Humans have a remarkable capacity for coordination. Our ability to interact and act jointly in groups is crucial to our success as a species. Joint Action (JA) research has often concerned itself with simplistic behaviors in highly constrained laboratory tasks. But there has been a growing interest in understanding complex coordination in more open-ended contexts. In this regard, collective music improvisation has emerged as a fascinating model domain for studying basic JA mechanisms in an unconstrained and highly sophisticated setting. A number of empirical studies have begun to elucidate coordination mechanisms underlying joint musical improvisation, but these findings have yet to be cached out in a working computational model. The present work fills this gap by presenting Tonal Emergence, an idealized agent-based model of improvised musical coordination. Tonal Emergence models the coordination of notes played by improvisers to generate harmony (i.e., tonality), by simulating agents that stochastically generate notes biased towards maximizing harmonic consonance given their partner’s previous notes. The model replicates an interesting empirical result from a previous study of professional jazz pianists: feedback loops of mutual adaptation between interacting agents support the production of consonant harmony. The model is further explored to show how complex tonal dynamics, such as the production and dissolution of stable tonal centers, are supported by agents that are characterized by (i) a tendency to strive toward consonance, (ii) stochasticity, and (iii) a limited memory for previously played notes. Tonal Emergence thus provides a grounded computational model to simulate and probe the coordination mechanisms underpinning one of the more remarkable feats of human cognition: collective music improvisation.
How do people use information from others to solve complex problems? Prior work has addressed this question by placing people in social learning situations where the problems they were asked to solve required varying degrees of exploration. This past work uncovered important interactions between groups’ connectivity and the problem’s complexity: the advantage of less connected networks over more connected networks increased as exploration was increasingly required for optimally solving the problem at hand. We propose the Social Interpolation Model (SIM), an agent-based model to explore the cognitive mechanisms that can underlie exploratory behavior in groups. Through results from simulation experiments, we conclude that “exploration” may not be a single cognitive property, but rather the emergent result of three distinct behavioral and cognitive mechanisms, namely, (a) breadth of generalization, (b) quality of prior expectation, and (c) relative valuation of self-obtained information. We formalize these mechanisms in the SIM, and explore their effects on group dynamics and success at solving different kinds of problems. Our main finding is that broad generalization and high quality of prior expectation facilitate successful search in environments where exploration is important, and hinder successful search in environments where exploitation alone is sufficient.
Often members of a group benefit from dividing the group’s task into separate components, where each member specializes their role so as to accomplish only one of the components. While this division of labor phenomenon has been observed with respect to both manual and cognitive labor, there is no clear understanding of the cognitive mechanisms allowing for its emergence, especially when there are multiple divisions possible and communication is limited. Indeed, maximization of expected utility often does not differentiate between alternative ways in which individuals could divide labor. We developed an iterative two-person game in which there are multiple ways of dividing labor, but in which it is not possible to explicitly negotiate a division. We implemented the game both as a human experimental task and as a computational model. Our results show that the majority of human dyads can finish the game with an efficient division of labor. Moreover, we fitted our computational model to the behavioral data, which allowed us to explain how the perceived similarity between a player’s actions and the task’s focal points guided the players’ choices from one round to the other, thus bridging the group dynamics and its underlying cognitive process. Potential applications of this model outside cognitive science include the improvement of cooperation in human groups, multi-agent systems, as well as human-robot collaboration.
Dubova, M., Moskvichev, A., & Goldstone, R. L. (2020). Reinforcement Communication Learning in Different Social Network Structures. International Conference on Machine Learning, 1st Language and Reinforcement Learning Workshop.
Social network structure is one of the key determinants of human language evolution. Previous work has shown that the network of social interactions shapes decentralized learning in human groups, leading to the emergence of different kinds of communicative conventions. We examined the effects of social network organization on the properties of communication systems emerging in decentralized, multi-agent reinforcement learning communities. We found that the global connectivity of a social network drives the convergence of populations on shared and symmetric communication systems, preventing the agents from forming many local “dialects”. Moreover, the agent’s degree is inversely related to the consistency of its use of communicative conventions. These results show the importance of the basic properties of social network structure on reinforcement communication learning and suggest a new interpretation of findings on human convergence on word conventions.
Strait, M., Lier, F., Bernotat, J., Wachsmuth, S., Eyssel, F., Goldstone, R. L., & Sabanovic, S. (2020). A three-site reproduction of the Joint Simon Effect with the NAO robot. 15th Annual International Conference on Human Robot Interaction. Cambridge, England. Association for Computing Machinery (ACM) & Institute of Electrical and Electronics Engineers (IEEE). (pp. 103-111). New York: ACM.
The generalizability of empirical research depends on the reproduction of findings across settings and populations. Consequently, generalizations demand resources beyond that which is typically available to any one laboratory. With collective interest in the joint Simon effect (JSE) – a phenomenon that suggests people work more effectively with humanlike (as opposed to mechanomorphic) robots – we pursued a multi-institutional research cooperation between robotics researchers, social scientists, and software engineers. To evaluate the robustness of the JSE in dyadic human-robot interactions, we constructed an experimental infrastructure for exact, lab-independent reproduction of robot behavior. Deployment of our infrastructure across three institutions with distinct research orientations (well-resourced versus resource-constrained) provides initial demonstration of the success of our approach and the degree to which it can alleviate technical barriers to HRI reproducibility. Moreover, with the three deployments situated in culturally distinct contexts (Germany, the U.S. Midwest, and the Mexico-U.S. Border), observation of a JSE at each site provides evidence its generalizability across settings and populations.
Setzler, M., & Goldstone, R. L. (2020). Coordination and consonance between interacting, improvising musicians. Open Mind: Discoveries in Cognitive Science, 4, 88—101. https://doi.org/10.1162/opmi_a_00036.
Joint action (JA) is ubiquitous in our cognitive lives. From basketball teams to teams of surgeons, humans often coordinate with one another to achieve some common goal. Idealized laboratory studies of group behavior have begun to elucidate basic JA mechanisms, but little is understood about how these mechanisms scale up in more sophisticated and open-ended JA that occurs in the wild. We address this gap by examining coordination in a paragon domain for creative joint expression: improvising jazz musicians. Coordination in jazz music subserves an aesthetic goal: the generation of a collective musical expression comprising coherent, highly nuanced musical structure (e.g. rhythm, harmony). In our study, dyads of professional jazz pianists improvised in a “coupled”, mutually adaptive condition, and an “overdubbed” condition which precluded mutual adaptation, as occurs in common studio recording practices. Using a model of musical tonality, we quantify the flow of rhythmic and harmonic information between musicians as a function of interaction condition. Our analyses show that mutually adapting dyads achieve greater temporal alignment and produce more consonant harmonies. These musical signatures of coordination were preferred by independent improvisers and naive listeners, who gave higher quality ratings to coupled interactions despite being blind to condition. We present these results and discuss their implications for music technology and JA research more generally.
