of culture. (“Social being determines con- expect to find a growing “fit” between cultural and so-
sciousness” (Marx & Engels 18471947) — or, once cial variables, as the former adapt to the latter. But by
the Hegelian gas has been released, social life shapes this point you will not be surprised to learn that neu-
thought.) In these views, culture varies with social po- tral transmission processes can also induce such cor-
 were “lower class” and “middle class”, and the cultural
traits “likes black velvet paintings” and “likes black and
2.5 white photographs”, the temptation to explain the cor-
relation by content would be overwhelming (Bourdieu
1984). Nonetheless, which way the correlation went
2.0

would be a matter of pure chance, or more exactly of

the reinforcement and amplification of small fluctua-
1.5

tions, though some such pattern forms with high prob-

χ2

ability. (This contrast between long- and medium- run

1.0

behavior is not uncommon in self-reinforcing network

processes (Pemantle & Skyrms 2004).) The strength
of the dynamically-induced correlations depends on the
0.5

assortativity of the social network; if it is not assorta-

tive, then the correlations between social and cultural
0.0

traits only rarely rise above the levels to be expected

by chance (Fig. 1).
0 500 1000 1500 2000 Social scientists interested in communications have
appreciated for a long time that network structure is
time very important to how information flows through a so-
cial group (Katz & Lazarsfeld 1955; Huckfeldt, Johnson,
& Sprague 2004), but they have not, so far as I know, re-
Figure 1: Neutral copying induces correlations between so- alized that it can create just the kind of correlation that
cial and cultural traits in assortative networks. The graph seems to cry out for an explanation by content. In fact,
has 100 nodes, randomly divided between two social types the real situation is somewhat worse than this, because
(equally probable), and a binary-valued cultural trait (ini- it really isn’t a given that people change because of in-
tially equally probable). Edges between nodes of the same teracting with their neighbors. It could well be that
type occur with probability p1 , those between different types people have neighbors who are similar to themselves,
have probability p2 . At each time step, a random node and so they all respond similarly to common exoge-
copies the cultural trait of a random neighbor. Horizontal nous causes, without any direct interactions (Steglich,
axis: time. Vertical axis: χ2 statistic for the correlation be- Snijders, & Pearson 2004).4 If one thinks of trying to
tween the social and cultural variables. Black line: behavior explain why certain users prefer certain kinds of news
of an assortative network (p1 = 0.09, p2 = 0.01, assortativ- stories, for example, one must account not only assor-
ity coefficient (Newman 2003) of realized graph r = 0.80). tativity, but also for common exposure to some outside
Note the eventual decline of χ2 as the network moves to- news source. — None of this, incidentally, requires that
wards a homogeneous equilibrium; in the very long run it people actually make decisions randomly, but only that
will reach 0. Grey line: behavior of a non-assortative net- the reasons which lead them to their decisions are ef-
work (p1 = p2 = 0.05, r = 0.045). fectively unpredictable from the other variables in the
system.
The moral is not that these kinds effect explain all
relations. To be specific, let’s implement the “voter correlations between social and cultural traits, or even
model” (Liggett 1985): at each discrete time step, a between different cultural traits. Rather, it shows that
node is chosen uniformly at random, independently of a neutral explanation is logically possible. To support
past and future choices. This node chooses a neigh- an adaptive explanation of a correlation, then, one must
bor (again uniformly and independently), and copies show some way in which the neutral model is not ad-
its value of the cultural trait. equate to the data. For example, additional experi-
Clearly, in a connected network, there are two ab- ments (not shown) indicate that, if I take the model
sorbing states, which are culturally homogeneous, and simulated in Fig. 1 and break the graph into commu-
eventually the network must settle into one or the other nities (following Newman & Girvan (2003)), then so-
of them, but the time it takes to do so will typically cial type and cultural traits are conditionally indepen-
be quite long (Sood & Redner 2005). In the mean- dent, given community membership, even in strongly-
while, if the network is socially assortative, numerical assortative networks. This conditional independence
experiments (Fig. 1) show that the social and the cul- does not hold when different social types have differing
tural traits tend to become correlated during a long biases for or against various cultural traits. Only when
“meta-stable” period.3 If I’d said that the social types we have found and verified such discrepancies between

