copyright notice
link to the published version: IEEE Computer, November, 2022

accesses since October 31, 2022


Hal Berghel

ABSTRACT: The recent tumult over the abuses of social medis suggests that this is a good time to revisit crowd science.

What does our recent experience with social media say about the intelligence of crowds? The central thesis of James Surowiecki's 2004 best-selling book The Wisdom of Crowds [SURO] held that “under the right circumstances, groups are remarkably intelligent, and are even smarter than the smartest people in them.” I shall argue that if his thesis holds, it does so only under very limited conditions. In fact, I will flip his thesis on its head and ask under what circumstances the opposite would obtain: where crowds are demonstrably less “intelligent” than the smartest people in them – or perhaps when crowds are less intelligent than anyone in them. I then contrast our recent experience with social media with that of Wikipedia. I will claim that experience with the latter is more supportive of Surowiecki's thesis than the former, although neither may be considered a validation.

I begin with the caveat that while I appreciate some of Surowiecki's observations, I have reservations about his general thesis regarding the value of crowds and groups. I confess a general suspicion of the value of group identification and cohesion as it so commonly leads to social dominance, intergroup conflict, and societal discord. So I'm by nature reluctant to assign any inherent value to group settings, and am inclined to attach any positive properties that might accrue to coincidence and hidden variables. That said, Surowiecki's analysis does have something more important to offer than his main thesis. This arises in his discussion of the criteria that inhibit crowds' collective judgments. There is considerable insight to be gained, particularly with respect to the darker sides of social media.

Surowiecki claimed that collective intelligence might bear on three categories of problems: those that deal with (1) cognition, (2) coordination, and (3) cooperation. He also lists three necessary conditions for crowds to be wise: (a) diversity, (b) independence, and (c) a particular kind of decentralization. My argument is twofold. First, while social media may help address problem categories (2) and (3), it is highly questionable whether social media has much to offer in terms of category (1). Second, in the case of social media, necessary condition (a) rarely obtains and (b) is frequently absent. Thus, it is only to the extent that Surowiecki is allowed to cherry pick his crowds that he will be able to identify confirming instances of his claim.


So what is a crowd? We begin by re-purposing a theory on social organization from economics and the social sciences called “convergence theory” [CONV] that holds that over time groups will converge toward “conditions of similarity” – i.e., differences will diminish. I don't want to carry this point too far, but I do want to emphasize that the idea of convergence as a driver of uniformity seems to be useful in characterizing social media. Crowds are, in this sense, collections of like-minded individuals coming together as one, or a reasonable approximation thereof. Of course, there are many subtleties involved: crowds attracted to crime scenes and spectacles may only be like-minded in the sense of a non-intellectual, morbid curiosity, whereas crowds at Ku Klux Klan meetings may be ideologically bound together by varieties of ethnocentrism. However, these distinctions are best left to social scientists to understand and will not affect our overview of the relationship between crowds and social media.

It is important to recognize that the darker sides of crowds have been observed for over a century. Elias Canetti observes that open crowds have a natural urge for growth and want to grow indefinitely in his 1960 book, Crowds and Power. [CANETTI] But, “one of the striking traits of the inner life of a crowd is the feeling of being persecuted, a peculiar angry sensitiveness and irritability directed against those it has once and forever nominated as enemies.” Canetti suggests that we view crowds as besieged cities, attracting more partisans form the countryside and bonding them together through a feeling of being persecuted. An external attack only serves to strengthen a crowd, so the ultimate destruction of a crowd will likely result from internal panic or disorder. Thus, on his view, crowds don't naturally become smarter. They become more partisan and defensive. This is in clear contrast to the main thesis in the Wisdom of Crowds.

Let's look at crowds from Canetti's perspective. Canetti identifies four fundamental qualities of crowds: the insatiable desire to grow, absolute and unquestionable equality between members, perceived density and indivisibility of members, and a shared, unattained goal. This does not describe either an intellectual bond or purposeful reflection.

The appropriateness of Canetti's observations to a study of present online crowds like social media should not be overlooked. Although his book was written in 1960, it remains relevant today. It should be mentioned that Canetti's work fits into a tradition of critical crowd analysis that spans more than a century and is inconsistent with The Wisdom of Crowds. Similarly, sociologist Gustave Le Bon's characterization of the psychology of crowds emphasized irresponsibility, herd mentality, irrationality, and impulsivity characteristic of “inferior forms of evolution…” which allows them to be “easily led into the worst excesses.” [LEBON] Le Bon argues that crowds are not influenced by reason, and what limited reflection they do sustain is of a “very inferior order.” Le Bon and Canetti hold views antithetical to those in the Wisdom of Crowds, but for different reasons.

