Preface
This is not another book on network analysis, despite the fact that both “networks” and “analyses” figure prominently in its pages. We were motivated to write it by a gap that we observed in network analysis as it relates to the epistemic underpinnings of social networks—specifically, the gap in our understanding and in our representations of what networked human agents know or believe. Bridging this gap is necessary for network-based explanations of behavior and a genuine representation of network dynamics, but it cannot be straightforwardly done by work in fields such as epistemic game theory or artificial intelligence, which emphasize formal models for dealing with interactive states of knowledge and belief. For this reason, this book introduces a language that researchers can use to explain, predict, and intervene in the epistemic fabric of social networks and interactions.
Because we are building a language that is meant to be used (and perhaps sometimes abused), it is useful to think of this not only as a book but also as an “application”—or “app”—in the computer software sense of the term. An app is a set of representations and the procedures for manipulating them that allows users to accomplish new tasks. Think of Microsoft Excel, Google Chrome, or the video game Rock Band. An app should be both usable and useful. Unlike a “theory,” which lives in a purely representational space, an app is embodied and made useful through repeated use. Thus, our goal is not simply to introduce another way to describe the cognitive and epistemic states of networked agents but to do so in a way that is “plug-in compatible” with the discursive and empirical practices of the fields that study social networks.
We owe a debt of gratitude to several people who have given generously of their time and energy to help us build this edifice. In particular, we thank Ron Burt, for his enthusiastic support and insightful commentary and suggestions throughout; Raluca Cojocariu, for her detailed, exacting, and attuned editorial and production assistance; Tim Rowley and Diederik van Liere for sharing network data and assisting with the collection of additional data used in the analysis of trust in Chapter 4; and two anonymous Stanford University Press reviewers for their detailed comments and suggestions for improvement.
Finally, this work would not have come to fruition at all but for the expert, caring, and patient guidance of our editor, Margo Beth Fleming, over the past three years. We thank her deeply.
MM and JB
Chapter 1
Using examples and unstructured intuitions that highlight the importance of knowledge and of beliefs, both individual and mutual, to the outcomes of social situations and interpersonal relations, we argue for the usefulness of explicit epistemic models of human interactions and networks. We introduce the notions of an epistemic state—that is, a link between individuals and propositions they may know or believe—and of an epistemic tie—that is, a connection between individuals’ epistemic states: if Alpha knows Beta knows Gamma knows that the park is closed after dark, then there is a set of epistemic ties connecting Alpha, Beta, Gamma, and the proposition that the park is closed after dark, which is part of the epistemic structure of the situation. We show how the structure of epistemic networks—epinets—formed by such links among individuals and their beliefs is relevant to the dynamics of human interactions, and how the dynamics of these networks are critical elements of complex interpersonal narratives.
. . .
What must human agents know about what other humans—with whom they are connected—know in order for the resulting patchwork of ties among them to function as a social network? Suppose that an anonymous survey of the friendship network of a seven-person executive team reveals that Beth, Harry, and Martha form a clique, with each describing the others as “friends.” We designate the triad as a clique, rather than as a patchwork of ties, because we expect these three to exhibit some special forms of cohesion that may be evidenced by, for example, an above-average ability to coordinate, collaborate, communicate, and collude. In other words, we expect the triad to function as a clique: we expect each member to know—and know that the other two know—sensitive information about an event of mutual importance, or we expect that such sensitive information will quickly propagate within the triad.
What each knows of and about the others and their knowledge is the “epistemic glue” of the clique; it is what allows Beth to react to an unforeseen disaster in ways she knows Harry and Martha will find justifiable, and it is what allows her to make sense of their intentions based on observing their reactions and knowing what they think about what she knows. The grammar is somewhat complicated, but its complexity closely tracks that of the phenomena we expect a clique to exhibit. This epistemic superstructure is what makes the clique a clique—an identifiable network substructure with very specific expected properties—rather than a patchwork of ties and connections that can offer no further insight or predictive value.