Setzler, M., & Goldstone, R. L. (2020). Quantifying Emergent, Dynamic Tonal Coordination in Collaborative Musical Improvisation. Proceedings of the 42nd Annual Conference of the Cognitive Science Society. (pp. 461-466). Toronto, CA. Cognitive Science Society.
Groups of interacting individuals often coordinate in service of abstract goals, such as the alignment of mental representations in conversation, or the generation of new ideas in group brainstorming sessions. What are the mechanisms and dynamics of abstract coordination? This study examines coordination in a sophisticated paragon domain: collaboratively improvising jazz musicians. Remarkably, freely improvising jazz ensembles collectively produce coherent tonal structure (i.e. melody and harmony) in real time performance without previously established harmonic forms. We investigate how tonal structure emerges out of interacting musicians, and how this structure is constrained by underlying patterns of coordination. Dyads of professional jazz pianists were recorded improvising in two conditions of interaction: a ‘coupled’ condition in which they could mutually adapt to one another, and an ‘overdubbed’ condition which precluded mutual adaptation. Using a computational model of musical tonality, we show that this manipulation effected the directed flow of tonal information amongst pianists, who could mutually adapt to one another’s notes in coupled trials, but not in overdubbed trials. Consequently, musicians were better able to harmonize with one another in coupled trials, and this ability increased throughout the course of improvised performance. We present these results and discuss their implications for music technology and joint action research more generally.
Campbell, C., Izquierdo, E. & Goldstone, R. L. (2020). How much to copy from others? The role of partial copying in social learning. Proceedings of the 42nd Annual Conference of the Cognitive Science Society. (pp. 916-922). Toronto, Canada: Cognitive Science Society.
One of the major ways that people engage in adaptive problem solving is by copying the solutions of others. Most of the work on this field has focused on three questions: when to copy, who to copy from, and what to copy. However, how much to copy has been relatively less explored. In the current research, we are interested in the consequences for a group when its members engage in social learning strategies with different tendencies to copy entire or partial solutions and different complexities of search problems. We also consider different network topologies that affect the solutions visible to each member. Using a computational model of collective problem solving, we demonstrate that strategies where social learning involves partial copying outperform strategies where individuals copy entire solutions. We analyze the exploration/exploitation dynamics of these social learning strategies under the different conditions.
In peer instruction, instructors pose a challenging question to students, students answer the question individually, students work with a partner in the class to discuss their answers, and finally students answer the question again. A large body of evidence shows that peer instruction benefits student learning. To determine the mechanism for these benefits, we collected semester-long data from six classes, involving a total of 208 undergraduate students being asked a total of 86 different questions related to their course content. For each question, students chose their answer individually, reported their confidence, discussed their answers with their partner, and then indicated their possibly revised answer and confidence again. Overall, students were more accurate and confident after discussion than before. Initially correct students were more likely to keep their answers than initially incorrect students, and this tendency was partially but not completely attributable to differences in confidence. We discuss the benefits of peer instruction in terms of differences in the coherence of explanations, social learning, and the contextual factors that influence confidence and accuracy.
Tump, A. N., Wu, C. M., Bouhlel, I., & Goldstone, R. L. (2019).The Evolutionary Dynamics of Cooperation in Collective Search. Proceedings of the 41st Annual Conference of the Cognitive Science Society. (pp. 883-889). Montreal, Canada: Cognitive Science Society.
How does cooperation arise in an evolutionary context? We approach this problem using a collective search paradigm where interactions are dynamic and there is competition for rewards. Using evolutionary simulations, we find that the unconditional sharing of information can be an evolutionary advantageous strategy without the need for conditional strategies or explicit reciprocation. Shared information acts as a recruitment signal and facilitates the formation of a self-organized group. Thus, the improved search efficiency of the collective bestows byproduct benefits onto the original sharer. A key mechanism is a visibility radius, where individuals have unconditional access to information about neighbors within a limited distance. Our results show that for a variety of initial conditions—including populations initially devoid of prosocial individuals—and across both static and dynamic fitness landscapes, we find strong selection pressure to evolve unconditional sharing.
Setzler, M., & Goldstone, R. L. (2019). Patterns of coordination in simultaneously and sequentially improvising jazz musicians. Proceedings of the 41st Annual Conference of the Cognitive Science Society. (pp. 1035-1040). Montreal, Canada: Cognitive Science Society.
In Joint Action (JA) tasks, individuals must coordinate their actions so as to achieve some desirable outcome at the grouplevel. Group function is an emergent outcome of ongoing, mutually constraining interactions between agents. Here we investigate JA in dyads of improvising jazz pianists. Participants’ musical output is recorded in one of two conditions: a real condition, in which two pianists improvise together as they typically would, and a virtual condition, in which a single pianist improvises along with a “ghost partner” – a recording of another pianist taken from a previous real trial. The conditions are identical except for that in real trials subjects are mutually coupled to one another, whereas there is only unidirectional influence in virtual trials (i.e. recording to musician). We quantify ways in which the rhythmic structures spontaneously produced in these improvisations is shaped by mutual coupling of co-performers. Musical signatures of underlying coordination patterns are also shown to parallel the subjective experience of improvisers, who preferred playing in trials with bidirectional influence despite not explicitly knowing which condition they had played in. These results illuminate how mutual coupling shapes emergent, group-level structure in the creative, open-ended and fundamentally collaborative domain of expert musical improvisation.
Sloman, S. J., Goldstone, R. L., & Gonzalez, C. (2019). Complex exploration dynamics from simple heuristics in a collective learning environment. Proceedings of the 41st Annual Conference of the Cognitive Science Society. (pp. 2818-2824). Montreal, Canada: Cognitive Science Society.
Effective problem solving requires both exploration and exploitation. We analyze data from a group problem-solving task to gain insight into how people use information from past experiences and from others to achieve explore-exploit trade-offs in complex environments. The behavior we observe is consistent with the use of simple, reinforcement-based heuristics. Participants increase exploration immediately after experiencing a low payoff, and decrease exploration immediately after experiencing a high or improved payoff. We suggest that whether an outcome is perceived as “high” or “low” is a dynamic function of the outcome information available to participants. The degree to which the distribution of observed information reflects the true range of possible outcomes plays an important role in determining whether or not this heuristic is adaptive in a given environment.
Andrade-Lotero, E., & Goldstone, R. L. (2019). Self-Organized Division of Cognitive Labor. Proceedings of the 41st Annual Conference of the Cognitive Science Society. (pp. 91-97). Montreal, Canada: Cognitive Science Society.