In response to Le Bon, Surowiecki claims

Gustave Le Bon had things exactly backward. If you put together a big enough and diverse enough group of people and ask them to “make decisions affecting matters of general interest,” that group's decisions will, over time, be “intellectually [superior] to the isolated individual,” no matter how smart or well-informed he is.” [SURO]

While my inclinations are to side with Canetti and Le Bon, Surowiecki should not be ignored. But we must recognize that his confidence in crowds is measured through, and hinges on, a major caveat: the necessary conditions for wise crowds being both diverse and independent. I shall claim that these necessary conditions are absent in many if not most online crowds, and for that reason the Canetti and Le Bon analyses of crowds seem to be a better fit for social media than Surowiecki's.


I admit to the bias that my life experience suggests that it is nearly impossible to over-estimate the credulity of crowds generally, whether religious, political, sports crowds, investment clubs, scout troops, concert goers, mobs, riots, protests, panics, etc. All crowds share at least one common feature: common focus. As a result, they are self-influencing and self-reinforcing. But these are not the only debilitating features of crowds. Crowds are far worse when they are populated through self-selection. Crowds tend to behave anti-magnetically – opposites are not attracted to one another because of this singularity of focus. If I may stretch the metaphor a bit more, as crowds grow so does the “ideological bond” that connects the members, which in turn further discourages diversity. Over time, crowds become herds/hives/swarms –whatever label one chooses to use. Think about this in terms of identifiable crowds within your sphere of observation. How welcoming are polygamous cults to ideologically monogamous potential recruits? How many Antifa signboards do you see at stop-the-steal rallies? Are arena seats randomly distributed to fans of opposing teams? Although there may be exceptions, self-selection works against the very variety and diversity that Surowiecki claims are necessary for wisdom to arise. Singularity of focus and self-selection seem to me to provide sufficient grounds for general distrust of the wisdom of any crowd.

Furthermore, crowds seem to face any challenge to their singular focus with reservation, if not outright rejection. My experience suggests that this feature is accountable for crowd willingness to accept disinformation, fake news, unsupported claims, conspiracies promoted by crowd influencers, and the like. In this way, as Canetti and Le Bon observed, crowd mentality inevitably tends toward herd mentality – especially as the membership grows and matures, and the foci narrow and/or multiply.

But I am not willing to discard Surowiecki's claims altogether. I concede that some crowds can exhibit sagacity, but only when the membership is carefully controlled. The primary villain is self-selection. Absent vetting, crowds will be most attractive to those who have illiberal, intolerant, and narrow-minded attitudes regarding the ideological polestars of the group. That is, if the crowd self-identifies with a particular cause, it is to be expected that measured reflection on fundamental principles will be unacceptable to the group. Convergence theory suggests that such a herd instinct is an inevitable feature of focused, mature crowds. So while I'm willing to admit that some crowds can make good choices as Surowiecki suggests, his objection to Le Bon was over-zealous. Even larger and more diverse crowds are capable of making larger and more diverse mistakes. I will call the thesis that crowds naturally decay into herds naïve crowd psychology .

Of course, Surowiecki's caveat allows an effective – though circular - escape from our intuitive and naïve crowd psychology. He makes his most plausible argument against this in his chapter on the value of diversity. The problem with his argument lies in its circularity. Suppose that one may legitimately describe a position taken by a certain crowd as mistaken, incorrect, or unjustified, but that conditions b) and c) above were satisfied. The caveat that the crowd was insufficiently diverse could always be used to explain the error. The problem is that the diversity caveat does not allow a non-vacuous alternative – i.e., there is no way to falsify it. We don't have to go full-tilt Karl Popper here with a carte blanche endorsement of the falsifiability principle, but rather content ourselves with the fact that Surowiecki's three conditions aren't testable. They're definitional. This is reminiscent of the elephant bane gambit I described some years back [ELEPH] whereby the use of chemical repellant could be used to explain the absence of pachyderms on Antarctica. If elephants are never found on the Ross ice sheet, it could be claimed that the elephant bane worked; else, not enough was used. It isn't difficult to find examples of the elephant bane rhetorical tactic in the media - especially in online resources that deal with religion, politics, and vexing social issues.