3
One could say that the cultural trait of a node is still (Galles & Pearl 1997).
4
“in the final analysis” determined by its social type, but This possibility seems to confound the claims of the re-
only with the proviso that the over-all structure of the net- cent, and widely-publicized, study of the spread of obesity
work “screens off” the latter, rendering it causally irrelevant in a social network (Christakis & Fowler 2007).
our data and the predictions of a good neutral model the movies or bookmarks which get recommended by
can we say that the adaptive explanation has passed a collaborative filtering services are the emergent prod-
severe test and truly has evidence in its support (Mayo ucts of the interactions of many participants (Lerman
1996). 2007). What social media offer us, again, is the possi-
bility to automatically collect large-scale data on such
Collective Cognition phenomena, combined with a clear understanding of
the interaction structure (or at least a lot of it), as well
It’s been recognized since the 1930s that market
as much of the external circumstances and the goals
economies are “collective calculating devices” (Lange
of the group. We can thus begin, at least at a small
& Taylor 1938; Hayek 1948). A market-clearing allo-
scale, to begin building and systematically testing the-
cation of good and services is simply too big for any-
ories which explain how social information processing
one to grasp, let alone find. Instead it is the process
and collective cognition succeed when they do.
of exchange itself which adaptively finds and imple-
ments this allocation.5 This is an example of what It might be thought that the theoretical explanation
we might call collective cognition, by analogy to the is rather simple, and goes (currently) under the name
classical (Mancur Olson 1971) “collective action”. Sim- of “the wisdom of crowds” (Surowiecki 2004): individ-
ilarly, the problems of designing policies for govern- uals make noisy guesses, which on average are unbiased
ments are largely beyond the scope of what anyone and uncorrelated, so simple averaging leads to conver-
can actually do, but not beyond the scope of demo- gence on the appropriate answer. Taken seriously, this
cratic deliberation, which reduces the problem from explanation implies that our economy, our sciences and
solving for the optimal policy in one stroke, to criti- our polities manage to work despite their social organi-
cizing and improving policies piecemeal (Braybrooke & zation, that science (for example) would progress much
Lindblom 1963), in light of the information and ideas faster if scientists did not collaborate, did not read each
of many participants. (Popper 1945; Lindblom 1965; others’ papers, etc. While every scientist feels this way
Ober 2005) (Historically, democratic decision-making occasionally, it is hard to take seriously. Clearly, there
has been associated with more social power than other has to be an explanation for the success of social in-
forms of government (McNeill 1982), but the causality formation processing other than averaging uncorrelated
is unclear.) Similar remarks apply to bureaucratic or- guesses, something which can handle, and perhaps even
ganizations, such as corporations, and to scientific dis- exploit, statistical dependence between decision mak-
ciplines. ers.
It is notable that modern societies are vastly better A particularly interesting line of attack on these prob-
at collective cognition than earlier ones. The degree lems is suggested by the analogy with ensemble meth-
of organization, and its precision, which we take for ods in machine learning. As Domingos (1999) has
granted would have been astonishing for even the in- pointed out, the success of these methods seems to con-
habitants of the most advanced societies c. 1600, to found naive interpretations of Occam’s Razor, in much
say nothing of c. 100. Historians have explored some the same way that the success of social information pro-
of the technical and institutional underpinnings of these cessing confounds the simple “wisdom of the crowds”