At a more fundamental level, what must human agents know or believe about what others know or believe for their interactions to have joint or shared sense and meaning and to lead to stable patterns of interpersonal behavior? Game theory has contributed a basic canon of coordination, cooperation, and collaboration “games” that require coherent mutual beliefs (players’ beliefs about other players’ beliefs, about other players’ beliefs about their own beliefs, and so on) whose “epistemic structures” can be analyzed to arrive at the preconditions for coordination, cooperation, collaboration, and even coherent communication. However, these neat analytical structures come at the cost of oversimplifying what humans believe and how they believe it as well as what they know and how they know it. States of knowing, like “oblivion” (not knowing and not knowing you do not know) or “forgetting” (knowing but not recalling what you know), are ruled out by assumptions such as those of “common priors” and “common knowledge,” even though these states are all important to the unfolding of real human interactions. Moreover, because the event spaces of game theory do not admit interpretations and shadings, the resulting analyses lack the subtlety required to understand that humans interpret “Can I pray while I smoke?” very differently from the way they interpret “Can I smoke while I pray?” The conjunctive “while” functions very differently in first-order logic from the way it functions in plain English.
The contemporary importance of epistemic moves and games to an understanding of social interactions is clear from the direction of technological progress and innovation. Homo sapiens is Homo communicans and makes use of the full range of methods for passing information-bearing signals to shape, control, and predict the social milieu of being in the world. Consider the “cc” (carbon copy) and “bcc” (blind carbon copy) functions of everyday e-mail, which act as levers for shaping the informational structure of an interaction: “cc” creates pools of mutual knowledge about the contents of a message and serves as an aggregation tool; “bcc” oppositely brackets cliques that are “in the know” from individuals outside a circle of trust or power. But these are just the rudiments: new technologies allow senders to control a message after they have sent it—and possibly delete it—making it possible for them to deny ever having sent such a message even though they know the recipients know its contents; to “hack” into each other’s e-mail servers to access a critical message without the message’s author or recipient knowing that the hacker knows its contents; and to encrypt a message so that only intended recipients can decode it on the basis of access to the public or private key with which it has been encoded. The complexity of “interactive epistemology” has multiplied over the past few decades and continues to do so. A new language and new models are needed to understand the epistemic glue of human social interactions.
Although we are interested in building intuitive, yet precise, models of this epistemic glue, we are assuredly not pioneers of the epistemic dimension of social interactions. Nuanced treatments of epistemic structures and effects have appeared in the fields of artificial intelligence (Fagin et al. 1995), epistemic game theory (Aumann 1989), and analytic philosophy (Kripke 2011). Nor are we the first to point out that social networks (and social structures more generally) require descriptions sensitive to differences between what social agents think, and what we think social agents think, about such structures (Krackhardt 1987). What we are after is a tool kit for modeling, measuring, and manipulating the epistemic glue of human interactions and networks in ways that are as accessible to social network analysts as they are engaging to logicians, epistemic game theorists, and artificial intelligence researchers. We are building an application—an “app”—as much as we are theorizing, modeling, or philosophizing.
The Epistemic Structure of a “Friendship Tie”
Because we are building an app as much as a theory, we need to become intimate with “user requirements”—that is, the kinds of uses to which our modeling tool kit may be put. To that end, consider the friendship tie between Beth and Harry in our earlier example. Beth “knows” Harry: she sees him daily, is familiar with his latest setbacks and successes, works with him on a joint project, and sees him socially about every other week. That is her longhand unpacking of the shorthand answer “Harry is my friend,” which she gave on the anonymous survey. Now what we want to know is this: when Beth needs Harry to convey to her, quickly and covertly, a sensitive piece of information she believes he has received from one of his acquaintances with whom Beth has no connection, what must Beth know about Harry for her to have good reason to believe that he will come through with it?