The division of labor phenomenon has been observed with respect to both manual and cognitive labor, but there is no clear understanding of the intra- and inter-individual mechanisms that allow for its emergence, especially when there are multiple divisions possible and communication is limited. Situations fitting this description include individuals in a group splitting a geographical region for resource harvesting without explicit negotiation, or a couple tacitly negotiating the hour of the day for each to shower so that there is sufficient hot water. We studied this phenomenon by means of an iterative two-person game where multiple divisions are possible, but no explicit communication is allowed. Our results suggest that there are a limited number of biases toward divisions of labor, which serve as attractors in the dynamics of dyadic coordination. However, unlike Schelling’s focal points, these biases do not attract players’ attention at the onset of the interaction, but are only revealed and consolidated by the in-game dynamics of dyadic interaction.
The utility of our actions frequently depends upon the beliefs and behavior of other agents. Thankfully, through experience, we learn norms and conventions that provide stable expectations for navigating our social world. Here, we review several distinct influences on their content and distribution. At the level of individuals locally interacting in dyads, success depends on rapidly adapting pre-existing norms to the local context. Hence, norms are shaped by complex cognitive processes involved in learning and social reasoning. At the population level, norms are influenced by intergenerational transmission and the structure of the social network. As human social connectivity continues to increase, understanding and predicting how these levels and time scales interact to produce new norms will be crucial for improving communities.
Low-level “adaptive” and higher-level “sophisticated” human reasoning processes have been proposed to play opposing roles in the emergence of unpredictable collective behaviors such as crowd panics, traffic jams, and market bubbles. While adaptive processes are widely recognized drivers of emergent social complexity, complementary theories of sophistication predict that incentives, education, and other inducements to rationality will suppress it. We show in a series of multiplayer laboratory experiments that, rather than suppressing complex social dynamics, sophisticated reasoning processes can drive them. Our experiments elicit an endogenous collective behavior and show that it is driven by the human ability to recursively anticipate the reasoning of others. We identify this behavior, “sophisticated flocking”, across three games, the Beauty Contest and the “Mod Game” and “Runway Game”. In supporting our argument, we also present evidence for mental models and social norms constraining how players express their higher-level reasoning abilities. By implicating sophisticated recursive reasoning in the kind of complex dynamic that it has been predicted to suppress, we support interdisciplinary perspectives that emergent complexity is typical of even the most intelligent populations and carefully designed social systems.
Bouhlel, I., Wu, C. M., Hanaki, N., & Goldstone, R. L. (2018). Sharing is not erring: Pseudo-reciprocity in collective search. Proceedings of the 40th Annual Conference of the Cognitive Science Society. (pp. 156-161). Madison, Wisconsin: Cognitive Science Society.
Information sharing in competitive environments may seem counterintuitive, yet it is widely observed in humans and other animals. For instance, the open-source software movement has led to new and valuable technologies being released publicly to facilitate broader collaboration and further innovation. What drives this behavior and under which conditions can it be beneficial for an individual? Using simulations in both static and dynamic environments, we show that sharing information can lead to individual benefits through the mechanisms of pseudoreciprocity, whereby shared information leads to by-product benefits for an individual without the need for explicit reciprocation. Crucially, imitation with a certain level of innovation is required to avoid a tragedy of the commons, while the mechanism of a local visibility radius allows for the coordination of self-organizing collectives of agents. When these two mechanisms are present, we find robust evidence for the benefits of sharing—even when others do not reciprocate.
Izquierdo, E. J., Campbell, C. M., & Goldstone, R. L. (2018). The Great Melting Pot: generating diversity by combining solutions across a global population. Annual Meeting of Collective Intelligence. Zurich, Switzerland.
One of the major ways that people engage in adaptive problem solving is by copying or imitating the solutions of others. Imitation saves an individual time and mitigates potential risks from individual trial-and-error learning. When an individual finds a neighbor with a better solution than theirs, copying their entire solution guarantees an improvement over the individual’s current condition. However, this reduces the diversity of solutions in the group and can lead the group to getting stuck in a local optima. One alternative is to copy the neighbor’s solution only partially, although this comes at a risk for the individual. Mixing two solutions may or may not lead to an improvement over their previous solution, but mixing has the potential to allow the group to explore entirely new areas of solution space. So, although partial copying comes at a cost to the individual, under what conditions does it benefit the group? In the current research, we are interested in the consequences for the group when its members engage in social learning strategies with different tendencies to copy entire or partial solutions, with different network topologies that affect the neighbors’ solutions visible to each member, and with different complexities of search tasks.
Learners often struggle to grasp the important, central principles of complex systems, which describe how interactions between individual agents can produce complex, aggre-gate-level patterns. Learners have even more difficulty transferring their understanding of these principles across superficially dissimilar instantiations of the principles. Here, we provide evidence that teaching high school students an agent-based modeling language can enable students to apply complex system principles across superficially different domains. We measured student performance on a complex systems assessment before and after 1 week training in how to program models using NetLogo (Wilensky, 1999a). Instruction in NetLogo helped two classes of high school students apply complex sys-tems principles to a broad array of phenomena not previously encountered. We argue that teaching an agent-based computational modeling language effectively combines the benefits of explicitly defining the abstract principles underlying agent-level interac-tions with the advantages of concretely grounding knowledge through interactions with agent-based models.
We lay out a multiple, interacting levels of cognitive systems (MILCS) framework to account for the cognitive capacities of individuals and the groups to which they belong. The goal of MILCS is to explain the kinds of cognitive processes typically studied by cognitive scientists, such as perception, attention, memory, categorization, decision-making, problem solving, judgment, and flexible behavior. Two such systems are considered in some detail—lateral inhibition within a network for selecting the most attractive option from a candidate set and a diffusion process for accumulating evidence to reach a rapid and accurate decision. These system descriptions are aptly applied at multiple levels, including within and across people. These systems provide accounts that unify cognitive processes across multiple levels, can be expressed in a common vocabulary provided by network science, are inductively powerful yet appropriately constrained, and are applicable to a large number of superficially diverse cognitive systems. Given group identification processes, cognitively resourceful people will frequently form groups that effectively employ cognitive systems at higher levels than the individual. The impressive cognitive capacities of individual people do not eliminate the need to talk about group cognition. Instead, smart people can provide the interacting parts for smart groups
An individual can interact with the same set of people over many different scales simultaneously. Four people might interact as a group of four and, at the same time, in pairs and triads. What is the relationship between different parallel interaction scales, and how might those scales themselves interact? We devised a four-player experimental game, the Modular Stag Hunt, in which participants chose not just whether to coordinate, but with whom, and at what scale. Our results reveal coordination behavior with such a strong preference for dyads that undermining pairwise coordination actually improves group-scale outcomes. We present these findings as experimental evidence for competition, as opposed to complementarity, between different possible scales of multi-player coordination. This result undermines a basic premise of approaches, like those of network science, that fail to model the interacting effects of dyadic, triadic, and group-scale structure on group outcomes.
