The second workshop panel showcased research focused on understanding how people’s cognitive and affective states are influenced by and influence social networks. According to panel moderator Noshir Contractor, Northwestern University, the presentations in this panel represented a new line of research in the area of social network thinking. This research, he said, which combines investigations on cognition, neural influences, and social psychology, presents an opportunity to move beyond identifying the attributes and traits of individuals to examine the physiological underpinnings of individuals’ behaviors as a result of social network interactions. He pointed out that research in this area addresses both how social networks influence the mind and how what is occurring inside the mind influences social networks. Emily Falk, University of Pennsylvania, a virtual moderator and presenter in this panel, added that researchers are interested in understanding how people use and perceive their networks, and are looking at individuals’ use of social networks from a number of perspectives, such as identity formation and social, cognitive, and affective features.
Carolyn Parkinson, University of California, Los Angeles, provided examples of studies that integrate approaches from social neuroscience and psychology with those from social network thinking to investigate information processing within individual minds. This integrated work is just beginning, she noted, but shows much promise. She pointed out that previous
work in social neuroscience has studied single relationships in isolation, citing the example of studies in which brain responses are compared as individuals view pictures of friends versus strangers. While limited, she said, these studies have provided insight into how individuals process social cues.
Parkinson called attention to the social brain hypothesis in social neuroscience. According to this hypothesis, many distinctive features of the human brain, its size and associated sophisticated cognitive abilities, have evolved specifically to enable humans to track and navigate large complexly bonded social groups. According to Parkinson, it makes sense intuitively that people are embedded in social networks and that the brain must store and retrieve a wide range of social knowledge to manage social interactions effectively on a daily basis. However, she said, little is known from science about the process of inferring and retrieving information about social networks and how such information impacts individuals’ thoughts and behaviors. She suggested that integrating approaches from social neuroscience and social network thinking would help elucidate various facets of social networks and their impacts on people’s behaviors. She then presented findings from studies that have yielded some preliminary understanding in this area.
One study described by Parkinson examined the social network of a cohort of MBA students. The social network was first characterized by mapping out all friendships among the 277 students. A subset of 21 of these students underwent functional magnetic resonance imaging (fMRI) scans to record brain responses while they watched videos of their classmates. The fMRI participants were later asked to estimate characteristics of these classmates’ positions in their shared social network: social distance (degrees of separation between classmates and participants), eigenvector centrality (how well connected a person is to well-connected others), and brokerage (the extent to which classmates bridge different areas of the network). Parkinson reported that the 21 participating students were found to know a great deal about where their classmates were positioned in the network. The researchers then used the fMRI data to investigate whether this network information was retrieved spontaneously in the brain when a participant encountered a familiar classmate. According to Parkinson, it was found that a distinct set of brain regions encoded each of the different network characteristics. Combining these results with knowledge about the functions of the different brain regions, she explained, can provide insight into how particular facets of social networks impact social responses.
Parkinson described another study of brain response and networks in which researchers observed individual differences in how brain regions tracked the popularity of other network members. The relationship between brain activity and the popularity of target group members appeared to be modulated by the perceiver’s own popularity, suggesting, she said, “that
some people’s brains might be more finely tuned to social structure than others.” She believes that a useful direction for future research would be to investigate the impact of context and individual differences on how the structure of the social world is processed in the brain.
Parkinson then drew attention to homophily, the notion that people tend to be surrounded by and frequently interact with others who are like themselves in a number of dimensions. Some research on this notion, she said, has focused on coarse variables, such as demographics, to define similarities among people. However, she noted, work in social neuroscience is looking at similarities from the perspective of interpretations of or responses to the world (e.g., similarities in emotions and the allocation of attention). One such study examined the brain responses of a group of people watching the same movie. Over time as the movie continued, Parkinson reported, their brain activity tended to synchronize. The extent to which brain activity aligned across individuals within particular brain regions, she continued, reflected similarities among them in interpretation, memory, and emotional reaction. Parkinson also cited a similar study that examined neural responses to video clips in different dyads among a cohort of graduate students whose social network had been characterized. In this study, the dyads were characterized by the degree of social distance: friends, friends of friends, friends of friends of friends, or dyads who were farther apart in terms of “degrees of separation” in their shared social network. Parkinson reported that this study found greater similarities in neural response among friends than among pairs of people who were farther apart in the social network. She suggested that future research address whether observed similarities in neural processing are a cause or consequence of friendship, and consider the questions of what kinds of similarities predict whether people become friends and the ways in which friends become more similar over time as they associate.
In closing, Parkinson asserted that neuroimaging is a useful research tool because it captures different kinds of processing in parallel and provides a level of information beyond study participants’ self-reports. She pointed out that self-reports from study participants may be of limited value because people may be unwilling to share information or unable to provide an accurate account of their thought processes or memories. She argued that combining the tools of neuroscience with methods for characterizing patterns of social relationships would advance research on understanding social cognition and behavior.