Jaron Lanier likens the use of the term “wisdom” in the context of crowds to Adam Smith's “invisible hand” in the context of market exchange. [LANIER] His point is provocative. In both of these cases, the attribution has a spoofy and gratuitous character. In the spirit of Bob Dylan, we might say that there seems to be something happening here, but we don't know what it is. The question arises whether wisdom and invisible hands are appropriate descriptors in these contexts. Tendencies to impart human qualities to non-human, and in many cases imaginary, objects have accompanied the entire human experience, so use in this context should not be surprising. These tendencies are so common that social scientists have given them names such as anthropomorphism, apotheosis, and euhemerism. That the use of this phenomena to rationalize mythology and religion has been documented for millennia. It is even a staple in fables, fairy tales, animated media, video games, and emojis for that matter. But one question always remains: do such uses actually add any explanatory value?

Lanier suggests that crowd wisdom might be a corollary to the Delphi method of forecasting. [DALKEY][HELMER] But that seems only partially right. The Delphi method relies on a structured panel of experts, not the collective wisdom of relatively random crowds. Surowiecki is clear about the difference between crowds and panels of experts: “Even if most of the people within a group are not especially well-informed or rational, it can still reach a collectively wise decision.” [SURO] I'm unwilling to concede the congruence between crowds and the Delphi method that Lanier sees.

But Lanier's overall skepticism about crowds seems reasonable. He circumscribes the limits of crowds this way: “ The collective is good at solving problems which demand results that can be evaluated by uncontroversial performance parameters, but bad when taste and judgment matter,” while admitting that “Collectives can be just as stupid as any individual, and in important cases, stupider.” [LANIER] He then offers a set of conditions where crowds and collective assessment may be superior to that of an individual: (1) when the crowd isn't defining its own question, (2) when the question leads to a simple result (e.g., Y/N or numeric value), and (3) when the information sources behind the assessment are appropriately filtered. “Break any one of this conditions and the collective becomes unreliable or worse.” Once again, caveats rear their ugly heads and we're back to the elephant bane gambit. What criteria do we apply to ensure that our information sources have been appropriately filtered? The problem is in general that we have no better insights into whether these caveats are satisfied than we have of whether the opining of a crowd/collective/herd/hive is reliable in the first place.

In Lanier's terms, crowds exhibit their worst behavior when they take on a “hive mentality”: a hive mind is a cruel idiot when it runs on autopilot.” [LANIER] All too frequently what we see in current social media is unreflective partisan tribe/crowd/herd/hive outbursts, and sub-cerebral emanations from ill-suited, un-prepared, and undisciplined minds. QAnon is a perfect example of the “cruel idiocy” of a hive mind. [QANON] Although the same would apply to other online resources - Breitbart, Newsmax, InfoWars, and the One America News Network come to mind. It is with social media outlets that the hive mind achieves maximal effect, and for this reason should be of greater concern to society. [ANTI]


About ten years ago I wrote in Computer that since n ot all crowd members are equally well-informed, trustworthy, or reliable, you can't rely on a crowd to filter out nonsense. As I put it then, “ Crowds,
like landfills, may produce treasures, but the yield rate isn't encouraging.“ [STICKY] As an illustration of the problem, I drew attention to an edit skirmish that took place in January, 2013. I documented that according to the first sentence in the Wikipedia article about him, the characterization of then Secretary of Defense Chuck Hagel went from “an American politician who was a United States Senator” to “an American politician, anti-Semite and pro-terrorist who was a United States Senator,” and back again in the span of a few hours. In this case, the combative contributor qua tribalist wasn't trying to bury the lede, but rather bury Hagel's reputation. This is a glaring example of the edit war problem that wikis face when contributors try to inject self-serving, malicious, defamatory disinformation (aka: nonsense) into a record or narrative in furtherance of their non-reality-based world view. To be sure, submissions are routinely reviewed by wiki volunteers and overseen by in-house editors who also check them for appropriateness. In the case of the Hagel edit skirmish, the disinformation was so blatant that it was caught quickly. But subtleties are not so easily spotted and nuanced suggestions may go unnoticed. In the case of Wikipedia, the editing oversight issue is significant. [WIKIEDIT] While Wikipedia articles may fray around the edges, and would not meet the peer-review standards of scholarly publications, in general Wikipedia works well enough to be useful so long as (1) the topics are not controversial, (2) the issues are not nuanced, and (3) not much rests on the accuracy of the content – i.e., the topic is relatively unimportant. Despite these advantages, it is potentially toxic to serious scholarship and for that reason is largely avoided.