The minimal set of beliefs about Harry that Beth needs to rationalize her expectations may include the following: she believes that Harry knows the information is useful to her, that it is important to her that their office mate, Martha, does not know Beth has come to know it, and perhaps that Harry knows Beth will not divulge her source after he has passed along the information. Complications can arise: if Beth knows Harry’s boss knows of Harry’s ties to Beth and is monitoring Beth’s actions to detect any sign that Harry has leaked to Beth the information he was entrusted to hold in confidence, then Beth may have to believe Harry knows of this threat and trusts in her integrity and competence not to “blow his cover.” Alternatively, she may have to believe Harry does not know about this threat (in which case she may choose to inform him of it as befits the level of trust she assumes they share).
In each case, there is a structure to the knowledge that these social agents “share” that is both intelligible and intuitive, although it grows quickly in logical complexity with the addition of new information and people. The structure of this “epistemic glue” is rendered intuitive and intelligible by the recursive and interactive nature of what this “social knowledge,” as it should properly be called, relates to, which is often knowledge about knowledge: Beth’s knowledge about Harry’s knowledge, which includes Harry’s knowledge about Beth’s knowledge about Harry’s knowledge, and so on. The structure of the epistemic glue is “interactive”: it links not only an agent’s mind to a proposition but also one agent’s mind to a proposition via another agent’s mind: Beth knows Harry knows that his boss is monitoring Beth’s actions for any sign of information leaked by Harry.
Of course, it is not only knowledge but beliefs, conjectures, and even barely articulated hunches that we want to capture and address with our language. Beth may not know—by any acceptable use of the term “knowing”—that Harry knows that the piece of information he possesses should be transmitted to her in a way that guards against eavesdropping—but she may simply believe it for reasons having to do with a complex of other prior beliefs. Harry may merely “sense” that Beth needs him to transmit the information covertly, without really having fully articulated that hunch as a proposition. All of these are legitimate objects of modeling, measurement, and manipulation for our app. We need a comprehensive conceptual framework to engage the range of states in which humans find themselves vis-à-vis propositions about the world. To study the gamut of mental objects playing pivotal roles in the relationships that form the fabric of human networks, we use the covering term epistemic states and represent these states as directed ties between social agents and propositions, which are grammatically correct sentences in natural language or well-formed formulas in propositional logic. And we use the term epistemic networks or epinets to refer to the networks that comprise a group of social agents, potentially true or possibly true propositions, and the interactive (“I think you think . . .”) and recursive (“I know I know . . .”) epistemic ties among them.
Let us give meaning to these words through usage. If Beth knows Harry knows that she needs him to convey a piece of sensitive information to her quickly and covertly, there is an epistemic sequence of ties connecting Beth to Harry to the fact that she needs him to act in such and such a way: as an agent to another agent to a proposition. If Harry knows that Beth knows this, then there is an epistemic tie sequence that connects Harry to Beth to Harry to the proposition in question; if Beth knows Harry knows she knows this, then there is a further epistemic tie sequence representing the correspondent connections; and so on. One can add agents (Martha, Harry’s boss) and propositions (the boss’s vigilance) to the epistemic network, as well as additional interactive epistemic ties involving one agent (“Beth knows that she knows what she knows”), two agents (“Beth knows Harry knows she knows . . .”), or more (“Beth believes Harry believes that his boss believes that she believes . . .”).
Epinets in Shakespeare, Kesselring, and Durrenmatt. Epinets evolve and, as they do, “things happen,” socially speaking. Far from being epiphenomenal—appearances sine consequences—changes in epistemic networks are those on which “the plots turn.”
In Shakespeare’s Othello (circa 1603; Ridley 1963), epinets representing the epistemic states of Othello, Desdemona, Iago, and the audience are essential to the interpretive schemata that allow the audience to understand and become involved in the play. The drama can be understood as the evolution of the epistemic states of its lead protagonist, Othello, from one of oblivion and trust through those of doubt and suspicion to one of certainty about a false belief regarding Desdemona’s infidelity. The epistemic state changes can be traced to manipulations by Othello’s lieutenant, Iago, that make careful use of the structure of interactive epistemic states—what Othello thinks Desdemona thinks when she says what she says; what he thinks she thinks he thinks; and so on).