Falk began her presentation by saying she would make the argument that studying brain function within social networks would lead to ways of
predicting individual-, group-, and population-level behaviors. To consider how this might work, she provided examples of studies of brain activity in relation to trends in ideas and behaviors. She emphasized that to make progress toward predicting behaviors, it will be necessary for research to consider both the individual and the web of the individual’s relationships. She asserted that future research can learn more about the dynamics within social networks by examining individuals (and brains) within the networks and can learn more about brain function by integrating broader contextual information about social network structure and composition with neuroimaging results.
Recent research, Falk noted, has addressed questions about brain activity in individuals receptive to change. Researchers, she said, have found that behavior change is the result of “finding personal value in a new set of ideas or behaviors.” She pointed out that a core system in the brain involving the ventromedial prefrontal cortex engages when people make decisions and is part of the brain’s “value system.” She cited a series of studies that have looked at brain activity, self-reports, and pre/post behaviors of people exposed to persuasive messages about health behaviors ranging from smoking to physical activity to peer influence on risk taking. Researchers, she explained, surveyed participants’ attitudes toward such a behavior, their intentions with respect to changing the behavior, and a number of other constructs (e.g., relevant beliefs). In study after study, she reported, brain activity in the value system predicted individual behavior change, explaining variance above and beyond people’s self-reports of their intentions and other constructs.
Falk also noted that at the moment people are presented with new information or a new idea, they tend to be vulnerable to discounting it. Initial messages that are confrontational, she elaborated, often elicit defensiveness, whereas specific psychological priming techniques (e.g., value affirmation) reduce the threat of the proposed behavior change and allow a person to perceive greater value in the persuasive messages. She added that studies of brain activity have shown that such use of value affirmation and other priming techniques does increase activity in the brain value system and predict subsequent behavior change.
Falk also pointed out that brain activity has been investigated in small groups of people to examine group behavior in a number of domains, such as information sharing. These studies have observed synchronization in brain activity across members of a group engaged in similar activities (see, for example, findings from group studies previously cited in the summary of Parkinson’s presentation). Falk suggested that the value system in the brain of the initial message recipient may be important not only “because it affects individual behavior but it also affects how well a message spreads through a network.” She noted that brain activity in the value system of a
communicator in response to health news increases the likelihood of sharing, and that preferences spread from communicators’ to receivers’ brains. Future research, she suggested, could measure brain response to assess the cognition involved in calculating the value of new information. She proposed two questions that may be involved in an individual’s decision making: (1) Is this new information relevant to me?, and (2) If other people receive this information, how will sharing it affect my relationships or social status? These types of questions have been studied empirically, she said, and preliminary evidence shows that brain activity significantly predicts the action of sharing information among one’s networks.
Falk also noted that social network structure and composition can moderate brain–behavior relationships. To illustrate this point, she provided an example of the effect the type of network surrounding an individual has on decision making. She cited one study that entailed monitoring the behavior and brain activity of about 200 sedentary adults in response to messaging about increasing their physical activity. She reported that people whose friend networks comprised more physically active individuals showed more brain activity in the value system, with corresponding change in behavior involving increased physical activity.
Falk presented another example in which the behavior change among people with different types of social networks was the same, but the underlying brain mechanisms turned out to be different. In this study, she said, participants learned about a mobile phone application and what their peers thought about it, and had to decide whether to recommend it to friends. Participants were classified according to their type of social network—brokerage versus high-closure. She defined brokerage networks as those in which a broker connects people who are otherwise not friends with one another, and hence may be the source of the translation and transmittal of information among network members. A high-closure network, on the other hand, is one in which information flows are confined and shared among network members. According to Falk, the study found no difference in how peer information was ultimately used among participants. However, she said, the study did find differences in brain activity for people with different network structures, suggesting that network association played a role in how brain mechanisms were employed: people with higher levels of information brokerage showed more activation in regions of the brain that have been linked to understanding other people’s mental states when deciding what to recommend to peers. She noted that other studies have also demonstrated predictable patterns of brain activity during group interactions that can distinguish participants according to their type of social network.
In closing, Falk argued that future research should continue to examine brain dynamics in relation to social network dynamics and that new
multilayer network models are needed to combine these levels of analysis. She proposed some research questions to consider: “Why do ideas spread in some contexts and not others? Who is likely to be most influential in different kinds of social contexts? How can motivation, learning, and performance [be optimized]? How do people learn the structure of their social world? How can . . . optimal interventions [be constructed given] these factors to promote well-being?
Jesse Hoey, University of Waterloo, described his work on developing artificial intelligence (AI) that is capable of operating with groups of human beings on social and emotional levels. Some of the motivation for this work, he said, has come from observing the challenges entailed in collaborations in online networks. He noted that a group of people interacting through and with technology is referred to as a “sociotechnical system.”
According to Hoey, the number of people creating online networks for social purposes and collaboration has increased exponentially. He cited the examples of an online community of engineers developing a do-it-yourself autonomous car system and recruiting a network of drivers to help reduce congestion and traffic-related pollution; another group creating apps to aid refugees with migrant issues in Europe; and computer programmers working together virtually to develop new software.1 He noted that the comments and interactions among people in these online networks can be cordial, but they can also be negative and mean-spirited. To illustrate the challenges of online collaboration, he pointed to the findings of the Open Source Survey, which showed that among the top six reported problems, four could be linked to social and emotional challenges within the network. These challenges included unresponsiveness, dismissive responses, conflict, and unwelcoming language.2
A goal of Hoey’s research is to build computational models of group behavior in such online teams, or sociotechnical systems. These models, he said, would be used to develop AI agents that could become members of the team, with social, emotional, and cultural grounding “to help change the way that people are behaving on these networks and hopefully make the networks more engaging, more inclusive, and more effective.”