As with social media, controversial wiki topics attract tribalists and partisans of every stripe. Armed with disinformation, they seek to manipulate a public narrative. When it comes to social media, their weapons of choice include sockpuppeting (i.e., pseudonymous manipulation of online resources to simultaneously distance themselves from a position or action and give the appearance of objectivity), catfishing (the use of fictional identities to target online victims), and gaslighting (i.e. pseudonymous manipulation to produce victim self-doubt and distress). These techniques rely on anonymity, obfuscation, and trickery to avoid criticism and backlash directed back at the sources. To paraphrase Jaron Lanier, this veil of anonymity amounts to an online cultural denial-of-service attack on users. An “invisible social vandalism” results. [LANIER2]

The reality is that these tactics are as difficult to detect [KATS] as the underlying personality disorders behind them, [MILLON] thus providing an insurmountable challenge for media or news-oriented online platforms who report on them. In just this way, wiki and social media platforms are also challenged to recognize, register and respond to subtlety and nuance which makes the problem of detecting half-truths, vagaries, and misinformation as difficult as detecting falsehoods, lies, and disinformation. Any intellective product of an anonymous crowd will be more difficult to unravel than that of an identifiable individual. As I suggested in my earlier article, the inability to achieve consistent, reliable vetting through peer review by knowledge domain experts was the reason that Tony Ralston and I abandoned our wiki, the ACM Timeline of Computing in the late 1990s. [STICKY] Finally, Wikipedia consumers are not trying to replicate scientific experimental results or conduct scholarly research based on primary sources and ground truth data. They are looking for an entry-level expedient overview where imprecision and inexactitude are acceptable. Wikis can have utility as long as we don't place much confidence in them.

From my experience, Wikipedia excels at the mundane: dates, quantities, names, places – information that is incontestable and uncontroversial. If there is information that is beyond dispute, there is a good probability Wikipedia's presentation will be reasonable. But just one step beyond the incontestable, credibility quickly wanes. This is not to say that credibility vanishes altogether, but it suffers considerably. Even though Wikipedia has added sophistication to the editorial process and seems to have eliminated the pendulous swings of the edit wars, one must remain mindful of its limitations.


So the core question should not be whether or to what extent crowds of any stripe are wise. That's a category mistake. Rather, we should ask whether they will be naturally drawn to the dark side. Our recent experience with social media, especially when it comes to issues of politics, religion and anti-social behavior, demonstrates the enormous potential of crowds for banality. While Wikipedia shows that online crowds can be reliable sources of information in some situations, the same can't be said for QAnon and 4chan. There is ample evidence that crowds can have a far darker, anti-social character. They can be untrustworthy, [QANON] abusive [LANIER2][BULLY][TWITLIT], and easily manipulated [CA] [NEOPETS], for example. We might go so far as to say that W ikipedia reveals crowds at their best; social media, at their worst.

So what we're left with is general suspicion of online crowds in terms of reliability, tempered by the observation that sometimes, and under controlled circumstances, crowds can have utility. We are forced, however, to recognized that the value of social media crowds can be inferred from their collective behavior. While our experience with online crowds and social media does not completely undermine Surowiecki's confidence – remember the all-important caveats that he built into the Wisdom of Crowds - it does suggest that there was much more to be gained from a careful study of Le Bon, Canetti and Lanier than Surowiecki.

While a formal study of the interrelationships between social media and crowds would involve a social science research effort, a useful informal approximation may be obtained by listening to AM talk radio, as the call-in crowds share similar motives with online crowds. From my experience both sources seem to display comparable levels of fervor, focus, resentment, alienation, and hostility. While not a scholarly study, listening to AM radio is a window into the darker sides of crowds. But unlike social media crowds, AM radio presents “crowdspeak” to all who care to listen without the filter of self-selection. But in both cases, crowds will continue to morph into anti-social packs, herds, hives, tribes, mobs, etc.. With AM radio, however, at least this can be monitored.