To investigate the effect of competitive incentives under peer review, we designed a novel experimental setup called the Art Exhibition Game. We present experimental evidence of how competition introduces both positive and negative effects when creative artifacts are evaluated and selected by peer review. Competition proved to be a double-edged sword: on the one hand, it fosters innovation and product diversity, but on the other hand, it also leads to more unfair reviews and to a lower level of agreement between reviewers. Moreover, an external validation of the quality of peer reviews during the laboratory experiment, based on 23,627 online evaluations on Amazon Mechanical Turk, shows that competition does not significantly increase the level of creativity. Furthermore, the higher rejection rate under competitive conditions does not improve the average quality of published contributions, because more high-quality work is also rejected. Overall, our results could explain why many ground-breaking studies in science end up in lower-tier journals. Differences and similarities between the Art Exhibition Game and scholarly peer review are discussed and the implications for the design of new incentive systems for scientists are explained.
Without ever explicitly discussing it, groups often times establish norms. A family or committee might develop a norm about when it is acceptable or not for members to interrupt each other. People greeting each other in different countries have very different norms for whether to shake hands or kiss, and if to kiss, how many times and in what cheek order. In some countries, tipping is not the norm, but if it is, violating the tipping norm could make you a persona non grata at a restaurant. We (Hawkins & Goldstone, 2016) were interested in how social norms emerge in a group without its members explicitly deciding on them, and the factors that promote effective norms.
To help explore these questions, we started by considering a simple scenario we call “Battle of the Exes.” You and your romantic partner live in a small town and both love coffee. Your shared loved of coffee was not, alas, enough to keep you together, and you have now broken up. There are only two coffee shops in your town, one with much better coffee than the other. Both you and your ex want to go every day for coffee during your simultaneously occurring coffee breaks, but if you pick the same place and run into one another, neither of you will enjoy your break at all.
Neither you nor your ex want to sit down to negotiate a schedule, but can you nonetheless develop a satisfactory routine? One of you could always go to the better coffee shop, but that would not be fair. Each of you could choose randomly, but that would end up with you and your ex often seeing each other, which would not maximize your duo’s happiness, and would not provide a stable solution in the long run.
These three features — fairness, happiness maximization, and stability are generally useful ways to assess the quality of a group’s behavior. To study scenarios like “Battle of the Exes” in the laboratory, we developed an interactive, real-time, online game. On each of the 60 rounds of the game, two players are given the choice of moving their avatar to one of two circles — one that they can visibly see will give them a small monetary prize and one that will give them a large payoff. The only catch is that if both players move to the same circle, then neither player gets anything for that round. For half of the groups, there was a small discrepancy between the prizes (1 cent vs 2 cents), and for the other half, there was a large discrepancy (1 cent versus 4 cents). Also, for half of the groups, each of the players could see the other player’s moment-to-moment position as they moved to the circles (Dynamic movement), while for the other half of the groups, the players only see the final choice that the other player made (Ballistic movement).
568 players were matched together to create 284 two-player groups. Some groups developed behaviors that were fair and stable, and led to both players earning a lot of money. These groups tended to develop social norms even without explicit communication. For example, the players A and B would alternate over rounds who got the large payoff, first A then B then A…., leading to a pattern like ABABABABAB.
In terms of maximizing happiness, the dynamic condition led to better earnings for the players than the ballistic condition. When the players can see each others’ moment-to-moment inclinations, that helps them coordinate. The dynamic condition also led to fairer solutions than the ballistic condition, with players earning similar amounts of money. An implication of these results is that giving the members in a group more information about what each person in the group is currently thinking about doing can help the group achieve well-coordinated, fair and happy solutions. This is something for politicians, social network providers, and amusement parks to consider when they are trying to design social spaces for their groups. Mutual visibility of group members is often an effective way to promote coordination.
In terms of developing stable strategies, there was a striking interaction between payoffs and movement type. When there was not a large difference in payoffs, choices in the ballistic condition were more stable than in the dynamic condition. When the stakes were low, players in the dynamic condition simply relied on moment-to-moment visual information to figure out who should get the larger payoff on any given round. They did not feel a strong pressure to develop a norm because they could use their continuous information as a crutch to help them coordinate. However, when the stakes were high, with one circle earning four times what the other circle earned, then the dynamic condition developed significantly more stable solutions than the ballistic condition. For these particularly contentious, high stakes situations, it is useful for the players to develop strong norms to help them coordinate, and the moment-to-moment information about player positions helps to create these norms.
One clear measure of how much contention there is in a group is how long both players move toward the same high payoff option before one “peels off” and lets the other player have the high payoff prize. Using this objective measure, groups have more contention at the beginning of the experiment session than the end. The higher stakes condition has more contention early on than the lower stakes condition, but by the end of the experiment, that ordering is flipped. Groups that have more contention at the beginning of the experiment tend to have less contention by the end of experiment, and are more likely to develop clever strategies like alternating who gets the high payoff option from round to round. A take-home message from this result is that contention in groups is not something to be avoided. For the groups in our “Battle of the Exes” game, early contention gives rise to well-coordinated, fair, efficient, and happiness maximizing solutions by the end of the experiment. It may be tempting to try to pave over contention and disagreement in a group, but letting the group work through these contentions is often key to giving them the motivation and insight that they need to develop creative, well-coordinated norms like alternating who gets the better payoff over rounds. So, although it may have been contention that broke you and your ex up in the first place, there is hope that this kind of early contention may allow you to enjoy your superior cup of coffee in peace. At least on Mondays, Wednesdays, and Fridays.
Why are some behaviors governed by strong social conventions while others are not? We experimentally investigate two factors contributing to the formation of conventions in a game of impure coordination: the continuity of interaction within each round of play (simultaneous vs. real-time) and the stakes of the interaction (high vs. low differences between payoffs). To maximize efficiency and fairness in this game, players must coordinate on one of two equally advantageous equilibria. In agreement with other studies manipulating continuity of interaction, we find that players who were allowed to interact continuously within rounds achieved outcomes with greater efficiency and fairness than players who were forced to make simultaneous decisions. However, the stability of equilibria in the real-time condition varied systematically and dramatically with stakes: players converged on more stable patterns of behavior when stakes are high. To account for this result, we present a novel analysis of the dynamics of continuous interaction and signaling within rounds. We discuss this previously unconsidered interaction between within-trial and across-trial dynamics as a form of social canalization. When stakes are low in a real-time environment, players can satisfactorily coordinate `on the fly,’ but when stakes are high there is increased pressure to establish and adhere to shared expectations that persist across rounds.