organizational revolutions (McNeill 1982; Beniger 1986; story. Ensemble methods, in which large numbers of
Yates 1989), but at a deeper level we have little idea why low-capacity classifiers or predictors (e.g., shallow clas-
this is so, or why what we do works (when it does work). sification trees) are combined, effectively create a sin-
This makes it harder to improve the functioning of our gle model of what appears to be very high capacity,
institutions for collective cognition. Economic theories and so they appear to be nothing but an invitation to
of mechanism design attempt to do so, but largely ad- over-fitting. Worse, typically ensemble methods such as
dress the problem of motivating people to act in certain boosting (Hastie, Tibshirani, & Friedman 2001), bag-
ways, rather than of how to figure out what the right ging (Breiman 1996) and mixtures of experts (Jacobs
action is (Miller 1992). 1997) create correlated low-level predictors, so that the
These are all very large themes indeed, of course, and simple average-the-crowd story is inapplicable. In fact,
it might seem grandiose to even mention them in this it is precisely because the component predictors are cor-
context. I am not suggesting that studying social me- related, but not identical, that the actual capacity of the
dia will give us the key to all organization technologies. ensemble is much smaller than its apparent capacity.
What it can do, however, is give us a set of case studies A similar result holds for cooperative problem-solving
where, on a much humbler level, people are nonetheless (Hong & Page 2004). Under mild conditions, it can be
engaged in social information processing and collective shown that a large group of “weak” heuristic problem-
cognition. Just as no one market participant decides on solvers, whose performance in isolation is only slightly
or represents the over-all market allocation, and no one better than random search, will actually out-perform a
scholar ever grasps more than a small portion of what similarly-sized group of “strong” heuristics, ones whose
is known about conic sections or cellular slime molds, average performance in isolation is much better. One of
those conditions, however, is that the problem-solvers
5 must be able to communicate with each other, mak-
On the formal computational power of market-like sys-
tems, see (Walsh et al. 2003). ing their candidate solutions strongly dependent rather
than uncorrelated. There is good evidence that this Press.
beneficial effect of heuristic diversity and communica- Beniger, J. 1986. The Control Revolution: Technolog-
tion is actually seen in the cognitive performance of hu- ical and Economic Origins of the Information Society.
man groups (Page 2007). This suggests a very promis- Cambridge, Massachusetts: Harvard University Press.
ing direction for research on social information pro- Bourdieu, P. 1984. Distinction: A Social Critique of
cessing, namely to use the mathematical techniques the Judgement of Taste. Cambridge, Massachusetts:
of statistical learning theory to establish bounds on Harvard University Press.
the performance of suitable sorts of ensemble-learners
and group problem-solvers, and see how close actual Braybrooke, D., and Lindblom, C. E. 1963. A Strategy
social information processing systems come to attain of Decision: Policy Evaluation as a Social Process.
those bounds, and how the latter could be improved by Glencoe, Illinois: The Free Press.
changes to their architectures. Breiman, L. 1996. Bagging predictors. Machine Learn-
Both ensemble methods and the Hong & Page re- ing 24:123–140.
sults on diverse heuristics posit relatively simple forms Chamley, C. 2004. Rational Herds: Economic Models
of “social” organization, such as direct averaging, or of Social Learning. Cambridge, England: Cambridge
passing a problem to the next person able to improve University Press.
on the current solution. There is every reason to think, Christakis, N. A., and Fowler, J. H. 2007. The spread
however, that the optimal form of organization will ac- of obesity in a large social network over 32 years. The
tually depend on the structure of the problem being New England Journal of Medicine 357:370–379.