We note also that there is an important distinction to be made between crowd organizers and leaders on the one hand and crowd members and followers on the other. Our focus on the collective wisdom of crowds enables us to ignore this distinction, without diminishing its importance. We observe that while the desire to exercise power by leading a crowd may be qualitatively different than the desire to exercise power as an individual, both cases may share a common pathology. More refined analysis is best left to the social sciences.

It is also appropriate to question whether credit given online crowds to support social movements is justified. The evidence supporting the efficacy of the so-called “Twitter revolutions” is sketchy at best, and may be totally overblown. [MOROZOV][ESFAND]

Finally, we emphasize again that self-selection and singularity of focus are the most corrosive aspects of crowds. If online crowds encouraged unrestricted membership, diversity of opinion, and non-rectified information flows, they would take on a less acrid character. But then, they wouldn't serve the ultra- partisan, non-reality-based communities as well.


REFERENCES (all links current as of 7/28/22)

[SURO] J. Surowiecki, The Wisdom of Crowds, Doubleday, New York, 2004.

[CONV] "Convergence Theories ." Encyclopedia of Sociology . Retrieved June 03, 2022 from

[CANETTI] E. Canetti, Crowds and Power (tr. Carol Stewart), Farrar, Straus and Giroux, New York, 1984. (available online: (registration required))

[LEBON] G. Le Bon, The Crowd: A Study of the Popular Mind, Macmillan, New York, 1896. (available online: )

[ELEPH] H. Berghel, "Secretocracy," in Computer , vol. 49, no. 2, pp. 63-67, Feb. 2016, doi: 10.1109/MC.2016.61.

[LANIER] J. Lanier, DIGITAL MAOISM: The Hazards of the New Online Collectivism, Edge, 5.29.06. (available online: )

[DALKEY] N. Dalkey and O. Helmer, An Experimental Application of the Delphi Method to the Use of Experts, Memorandum RM-727 (abridged), Rand Corporation, July, 1962. (available online: )

[HELMER] O. Helmer and N. Rescher, On the Epistemology of the Inexact Sciences, Rand Report R-353, Rand Corporation, 1960. (available online: )

[QANON] H. Berghel, The QAnon Phenomenon: The Storm Has Always Been Among Us, in Computer , vol. 55, no. 5, pp. 93-100, May 2022, doi: 10.1109/MC.2022.3154125. (available online: )

[ANTI] H. Berghel, New Perspectives on (Anti)Social Media, in Computer , vol. 53, no. 3, pp. 77-82, March 2020, doi: 10.1109/MC.2019.2958448. (available online: )

[STICKY] H. Berghel, "Sticky Wikis," in Computer , vol. 47, no. 9, pp. 90-93, Sept. 2014, doi: 10.1109/MC.2014.263. (available online: )

[WIKIEDIT] Wikipedia: Editing Policy (online):

[LANIER2] J. Lanier, Ten Arguments for Deleting Your Social Media Accounts Right Now, Picador Reprint, New York, 2019.

[KATS] D. Kats, Identifying Sockpuppet Accounts on Social Media Platforms, Norton Labs blog, 29 APR, 2020. (available online: )

[MILLON] T. Millon, S. Grossman, C. Millon, S. Meagher, R. Ramnath, Personality Disorders in Modern Life (2 nd ed.), Wiley, New York, 2004 (available online: )

[BULLY] S. Hinduja, J. Patchin, Identification, Prevention, and Response, 2021 Edition, Cyberbullying Research Center, (available online: )

[TWITLIT] H. Berghel, Weaponizing Twitter Litter: Abuse-Forming Networks and Social Media, Computer , vol. 51, no. 4, pp. 70-73, April 2018. (available online: )

[CA] H. Berghel, Malice Domestic: The Cambridge Analytica Dystopia, Computer , vol. 51, no. 5, pp. 84-89, May 2018, doi: 10.1109/MC.2018.2381135.

[NEOPETS] S. Lock, Neopets security breach: users' data reportedly stolen, The Guardian, 22 Jul 2022. (available online: )

[MOROZOV] E. Morozov, Net Delusion: The Dark Side of Internet Freedom, PublicAffairs, New York, 2011).

[ESFAND] G. Esfandiari, The Twitter Devolution, Foreign Policy, June 8, 2010. (available online: )