At the beginning of the play, Othello believes unconditionally in Desdemona’s loyalty and faithfulness. He trusts her, in the sense of believing that she could not evince fidelity and love if she did not have these feelings. Also, Othello is oblivious to the possibility that Desdemona is attracted to the young Cassio—one of his lieutenants and a rival to his chief lieutenant and schemer, Iago—in the sense that he gives this possibility no thought. It is not that, if asked about Desdemona’s putative relationship with Cassio, he would say, “I do not know” or “I do not believe it is true”; rather, he would be shocked by Iago’s presumption and by the suggestion of such a possibility. At the same time, Othello is jealous of Desdemona, yet not aware of his jealousy, and so is an easy target for Iago, who seeds doubt in Othello’s mind regarding Desdemona’s fidelity and manages to augment it by playing with what Othello thinks Desdemona thinks and with what Othello thinks Desdemona thinks he thinks: when Desdemona realizes that Othello is now both jealous and suspicious, she attempts to endear herself through loving entreaties. However, Iago has also managed to plant in Othello’s mind that Desdemona is a master dissembler, and therefore Othello interprets these entreaties as masterful dissimulations of faithlessness rather than avowals of love.
Some outright lying and trickery are required of Iago to wean Othello from his trust of Desdemona, but his deceit succeeds because of his prior success in setting up the right epistemic states among Othello, Desdemona, Cassio, and himself. In particular, Othello believes Desdemona is unfaithful; he knows she does not know with whom he suspects her of being unfaithful; he believes she knows he is suspicious; and therefore he thinks she is likely to exaggeratedly “protest” her love for him, Othello, so as to lay his suspicions to rest. When Desdemona—oblivious to much of this epistemic structure—intervenes in favor of her presumed lover, Cassio (who has in the meantime fallen from Othello’s grace after a contretemps into which he has been cunningly pulled by Iago), the epistemic trap Iago has set for Othello springs exactly as planned. Othello sees the situation as one of figuring out with whom his wife is cheating rather than whether or not she is cheating at all. Iago’s soliloquies keep the audience informed of the dynamics of Othello’s epistemic states, contributing to the indignation the audience feels as it witnesses the epistemically trapped Othello smother Desdemona in a fit of jealous rage—an indignation that is assuaged, though perhaps only partly, when a finally awakened Othello at last sees Iago for the vile manipulator he has become and attempts, unsuccessfully, to kill him.
In Joseph Kesselring’s Arsenic and Old Lace (1942), Mortimer Brewster, the central character, is deciding whether or not to fulfill his promise to marry the woman he loves, within the complicated emotional landscape of his family, which includes two elderly spinsters, Abby and Martha—who, while passing as “good Christians,” specialize in poisoning lonely old men with a home brew of elderberry wine laced with cyanide and arsenic; his brother Teddy, who believes he is Teddy Roosevelt and digs graves for the spinsters’ victims in Mortimer’s cellar thinking he is digging locks for the Panama Canal; and a murderous brother, Jonathan. The characters are oblivious to each others’ beliefs and intentions—they do not know and do not know they do not know—and the only bits of information that are common knowledge are either false (the spinsters are good Christians) or relatively useless (Teddy thinks he is Roosevelt). It is this oblivion—of which the audience is aware—that creates the tension between what the characters say, do, and cause one another to think, and what the audience knows they know, as illustrated in Figure 1.1.