Hmeljak, D., & Goldstone, R. L. (2016). Avatars and behavioral experiments: methods for controlled quantitative social behavioral research in virtual worlds. In Y. Silvan (Ed.) Handbook on 3D3C Virtual Worlds. Zurich, Switzerland: Springer International Publishing.
Three-dimensional, Community, Creation, and Commerce (3D3C) worlds can support real-time, quantitatively controlled experiments for studying human group behavior. This chapter provides a review of social behavioral research in virtual worlds, their methodologies and goals, such as studies of socio-economical trends, interpersonal communications between virtual world residents, automated survey studies, etc. The chapter contrasts existing research tools in virtual worlds with the goals of studying human group behavior as a complex system—how interacting groups of people create emergent organizations at a higher level than the individuals comprising such groups. Finally, the chapter presents features of virtual world-based group behavior experiments that allow the recreation of controlled quantitative experiments previously conducted in supervised lab sessions or web-based games.
We consider a situation in which individuals search for accurate decisions without direct feedback on their accuracy, but with information about the decisions made by peers in their group. The “wisdom of crowds” hypothesis states that the average judgment of many individuals can give a good estimate of, for example, the outcomes of sporting events and the answers to trivia questions. Two conditions for the application of wisdom of crowds are that estimates should be independent and unbiased. Here, we study how individuals integrate social information when answering trivia questions with answers that range between 0% and 100% (e.g., “What percentage of Americans are left-handed?”). We find that, consistent with the wisdom of crowds hypothesis, average performance improves with group size. However, individuals show a consistent bias to produce estimates that are insufficiently extreme. We find that social information provides significant, albeit small, improvement to group performance. Outliers with answers far from the correct answer move toward the position of the group mean. Given that these outliers also tend to be nearer to 50% than do the answers of other group members, this move creates group polarization away from 50%. By looking at individual performance over different questions we find that some people are more likely to be affected by social influence than others. There is also evidence that people differ in their competence in answering questions, but lack of competence is not significantly correlated with willingness to change guesses. We develop a mathematical model based on these results that postulates a cognitive process in which people first decide whether to take into account peer guesses, and if so, to move in the direction of these guesses. The size of the move is proportional to the distance between their own guess and the average guess of the group. This model closely approximates the distribution of guess movements and shows how outlying incorrect opinions can be systematically removed from a group resulting, in some situations, in improved group performance. However, improvement is only predicted for cases in which the initial guesses of individuals in the group are biased.
[This paper is a commentary on the following article: Gintis, H., & Helbing, D. (2015). Homo Socialis: An Analytical Core for Sociological Theory. Review of Behavioral Economics.]
Explaining how patterns of collective behavior emerge from interactions among individuals with diverse, sometimes opposing, goals is a societally crucial and particularly timely pursuit. It is timely because humans are more tightly connected to one another now than ever before. From 1984 to 2014 there has been more than a million-fold increase in the number of devices that can reach the global digital network. Although web technology is new and transformative, from a broader perspective, it is also just a recent manifestation of humanity’s perpetual drive to become more intermeshed. Earlier manifestations of this drive include the printing press, global transportation networks, telecommunication systems, and the academy. These social networks have catalyzed the formation of otherwise unattainable social patterns. Understanding the origins and possible destinations of these social patterns is both scientifically and pragmatically consequential.
Our “Creature League” study has been mentioned at Science Daily, ScienceNewsline, IU’s News Room, Medical Xpress, EurekAlert!, and Science Codex. Here’s an audio description of the work, courtesy of Academic Minute. Participants in the group behavior experiment of Wisdom, Song, and Goldstone (2013) tried to assemble teams of Pokemon-like creatures that scored well. Each creature was associated with a score for itself, but some pairs of creatures also produced positive or negative scores. Because of these interactions between creatures, the problem of assembling high-scoring teams posed a difficult search problem for participants. Participants could assemble their teams by 1) using their previous teams (status quo), 2) taking creatures from their historically best team (retrieval), 3) dragging untested creatures from the league of creatures (innovating), or 4) dragging individual creatures or entire teams from other participants’ solutions (imitating).
Some of the interesting results from this study were:
1) Participants tend to do BETTER when surrounded by imitators. One of the primary mechanisms for this is that when a person comes up with a good solution, their peers copy the solution, and sometime improve upon it. The person who was originally imitated can then benefit from these subsequent solutions (cliff swallows show a similar collective dynamic, with birds benefitting by being imitated while foraging). Imitation also acts as a cultural memory for what has worked well in the past. If an innovator’s solution to a problem is preserved by imitators, then the innovator does not have to remember their solution themselves.
2) As problem increased in difficulty, solutions were less diverse, and exploration was less prevalent.
3) Participants were more likely to imitate popular choices. above and beyond what would be expected from random copying of solution elements.
4) Participants are more likely to imitate a solution that is increasing in popularity among peers.
5) Participants are more likely to imitate solutions that are similar to their current solutions. This helps avoid hybrids/cross-breeds that don’t score well.
6) Participants begin a game by imitating and innovating relatively often, and end by more conservatively sticking to their existing solution. The best scoring strategy was to stick close to an existing solution, and innovating was worst.
7) At a group level, diversity of solutions decreased over rounds of a game. Bigger groups did better, but bigger groups also showed less diversity.
Leydesdorff, L., & Goldstone, R. L. (2014). Interdisciplinarity at the Journal and Specialty Level: The changing knowledge bases of the journal Cognitive Science. Journal of the American Society for Information Science and Technology, 65, 164-177.
Using the referencing patterns in articles in Cognitive Science over three decades, we analyze the knowledge base of this literature in terms of its changing disciplinary composition. Three periods are distinguished: (1) construction of the interdisciplinary space in the 1980s; (2) development of an interdisciplinary orientation in the 1990s; (3) reintegration into “cognitive psychology” in the 2000s. The fluidity and fuzziness of the interdisciplinary delineations in the different visualizations can be reduced and clarified using factor analysis. We also explore newly available routines (“CorText”) to analyze this development in terms of “tubes” using an alluvial map, and compare the results with an animation (using “visone”). The historical specificity of this development can be compared with the development of “artificial intelligence” into an integrated specialty during this same period. “Interdisciplinarity” should be defined differently at the level of journals and of specialties.
Here are some reports of our PLoS One paper on human collective behavior studying cyclic patterns in a generalization of the familiar rock-scissors-paper game. We find situations in which groups of people grow increasingly predictable as they cycle around a set of choice options in a game similar to rock-scissors-paper but with 24 rather than 3 choices.