Hoey commented on the state of AI. AI, he said, is advanced in some areas, such as computers that can play difficult games and answer complex
questions and autonomous systems for well-defined tasks. However, he asserted, AI applied to group decision making is much further behind. He illustrated the state of AI with a recent grand challenge to create a team of robots able to play soccer. He pointed out that while these robots can coordinate with each other at a very superficial level, they are nowhere close to what humans can do and lack such features as empathy and altruism that aid decision making in the game. He argued that the robots lacked a model of emotion.
To develop a model of emotion, Hoey has built on pioneering work by Charles Osgood on semantic differential scales.3 A semantic differential scale, he explained, is a scale with opposing adjectives at each end. To illustrate, he used the concept “polite” with a scale ranging from “rough” to “smooth.” Even though polite has little to do with roughness and smoothness, Hoey reported that people sharing the same language and culture will rate polite toward smooth at the same point on the scale with “a remarkable degree of consistency and consensus.” This agreement exists, he said, because the terms “feel” the same to people within a culture, and as such, the ratings provide a basis for affective labeling of terms. He added that Osgood’s work identified three dimensions that could explain a large percentage of the variance in the data across a number of these semantic scales: (1) evaluation (good to bad), (2) potency (strong to weak), and (3) activity (hyper to sleep).
Hoey suggested that certain fundamental sentiments are implicitly agreed upon. He asserted that when human beings see or enter a social situation, they create “transient impressions” of the people involved, their behaviors, the setting, and so on. These transient impressions can differ from fundamental sentiments, he suggested; however, the tendency is for humans to align their behaviors or emotions so that the situation fits with their fundamental sentiments or such that the situation is emotionally consistent.
Finally, Hoey introduced affect control theory as a mathematical consistency theory that has been operationalized as a computer program.4 The premise behind affect control theory is that actors in a group behave in ways that cause them to experience transient impressions that are consistent with their fundamental sentiments. The transient impressions and fundamental sentiments are represented mathematically based on the scaling of the three dimensions identified by Osgood. Hoey reported that this theory and the associated computer software are being used in a number of stud-
3 Osgood, C.E. (1952). The nature and measurement of meaning. Psychological Bulletin, 49(3), 197–237.
4 For more information on affect control theory, see Heise, D. (2007). Expressive Order: Confirming Sentiments in Social Actions. New York: Springer.
ies to simulate interactions among a group of people and help predict, for example, who might emerge as the leader or who might be excluded. He suggested that as more is learned from this modeling of group interactions, AI can advance to develop artificial agents that may be capable of becoming functional members of human groups.
Kenneth Joseph, Northeastern University, began his presentation by emphasizing the importance of social identities in studies of social networks. He highlighted two points: (1) both social identities and social networks have strong and weak versions that combine to affect social behaviors, and (2) both social identities and social networks evolve, and they often evolve together. He defined social identities as “the words and phrases used to label [one’s self] and others.” His presentation focused on one’s own labels of one’s self and their effects on behavior. Weak social identities, he said, are situational: they may change as an individual moves from place to place—for example, from home to work. He pointed out that the dichotomy between weak and strong social identities is useful for discussion purposes, but in reality, people’s identities fall along a continuum of strength. He added that social identities evolve as individuals learn from their behaviors, as well as those of others. He suggested that data from social media can be a valuable resource for studying identities, and their dynamic nature, within social networks.
Joseph presented an example drawn from recent events to illustrate how strong and weak identities could have played a role in an observed behavior and change in a social network. The example centered on a longtime Buffalo Bills fan who quit his job at the Bills stadium suddenly after members of the team protested the national anthem. Joseph suggested that the fan’s connection to the Buffalo Bills fan network, although enduring, was probably a weak part of his social identity. A stronger part of the fan’s identity, he said, may have been his identity as a conservative American. In this case, he concluded, a strong identity encouraged a particular behavior that a weak identity, and the weak social ties connected to it, could not prevent.
According to Joseph, this example illustrates the connection between social identity and networks: eliminating part of one’s identity resulted in eliminating ties to a particular network. Joseph also emphasized the importance of thinking strategically about the networks being analyzed. He suggested that since people are part of multiple networks, research on the impact of networks on behaviors should consider the strength and stability of the different networks and which might most influence behavior in a given context.
Joseph closed by stating that much remains to be learned about the role of weak and strong identities in behavior. He views social media as an opportunity to observe behavior at scale dynamically over time. For example, with Twitter, people reveal their identities, and they interact in a dynamic social network. Joseph suggested that research look at how particular identities are built and accepted, how identities are used in different networks, how identities are dropped, and how network structures play a role in identity formation.
This page intentionally left blank.