Friedrich Durrenmatt takes the strategy of building dramatic tension based on interactive epistemic knots and tangles to a higher level in The Physicists (Die Physiker) written in 1961 (Kirkup 1994). The play is set in a sanatorium for the mentally ill run by Dr. Mathilde von Zahnd, a famed psychiatrist, and features three patients—Beutler (who “believes” he is Sir Isaac Newton), Ernesti (who “believes” he is Albert Einstein), and Mobius (who “believes” he is visited regularly by King Solomon). “Einstein” has just murdered one of his nurses, and the police investigation turns up the fact that Newton had earlier murdered another nurse. “Believes” is in quotation marks because it represents a far more complex interactive epistemic state than we expect, which the second act elucidates—specifically, the fact that each inmate pretends to believe in a false identity for the benefit of the staff and Dr. von Zahnd to more convincingly feign the madness required to cover up his true identity and so allow him to stay in the asylum. Mobius is a renowned physicist who has checked in to protect his inventions from government exploitation, and he is being tracked by two foreign spies—“Newton” and “Einstein”—who are interested in appropriating one of his inventions to their statist ends. Meanwhile, Dr. von Zahnd, in order to appropriate Mobius’s invention for herself, pretends to believe that “Newton” and “Einstein” are who they say they are and that Mobius is deranged.
FIGURE 1.1 Epinet: Arsenic and Old Lace
As illustrated in Figures 1.2 and 1.3, the plot of The Physicists turns not only on an epistemic network that includes states of knowledge, uncertainty, and oblivion regarding both facts and intentions, but also on an evolution of that network—between the two acts—even though the facts remain consistent. These facts are the murders of the nurses and the unfolding action in an asylum. However, this network upends the viewer’s interpretation of these facts (e.g., that “Einstein” and “Newton” are mad and that Dr. von Zahnd is their caregiver).
“Everyday” Epistemic State Tangles
If we are correct that interactive and reflective epistemic states make a large difference in explaining and understanding human social behavior, we should be able to show how epistemic states and the epistemic links that connect them produce insight into common interactive situations and predicaments. We do so through the following vignettes.
Vignette 1. Alice is seeking a job in Bob’s firm. She is being interviewed by Bob for this purpose. Alice made some false statements on her resume, which Bob has before him while the interview is proceeding—a notable one about having won an engineering contest in college in 2010. Alice is unaware of the fact that Bob knows that his own son had actually won an engineering contest that year at the same college. And she does not know (or know she does not know) that Bob suspects—though he does not know—that the contest he knows his son to have won is the very same contest Alice claims to have won. Thus, the credential that Alice has given on her resume is suspect to Bob—who also comes to suspect Alice of lying—but Alice believes her deception is working.
As the interview unfolds, Alice thinks Bob believes her. Bob does not believe her. He also knows, from her facial expression, that she believes her deception has been successful. Because she is certain of this success, Alice does not obsess any further about who thinks what about whom and proceeds oblivious to the inferential dynamics that are developing. Bob manages to keep his conjectures about Alice’s sincerity hidden, which leaves Alice in her turn deceived about the success of her own deception.
FIGURE 1.2 Epinet: The Physicists, Act I
FIGURE 1.3 Epinet: The Physicists, Act II
Vignette 2. Alan sends an electronic message to Belinda and, without specifying that he is sending it anywhere else, also sends the message to Charles, who is Belinda’s boss, via a “bcc.” Thus, Alan thinks Belinda knows the contents of the message, and Belinda thinks Alan thinks she knows (absent any other considerations); Alan thinks Charles knows, and Charles thinks Alan knows (absent any other considerations). However, Belinda does not think Charles knows—otherwise, the “cc” would have appeared in the message, even though Charles thinks she knows and also thinks she does not think he knows. Charles can therefore “play” with Belinda, assuming that she does not think he knows. Charles also knows Alan thinks Belinda does not think Charles knows, and therefore can also wield some power over Alan by credibly threatening to respond to Alan’s message with a “cc” of the response to Belinda, on which Belinda’s original “bcc” will appear.