When making decisions, humans can observe many kinds of information about others’ activities, but their effects on performance are not well understood. We investigated social learning strategies using a simple problem-solving task in which participants search a complex space, and each can view and imitate others’ solutions. Results showed that participants combined multiple sources of information to guide learning, including payoffs of peers’ solutions, popularity of solution elements among peers, similarity of peers’ solutions to their own, and relative payoffs from individual exploration. Furthermore, performance was positively associated with imitation rates at both the individual and group levels. When peers’ payoffs were hidden, popularity and similarity biases reversed, participants searched more broadly and randomly, and both quality and equity of exploration suffered. We conclude that when peers’ solutions can be effectively compared, imitation does not simply permit scrounging, but it can also facilitate propagation of good solutions for further cumulative exploration.
Recent theories from complexity science argue that complex dynamics are ubiquitous in social and economic systems. These claims emerge from the analysis of individually simple agents whose collective behavior is surprisingly complicated. However, economists have argued that iterated reasoning–what you think I think you think–will suppress complex dynamics by stabilizing or accelerating convergence to Nash equilibrium. We report stable and efficient periodic behavior in human groups playing the Mod Game, a multi-player game similar to Rock-Paper-Scissors. The game rewards subjects for thinking exactly one step ahead of others in their group. Groups that play this game exhibit cycles that are inconsistent with any fixed-point solution concept. These cycles are driven by a ‘‘hopping’’ behavior that is consistent with other accounts of iterated reasoning: agents are constrained to about two steps of iterated reasoning and learn an additional one-half step with each session. If higher-order reasoning can be complicit in complex emergent dynamics, then cyclic and chaotic patterns may be endogenous features of real-world social and economic systems.
Download PDF version of this paper
See a movie of actual humans (shown as Xs) playing the Mod Game. Notice the clumping of their moves and their regular progression around the circle of choices.
Unlike how most psychology experiments on learning operate, people learning to do a task typically do so in the context of other people learning to do the same task. In these situations, people take advantage of others’ solutions, and may modify and extend these solutions, thereby affecting the solutions available to others. We are interested in the group patterns that emerge when people can see and imitate the solutions, innovations, and choices of their peers over several rounds. In one series of experiments and computer simulations, we find that there is a systematic relation between the difficulty of a problem search space and the optimal social network for transmitting solutions. As the difficulty of finding optimal solutions in a search space increases, communication networks that preserve spatial neighborhoods perform best. Restricting people’s access to others’ solutions can help the group as a whole find good, hard-to-discover solutions. In other experiments with more complex search spaces, we find evidence for several heuristics governing individuals’ decisions to imitate: imitating prevalent options, imitating options that become increasingly prevalent, imitating high-scoring options, imitating during the early stages of a multiround search process, and imitating solutions similar to one’s own solution. Individuals who imitate tend to perform well, and more surprisingly, individuals also perform well when they are in groups with other individuals who imitate frequently. Taken together, our experiments on collective social learning reveal laboratory equivalents of prevalent social phenomena such as bandwagons, strategy convergence, inefficiencies in the collective coverage of a problem space, social dilemmas in exploration/exploitation, and reciprocal imitation.
Click here to learn more about this project
We implemented a problem-solving task in which groups of participants simultaneously played a simple innovation game in a complex problem space, with score feedback provided after each of a number of rounds. Each participant in a group was allowed to view and imitate the guesses of others during the game. The results showed the use of social learning strategies previously studied in other species, and demonstrated benefits of social learning and nonlinear effects of group size on strategy and performance. Rather than simply encouraging conformity, groups provided information to each individual about the distribution of useful innovations in the problem space. Imitation facilitated innovation rather than displacing it, because the former allowed good solutions to be propagated and preserved for further cumulative innovations in the group. Participants generally improved their solutions through the use of fairly conservative strategies, such as changing only a small portion of one’s solution at a time, and tending to imitate solutions similar to one’s own. Changes in these strategies over time had the effect of making solutions increasingly entrenched, both at individual and group levels. These results showed evidence of nonlinear dynamics in the decentralization of innovation, the emergence of group phenomena from complex interactions of individual efforts, stigmergy in the use of social information, and dynamic tradeoffs between exploration and exploitation of solutions. These results also support the idea that innovation and creativity can be recognized at the group level even when group members are generally cautious and imitative.
Frey, S., & Goldstone, R. L. (2011). Going with the group in a competitive game of iterated reasoning. Proceedings of the Thirty-Third Annual Conference of the Cognitive Science Society. (pp. 1912-1917). Boston, Massachusetts: Cognitive Science Society.
In some strategic games, thinking ahead about other players’ reasoning can lead to better predictions about what they will do. In other games, infinitely iterated reasoning ultimately prescribes random play. In an online experiment of strategic thinking in groups, we tested participants in a game with the formal structure of a random game, but the superficial struc- ture of a game that rewards iterated reasoning. We found that participants conformed to the superficial structure of the game, and earned more than they would have by playing randomly. We estimated how many steps participants thought ahead in the game and discovered implicit coordination at the group level. Participants unexpectedly “matched” their degree of iterated thinking to each other.
Roberts, M. E., & Goldstone, R. L. (2011). Adaptive Group Coordination and Role Differentiation. PLoS One, 6, 1-8.
Many real world situations (potluck dinners, academic departments, sports teams, corporate divisions, committees, seminar classes, etc.) involve actors adjusting their contributions in order to achieve a mutually satisfactory group goal, a win-win result. However, the majority of human group research has involved situations where groups perform poorly because task constraints promote either individual maximization behavior or diffusion of responsibility, and even successful tasks generally involve the propagation of one correct solution through a group. Here we introduce a group task that requires complementary actions among participants in order to reach a shared goal. Without communication, group members submit numbers in an attempt to collectively sum to a randomly selected target number. After receiving group feedback, members adjust their submitted numbers until the target number is reached. For all groups, performance improves with task experience, and group reactivity decreases over rounds. Our empirical results provide evidence for adaptive coordination in human groups, and as the coordination costs increase with group size, large groups adapt through spontaneous role differentiation and self-consistency among members. We suggest several agent-based models with different rules for agent reactions, and we show that the empirical results are best fit by a flexible, adaptive agent strategy in which agents decrease their reactions when the group feedback changes. The task offers a simple experimental platform for studying the general problem of group coordination while maximizing group returns, and we distinguish the task from several games in behavioral game theory.
Frey, S., & Goldstone, R. L. (2010). Functional Structure and Coordination Failure in Real-Time Group Behavior. Proceedings of the Thirty-Second Annual Conference of the Cognitive Science Society. (pp. 2093-2098). Portland, Oregon: Cognitive Science Society.
We reveal spontaneous group formation and differentiation in an online dynamic coordination experiment. We observe increased group stratification and attribute it to increases in pairwise cooperative behavior, rather than uncooperative behavior. Our network analyses document the fine scale structure of coordination failure in the face of many established determinants of coordination success. We explore previous work in coordination failure to frame our own findings. Factors that have been previously shown to improve coordination in discretetime, forced-decision experimental games do not prevent decisive coordination failure in our real-time, asynchronous group decision-making environment.