solved. (Cf. Braybrooke & Lindblom (1963) on how the
Cohen, G. A. 2000. Karl Marx’s Theory of History: A
social organization of policy analysts serves their cog-
Defense. Princeton, New Jersey: Princeton University
nitive strategy of “disjointed incrementalism.”) In par-
Press, second edition.
ticular, coordination over time is not an issue in ensem-
ble methods, and handled by assumption in the Hong Collins, R. 1998. The Sociology of Philosophies:
& Page model, but extremely important in real-world A Global Theory of Intellectual Change. Cambridge,
systems for social information processing and collective Massachusetts: Harvard University Press.
cognition. Dewey, J. 1927. The Public and Its Problems. New
This suggests a final line of research, one which draws York: Henry Holt.
together ideas from distributed systems, economics and Domingos, P. 1999. The role of Occam’s Razor in
statistical mechanics. Experience with distributed sys- knowledge discovery. Data Mining and Knowledge
tems shows that often the hardest part of their design is Discovery 3:409–425.
ensuring coordination over time, and that failure to do Elster, J. 1985. Making Sense of Marx. Cambridge,
so can lead to all manner of unwanted behavior, in par- England: Cambridge University Press.
ticular to wild oscillations and/or locking into deeply
undesirable configurations (Lynch 1996). In fact, the Frawley, W. D. 1997. Vygotsky and Cognitive Science:
failure modes of distributed systems are strongly remi- Language and the Unification of the Social and Com-
niscent of the pathologies of economic (Chamley 2004) putational Mind. Cambridge, Massachusetts: Harvard
and statistical-mechanical (Young 1998) models of so- University Press.
cial learning, when they are placed in suitable (that is, Galles, D., and Pearl, J. 1997. Axioms of causal rele-
unsuitable) situations. Designing, or reforming, a sys- vance. Artificial Intelligence 97:9–43.
tem for computer-mediate social information processing Harvey, P. H., and Pagel, M. D. 1991. The Compara-
is at once a problem of distributed algorithm design and tive Method in Evolutionary Biology. Oxford: Oxford
a problem of mechanism design, and they two modes or University Press.
aspects should inform one another, as well as empirical Hastie, T.; Tibshirani, R.; and Friedman, J. 2001.
results about what actually happens when real human The Elements of Statistical Learning: Data Mining,
beings use different systems for different tasks. Inference, and Prediction. New York: Springer-Verlag.
Hayek, F. A. 1948. Individualism and Economic Or-
Acknowledgments Thanks to P. Agre, P. Domin- der. Chicago: University of Chicago Press.
gos, C. Genovese, K. Lee, S. Page, W. Tozier, E. Smith, Hong, L., and Page, S. E. 2004. Groups of diverse
N. Snoad, and the participants of the 2002 workshop on problem solvers can outperform groups of high-ability
collective cognition and distributed intelligence at the problem solvers. Proceedings of the National Academy
Santa Fe Institute for valuable discussions, and to K. of Sciences 101:16385–16389.
Klinkner for many reasons (including valuable discus- Huckfeldt, R.; Johnson, P. E.; and Sprague, J. 2004.
sions). Political Disagreement: The Survival of Diverse Opin-
ions within Communication Networks. Cambridge,
References England: Cambridge University Press.
Balkin, J. M. 1998. Cultural Software: A Theory of Hutchins, E. 1995. Cognition in the Wild. Cambridge,
Ideology. New Haven, Connecticut: Yale University Massachusetts: MIT Press.
Jacobs, R. A. 1997. Bias/variance analyses of Newman, M. E. J. 2003. Mixing patterns in networks.
mixtures-of-experts architectures. Neural Computa- Physical Review E 67:026126.
tion 9:369–383. Nitecki, M. H., and Hoffman, A., eds. 1987. Neutral
Katz, E., and Lazarsfeld, P. F. 1955. Personal in- models in biology. New York: Oxford University Press.
fluence: the part played by people in the flow of mass Ober, J. 2005. Athenian Legacies: Essays on the Pol-
communications. Glencoe, Illinois: Free Press. itics of Going On Together. Princeton, New Jersey:
Kitcher, P. 1993. The Advancement of Science: Sci- Princeton University Press.
ence without Legend, Objectivity without Illusions. Ox- Page, S. E. 2007. The Difference: How the Power of
ford: Oxford University Press. Diveristy Creates Better Groups, Firms, Schools, and
Lange, O., and Taylor, F. M. 1938. On the Economic Societies. Princeton, New Jersey: Princeton Univer-
Theory of Socialism. Minneapolis: University of Min- sity Press.
nesota Press. Pemantle, R., and Skyrms, B. 2004. Network forma-
Lerman, K. 2007. Social information processing in tion by reinforcement learning: the long and medium
social news aggregation. IEEE Internet Computing run. Mathematical Social Sciences 48:315–327.
submitted. Popper, K. R. 1945. The Open Society and Its Ene-
Lieberson, S. 2000. A Matter of Taste: How Names, mies. London: Routledge.
Fashions, and Culture Change. New Haven, Connecti- Sood, V., and Redner, S. 2005. Voter model on het-
cut: Yale University Press. erogeneous graphs. Physical Review Letters 94:178701.
Liggett, T. M. 1985. Interacting Particle Systems. Sperber, D. 1996. Explaining Culture: A Naturalistic
Berlin: Springer-Verlag. Approach. Oxford: Basil Blackwell.
Lindblom, C. E. 1965. The Intelligence of Democ- Steglich, C.; Snijders, T. A. B.; and Pearson, M. 2004.
racy: Decision Making through Mutual Adjustment. Dynamic networks and behavior: Separating selection
New York: Free Press. from influence. Technical Report 95-2001, Interuniver-
Lupia, A.; McCubbins, M. D.; and Popkin, S. L., eds. sity Center for Social Science Theory and Methodol-
2000. Elements of Reason: Cognition, Choice, and the ogy, University of Groningen.
Bounds of Rationality. Cambridge, England: Cam- Surowiecki, J. 2004. The Wisdom of Crowds: Why the
bridge University Press. Many Are Smarter than the Few and How Collective
Luria, A. R. 1976. Cognitive Development: Its Wisdom Shapes Business, Economies, Societies, and
Cultural and Social Foundations. Cambridge, Mas- Nations. New York: Doubleday.
sachusetts: Harvard University Press. Toulmin, S. 1972. Human Understanding: The Collec-
Lynch, N. A. 1996. Distributed Algorithms. San Fran- tive Use and Evolution of Concepts. Princeton, New
cisco: Morgan Kaufmann. Jersey: Princeton University Press.
Mancur Olson, J. 1971. The Logic of Collective Action: Turner, S. P. 2002. Brains/Practices/Relativism: So-
Public Goods and the Theory of Groups. Cambridge, cial Theory After Cognitive Science. Chicago: Univer-
Massachusetts: Harvard University Press, revised edi- sity of Chicago Press.
tion. First edition, 1965. Vygotsky, L. S. 1934/1986. Thought and Language.
Marx, K., and Engels, F. 1847/1947. The German Cambridge, Massachusetts: MIT Press.
Ideology. New York: International Publishers. Vygotsky, L. S. 1978. Mind in Society: The Develop-
Mayo, D. G. 1996. Error and the Growth of Exper- ment of Higher Psychological Processes. Cambridge,
imental Knowledge. Chicago: University of Chicago Massachusetts: Harvard University Press.
Press. Walsh, W. E.; Yokoo, M.; Hirayama, K.; and Well-
McNeill, W. H. 1982. The Pursuit of Power: Tech- man, M. P. 2003. On market-inspired approaches
nology, Armed Force and Socety since A.D. 1000. to propositional satisfiability. Artificial Intelligence
Chicago: University of Chicago Press. 144:125–156.
McPherson, M.; Smith-Lovin, L.; and Cook, J. M. Yates, J. 1989. Control through Communication: The
2001. Birds of a feather: Homophily in social networks. Rise of System in American Management. Baltimore:
Annual Review of Sociology 27:415–444. Johns Hopkins University Press.
Mercer, N. 2000. Words and Minds: How We Use Young, H. P. 1998. Individual Strategy and Social
Language to Think Together. London: Routledge. Structure: An Evolutionary Theory of Institutions.
Miller, G. J. 1992. Managerial Dilemmas: The Po- Princeton: Princeton University Press.
litical Economy of Hierarchy. Cambridge, England: Ziman, J. 2000. Real Science: What It Is, and What
Cambridge University Press. It Means. Cambridge, England: Cambridge University
Newman, M. E. J., and Girvan, M. 2003. Finding and Press.
evaluating community structure in networks. Physical
Review E 69:026113.