Vignette 3. The CEOs of three large consumer goods firms selling undifferentiated products to a homogeneous market decide to coordinate their pricing of a particular product in order to prevent a new competitor from entering their market space. Because they do not want to get together and explicitly set prices for the critical period in which the new competitor’s threat of entry is high—lest they be found out and pursued by antitrust authorities—they must coordinate tacitly. Accordingly, by a series of sequential cuts or hikes, they aim to arrive at a mutually agreeable price. The goal of the exercise is common knowledge among the oligopolists—every one of them knows it, knows that every other one knows it, and so forth, ad infinitum. Also, common knowledge is the fact that any oligopolist can use the disguise of a rational response to the threat of entry by a new competitor to steal business from incumbent rivals. Should one oligopolist implement a severe price cut (down to or below his marginal cost, for instance), the others will infer a breach of trust that they would not have inferred had the structure of the preemption game—as well as the possible excuse given by the renegade oligopolist—not been common knowledge among them.
Trust, in this case, keeps the price-fixing game from turning into a “Bertrand game” in which the oligopolists dissipate their profits by bidding all of the prices down to their marginal costs. This trust is predicated on a base of mutually shared facts and premises that each oligopolist knows all of the others know, as well as a set of shared premises that allows each oligopolist to make sense of every other oligopolist’s actions in the context of the price-setting game (such as the fact that an oligopolist can fake panic at an imminent new threat and slash prices in response, but would do so only provided that every other incumbent takes the same view of the threat).
Vignette 4. The board of directors of XYZ Corporation has reached the decision to terminate the employment of the firm’s CEO at a board meeting that did not include him. The date of the public announcement of his departure has been set. The firm is in a delicate financial and organizational state, with many key employees on the verge of resignation, and with key accounts at stake upon the departure of key employees, some of whom are loyal to the outgoing CEO. The board expects the mode, manner, and timing of the announcement to play a key role in determining the ongoing viability of XYZ. To quell rumors and speculations that could lead the outgoing CEO to break confidence and detrimentally communicate the impending changes, a controlled and well-timed release of the announcement is desired, with some key employees being brought into a “circle of trust” that can manage the firm on a daily basis in advance of a public announcement.
Because of the board’s fragmentary and sporadic access to the firm’s everyday communications, it is important for the leadership to be able to anticipate, in advance, the social paths along which the news will propagate in order to predict the response and to control possible side effects. It is well known to the directors that conveying the news “in confidence” to key employees may generate an informational cascade in which each recipient also conveys the news “in confidence” to a few trusted employees, to show connectedness, signal their importance, or simply out of a conspiratorial inclination. Thus, the question for the board is this: Who at XYZ can be trusted, and who trusts whom? The trust is not “generalized”: it is not merely the inclination to act cooperatively. Rather, what the board needs is a depiction of the social relationships within XYZ that will enable the news to be communicated precisely, reliably, and securely. Clearly, a trust relationship turns on what each party thinks the other thinks—or knows—at a moment in time, and on what the relationship is between what someone says and what is actually the case. The trust that employee A will place in any employee B will turn both on A’s confidence in B’s integrity (safeguarding the source and content of the information) and on A’s confidence in B’s competence (saying no more and no less than what is true and relevant at a particular point in time, without colorful distortions or purposeful omissions). Such confidence may need to be safeguarded by A’s expectation of B’s own trust in A (“only the trusting can be trusted”), which in turn generates new complications in the description of what a trusting relationship is.
“Information games” of the type we have constructed in these vignettes are entertaining because the moves they comprise produce changes—potentially very significant changes—in a larger pattern of interactive epistemic states that also overturn the emotional landscape of the situation. Baseline assumptions that social agents make about each other—cooperativeness, trustworthiness, rationality, docility—are uprooted by single and often involuntarily conveyed signals. Exformation—what the sender of a message does not say but is nevertheless relevant to the receiver precisely because it goes unsaid—can be as informative as information, provided that the right “epistemic tangle” is in place: if Alice thinks Bob knows that Alice is about to be fired and Bob says nothing to Alice when they have lunch together, the absence of a signal from Bob will influence Alice’s trust in him and her estimate of his support.