Theiner, G, Allen, C., & Goldstone, R. L. (2010). Recognizing group cognition. Cognitive Systems Research, 11, 378-395.
In this paper, we approach the idea of group cognition from the perspective of the ”extended mind” thesis, as a special case of the more general claim that systems larger than the individual human, but containing that human, are capable of cognition (Clark, 2008; Clark & Chalmers, 1998). Instead of deliberating about ”the mark of the cognitive” (Adams & Aizawa, 2008), our discussion of group cognition is tied to particular cognitive capacities. We review recent studies of group problem solving and group memory which reveal that specific cognitive capacities that are commonly ascribed to individuals are also aptly ascribed at the level of groups. These case studies show how dense interactions among people within a group lead to both similarity-inducing and differentiating dynamics that affect the group’s ability to solve problems. This supports our claim that groups have organization-dependent cognitive capacities that go beyond the simple aggregation of the cognitive capacities of individuals. Group cognition is thus an emergent phenomenon in the sense of Wimsatt (1986). We further argue that anybody who rejects our strategy for showing that cognitive properties can be instantiated at multiple levels in the organizational hierarchy on a priori grounds is a ”demergentist,” and thus incurs the burden of proof for explaining why cognitive properties are ”stuck” at a certain level of organizational structure. Finally, we show that our analysis of group cognition escapes the ”coupling-constitution” charge that has been leveled against the extended mind thesis (Adams & Aizawa, 2008).
Wisdom, T. N., & Goldstone, R. L. (2010). Social Learning and Cumulative Mutual Improvement in a Networked Group. Proceedings of the Thirty-Second Annual Conference of the Cognitive Science Society. (pp. 1405-1410). Portland, Oregon: Cognitive Science Society.
We used a simple problem-solving game task to study imitation and innovation in groups of participants. Guesses were composed of multiple elements with linear and interactive effects on score, and score feedback was provided after each of a number of rounds. Participants were allowed to view and imitate the guesses of others during each round, and the score information accompanying others’ guesses was either shown or hidden in two conditions. When scores were not visible, social learning was impeded; participants were less efficient in their searching of the problem space and achieved lower performance overall. When scores were visible, higher performance was observed, and results indicated a more equitable sharing of productive exploration among participants within groups as a result of selective imitation and cross-participant cumulative mutual innovations.
Gureckis, T. M., & Goldstone, R. L. (2009). How you named your child: Understanding the relationship between individual decision-making and collective outcomes. Topics in Cognitive Science, 1, 651-674.
We examine the interdependence between individual and group behavior surrounding a somewhat arbitrary, real world decision: selecting a name for
one’s child. Using a historical database of the names given to children over the last century in the United States, we nd that naming choices are influenced by both the frequency of a name in the general population, and by its “momentum” in the recent past in the sense that names which are growing in popularity are preferentially chosen. This bias toward rising names is a recent phenomena: in the early part of the 20th century, increasing popularity of a name from one time period to the next was correlated with a decrease in future popularity. However, more recently this trend has reversed. We evaluate a number of formal models that detail how individual decision-making strategies, played out in a large population of interacting agents, can explain these empirical observations. We argue that cognitive capacities for change detection, the encoding of frequency in memory, and biases towards novel or incongruous stimuli may interact with the behavior of other decision makers to determine the distribution and dynamics of cultural tokens such as names.
Goldstone, R. L. & Gureckis, T. M. (2009). Collective behavior. Topics in Cognitive Science, 1, 412-438.
The resurgence of interest in collective behavior is in large part due to tools recently made available for conducting laboratory experiments on groups, statistical methods for analyzing large data sets reflecting social interactions, the rapid growth of a diverse variety of online self-organized collectives, and computational modeling methods for understanding both universal and scenario-specific social patterns. We consider case studies of collective behavior along four attributes: the primary motivation of individuals within the group, kinds of interactions among individuals, typical dynamics that result from these interactions, and characteristic outcomes at the group level. With this framework, we compare the collective patterns of noninteracting decision makers, bee swarms, groups forming paths in physical and abstract spaces, sports teams, cooperation and competition for resource usage, and the spread and extension of innovations in an online community. Some critical issues surrounding collective behavior are then reviewed, including the questions of ‘‘Does group behavior always reduce to individual behavior?’’ ‘‘Is ‘group cognition’ possible?’’ and ‘‘What is the value of formal modeling for understanding group behavior?’’
Roberts, M. E., & Goldstone, R. L. (2009). Adaptive group coordination, Proceedings of the Thirty-First Annual Conference of the Cognitive Science Society. 2698-2704. Amsterdam, Netherlands: Cognitive Science Society.
Human groups exhibit poor performance in many social situations because task constraints promote either individual maximization behavior or diffusion of responsibility. We introduce a group task that tests human coordination when only a shared group goal exists. Without communication, group members submit numbers in an attempt to collectively sum to a randomly selected number. After receiving group feedback, members adjust their submitted numbers in the next round. Small groups generally outperform large groups, and for all groups, performance improves with task experience, and reactivity to feedback decreases over rounds. Our empirical results and computational modeling provide evidence for adaptive coordination in human groups despite minimal shared history and only indirect communication, and perhaps most interestingly, as the coordination costs increase with group size, large groups adapt through spontaneous role differentiation and self-consistency among members.
Roberts, M. E., & Goldstone, R. L. (2009). Sub-optimalities in group foraging and resource competition. , Proceedings of the Thirty-First Annual Conference of the Cognitive Science Society 2371-2377. Amsterdam, Netherlands: Cognitive Science Society.
Previous group foraging research has shown that human groups sub-optimally distribute themselves to resources and display undermatching, with a smaller-than-expected proportion of individuals at the more abundant resource pool. In order to further explore these sub-optimalities, we extended a group foraging paradigm to test three variables: the effects of three resource pools, travel cost between pools, and the size of the pools. Although each condition led to undermatching, the conditions showed significant differences in the extent of undermatching, the frequency of switching between resource pools, the wealth inequality among foragers, and the comparative wealth inequality at different resource pools. The results for the three pool conditions suggest that human groups have difficulty in discriminating the relative value of resource pools. The results for the travel cost conditions indicate that human groups distribute themselves to resources more optimally when individuals can easily switch between pools, which is the opposite of the result found with foraging pigeons. Finally, the results for the pool size conditions indicate that larger pool sizes promote greater undermatching, apparently because individuals inefficiently compete over large areas rather than effectively parceling the pools into smaller, distinct regions.
Searching in space and minds: IU research suggests underlying linkarticle in E Science News (September 12, 2008),UPI, Scientific American and Science Daily
Goldstone, R. L., Roberts, M. E., & Gureckis, T. M. (2008). Emergent Processes in Group Behavior. Current Directions in Psychological Science, 17, 10-15.