Ambiguous or complex epistemic states can be useful for either creating or breaking the trust required for information to flow freely; the local and temporary topology of an epinet can change the meaning of signals exchanged between the agents within it. Differences among the epistemic states of agents and among the relationships between them—which are what epinets are designed to capture—make a difference, and often a very large one, in the ensuing dynamics of the relationships or of the network.
Epistemic States and Epistemic Networks as Explanation Generators
The epistemic glue whose microstructure we aim to model via epistemic networks is intimately and ubiquitously involved in empirical studies of network phenomena. Networks of interorganizational collaboration (Powell, Koput, and Smith-Doerr 1996) rest on shared knowledge about tasks, technologies, and capabilities of other organizations in the same network, as do interpersonal networks of collaborative creative work (Uzzi and Spiro 2005). The cohesiveness and robustness of such networks hinge on regimes of trust among agents that often make possible “private games” (Burt and Knez 1995) that may be “too dangerous” (Burt 1999) in their absence. Co-mobilization in networks is sensitively dependent on what agents believe other agents will do if the former decide to mobilize (Chwe 1999, 2000, 2001), as well as on the fact that they know what other agents (validly) know they believe.
Nevertheless, their own social networks often confront agents with “horizons of observability” (Friedkin 1983) that constrain their knowledge of agents in the network to those corresponding to adjacent or alter-adjacent “nodes,” thus limiting the explanatory power of models of affiliative behavior predicated on an agent’s knowledge of the position of every other agent within the network (Gould 1993; Jackson 2005). The explanatory success of network theories of interpersonal and interorganizational phenomena thus depend on researchers’ assumptions about what agents know, what they know about what other agents know, and to what extent they trust what they and others know.
We are about to develop a precise way of representing states of knowledge, awareness, ignorance, and the like—jointly known as epistemic states—of agents in a social network. This representation will permit development of a new theory about the relevance and importance of epistemic states to the structure and dynamics of (interpersonal and interorganizational) networks (henceforth networks tout court), as well as development of more precise measurement instruments and techniques for testing and validating the theory. When we model a social structure as a social or economic network of agents, we pay a price in explanatory depth and generality when we omit an epistemic model of that structure. Consider the problem of predicting which among a large number of possible exchange networks will form among a set of agents interested in net-benefit-of-affiliation maximization (Jackson 2005; Ryall and Sorenson 2007). Existing analyses typically assume that the value associated with the formation of different ties is given and known to all agents, and they concern themselves with the calculation of the feasible or efficient networks that emerge as a result of different agents forming the ties that are most beneficial to them. However, if we assume that agents form ties on the basis of cost-benefit calculations of the relative values of possible ties that are in turn based on their knowledge of the value of each possible tie they could form, then we see that the set of possible ties must be conditioned by what agents know or believe about the value of forming any one tie in particular. In order for the most efficient of possible networks to form, the function that assigns values to each possible tie also has, itself, to be known to all network agents.
Two options are available to the realistically minded modeler who wants to examine the conditions under which this assumption is valid. One is to posit that, even though not all agents are informed about the value of all possible ties they can form, they can become informed over a finite period of time. In this case, the explanatory problem shifts to the mechanisms by which a full-knowledge state can be achieved. Different mechanisms for information dissemination are likely to create different knowledge regimes within the network. Broadcasting the value function, for instance, achieves full knowledge provided that (1) every agent is tuned in to the broadcaster and (2) every agent considers the presence of the broadcaster’s signal to be sufficient reason for believing the signal to be valid. Stipulation 1 raises questions about the marginal proclivity of agents to tune in to the broadcaster (e.g., it assumes that they believe the broadcaster can and has reason to supply information that is useful to them, and that they are aware that they are ignorant of, not oblivious to, the information). Stipulation 2 raises questions about the trustworthiness of the broadcasting source. Alternatively, the value function can be assumed to “percolate” through the network through word of mouth and rumor. The relative success of this explanation hinges on the marginal proclivity of agents (1) to inform each other truthfully and (2) to believe each other authentically; that is, success hinges on the trustworthiness and trustingness of network agents—qualities that are likely to be heterogeneous.
The second option is to assume that not all agents under examination know the value function. This creates a different explanatory agenda, one aimed at figuring out “who knows what” at any point in time. One possibility is to “survey the field,” but this has the disadvantage of (1) assuming that agents know what they believe or (2) inducing knowledge states that these agents would not otherwise have and that therefore cannot function as valid explanatory variables for what “would have happened had there been no intervention” (see Seidel and Westphal (2004) for evidence of such researcher-initiated information contagion). Since, moreover, we are interested in predicting what actions agents are likely to take by virtue of their knowledge, it is also important to understand whether or not different agents “believe or trust what they know,” which leads to the need to make finer distinctions in what we mean about “knowledge” in terms of the credibility or trustworthiness of the knowledge source.
What emerges from this discussion is the need to be precise about knowledge even when trying to explain a relatively simple network phenomenon such as purposeful tie formation. Being precise about knowledge requires more than just an effort to specify a function that maps bits of knowledge to individual network agents; the ways in which these agents know what they know are also relevant. Higher-level epistemic states such as awareness, oblivion, and ignorance are as important as first order states such as risky and uncertain belief. Moreover, interactive epistemic states such as trust, trustworthiness, trustingness, and credibility are crucially important to the kinds of plausible and testable stories that we can devise about network formation. By distinguishing epistemic differences likely to make a material difference, we aim to make the study of network epistemics an important component of social network modeling and inquiry.
The Way Forward
In this chapter, we have introduced the notion of epistemic glue as a prerequisite for the explanatory power of network models of social interaction and social action. We have shown that this glue has structure, comprising the epistemic states and ties among networked agents. We have illustrated how the evolution of this structure produces changes in the outcomes of interactions (“narratives”) and taken this to be a telling sign of the causal significance of such structures. We have defined epinets as symbolic representations of the glue underlying social networks and established that they matter and, more loosely, how they matter.
The investment to be made in appropriating a new language for modeling and representing phenomena is substantial. In spite of what we hope is a persuasive argument for considering epistemic states in detail and incorporating them systematically into the apparatus of social network analysis, we recognize that some surplus of insight and explanatory power is likely required to generate the necessary inducement to compensate for the required cognitive effort. To this end, we summarize here and elaborate in subsequent chapters the applications that epistemic imagery enables in key areas of social network theory and research. FIGURE 1.4 illustrates the organization of the book. We begin in Chapter 2 with an outline of the primitives of an epistemic description language (EDL) that can be used to describe epinets comprisingindividual agents, propositional beliefs, and epistemic states and ties. In Chapter 3, we deploy the EDL to show how interactive epistemic states (what agents think about what other agents think) can be used to understand fundamental network phenomena such as brokerage and closure via the epistemic analysis of basic mechanisms of network formation, co-mobilization, coordination, and communication. In Chapter 3, we show how self-referential epistemic states (beliefs about the structure of the network itself) can be used to analyze components of status, including knownness, fame, glory, and clout. In Chapter 4, we examine in detail how interactive epistemic states can be used to unpack and analyze trust-based relationships and network structures, such as “circles of trust” and superconductivity. We transition from epinet structure to dynamics in Chapter 5, introducing a series of moves and strategies that are defined as operations that agents perform on the structures of epinets of which they are part. We also introduce concepts that function in a manner similar to that of equilibrium in game-theoretic accounts, and we show how these concepts can be used to understand the long-run dynamics of epinets.
FIGURE 1.4 Organization of the Book