Just as networks of neurons create structured thoughts beyond the ken of any individual neuron, so people spontaneously organize themselves into groups to create emergent organizations that no individual may intend, comprehend, or even perceive. Recent technological advances have provided us with unprecedented opportunities for conducting controlled, laboratory experiments on human collective behavior. We describe two experimental paradigms where we attempt to build predictive bridges between the beliefs, goals, and cognitive capacities of individuals and group-level patterns, showing how the members of a group dynamically allocate themselves to resources, and how innovations are spread in a social network. Agent-based computational models have provided useful explanatory and predictive accounts. Together, the models and experiments point to tradeoffs between exploration and exploitation, compromises between individuals using their own innovations and innovations obtained from their peers, and the emergence of group-level organizations such as population waves, bandwagon effects, and spontaneous specialization.
Goldstone, R. L., Roberts, M. E., Mason, W., & Gureckis, T. (2008). Collective search in concrete and abstract spaces. In T. Kugler, J. C. Smith, T. Connelly, and Y. Sun (Eds.) Decision modeling and behavior in complex and uncertain environments. New York: Springer Press. (pp. 277-308).
Our laboratory has been studying the emergence of collective search behavior from a complex systems perspective. We have developed an internet-based experimental platform that allows groups of people to interact with each other in real time on networked computers. The experiments implement virtual environments where participants can see the moment-to-moment actions of their peers and immediately respond to their environment. Agent-based computational models are used as accounts of the experimental results. We describe two paradigms for collective search – one in physical space and the other in an abstract problem space. The physical search situation concerns competitive foraging for resources by individuals inhabiting an environment consisting largely of other individuals foraging for the same resources. The abstract search concerns the dissemination of innovations in social networks. Across both scenarios, the group-level behavior that emerges reveals influences of exploration and exploitation, bandwagon effects, population waves, and compromises between individuals using their own information and information obtained from their peers.
Janssen, M. A., Goldstone, R. L., Menczer, F., & Ostrom, E. (2008). Effect of rule choice in dynamic interactive spatial commons. International Journal of the Commons, 2, 288-312.
This paper uses laboratory experiments to examine the effect of an endogenous rule change from open access to private property as a potential solution to over-harvesting in commons dilemmas. A novel, spatial, real-time renewable resource environment was used to investigate whether participants were willing to invest in changing the rules from an open access situation to a private property system. We found that half of the participants invested in creating private property arrangements. Groups who had experienced private property in the second round of the experiment, made different decisions in the third round when open access was re-instituted in contrast to groups who experienced three rounds of open access. At the group level, earnings increased in Round 3, but this was at a cost of more inequality. No significant differences in outcomes occurred between experiments where rules were imposed by the experimental design or chosen by participants.
Mason, W. A., Jones, A., & Goldstone, R. L. (2008). Propagation of innovations in networked groups. Journal of Experimental Psychology: General, 137, 422-433.
A novel paradigm was developed to study the behavior of groups of networked people searching a problem space. We examined how different network structures affect the propagation of information in laboratory-created groups. Participants made numerical guesses and received scores that were also made available to their neighbors in the network. The networks were compared on speed of discovery and convergence on the optimal solution. One experiment showed that individuals within a group tend to converge on similar solutions even when there is an equally valid alternative solution. Two additional studies demonstrated that the optimal network structure depends on the problem space being explored, with networks that incorporate spatially-based cliques having an advantage for problems that benefit from broad exploration, and networks with greater long-range connectivity having an advantage for problems requiring less exploration.
Wisdom, T. N., Song, X., & Goldstone, R. L. (2008). The effects of peer information on problem-solving in a networked group. Proceedings of the Thirtieth Annual Conference of the Cognitive Science Society, , (pp. 583-588). Washington, D.C.: Cognitive Science Society
In this experiment, we implemented a problem-solving task in which groups of participants simultaneously play a simple puzzle game, with score feedback provided after each of 24 rounds. Each participant in a group is allowed to view and imitate the guesses of others during the game. Results show that when the utility of others’ innovations is unambiguous, individuals base their own solutions on personal innovation and productively imitate other players’ innovations early on, and that this tendency to imitate is proportional to the relative amount of information available from others. Average trends of innovation and imitation decreased across rounds as player guesses stabilized and scores increased. Mean scores and imitation increased with group size, while individual innovation decreased. Results are consistent with previously studied social learning strategies in several taxa.
Back when Dwight Eisenhower was president of Columbia University, he was asked how the university should arrange the sidewalks to best interconnect the campus buildings. He responded that they should first plant grass seed, let the grass grow, see where the grass became worn by people’s footsteps, and install the sidewalks in the most worn patches. The Percepts and Concepts Laboratory (directed by Chancellor’s professor Robert Goldstone, also director of the Cognitive Science Program) at Indiana University has put Eisenhower’s proposal to the empirical test, asking what kinds of trails people will spontaneously form when they are motivated to take advantage of the trails left by their predecessors. Early trail blazers through a jungle use machetes to make slow progress in building paths – progress that is capitalized on and extended by later trekkers, who may then widen the trail, then later put stones down, gravel, asphalt, and eventually an eight-lane highway.
In the article “Self-organized Trail Systems in Groups of Humans” (appearing in the July/August issue of the journal Complexity, available at http://cognitrn.psych.indiana.edu/papers.html), Robert Goldstone and Michael Roberts report the results of a group experiment in which people collectively travel among random destinations in a virtual world. As they step on a location, they change their environment, making it easier for subsequent walkers to step on the same location. In this way, a trail left by a walker often leads other walkers to follow the same trail, thereby reinforcing and extending the trail.
The trails that our experimental groups of participants created are compromises between people going directly to their destinations, and taking paths of least effort. The trail network that completely connects a set of destinations using the minimal amount of total trail length is called a Minimal Steiner Tree. While soap films reliably create Minimal Steiner Trees, our human collectives did not. However, their paths did deviate away from bee-line paths to destinations, in the direction of Minimal Steiner Trees.
We modeled our results by adapting a model from biophysics (Helbing, Keltsch, & Molnár, 1997) that is based on Brownian motion within a field potential, and has been applied to ant trails. This model, which assumes that travelers’ steps are a compromise between going where they want to go and where others have gone before, did a good job of reproducing the trails that our groups formed. The growth of our collectively produced trails offers the promise of revealing principles about how future progress is achieved by exploiting and extending prior innovations. Our experiments and simulations also provide a rigorous way of following the poet Antonio Machado’s exhortation: “Traveler, there is no path. Paths are made by walking.”
Goldstone, R. L., & Roberts, M. E. (2006). Self-organized trail systems in groups of humans. Complexity, 15, 43-50.
Helbing, D., Keltsch, J., & Molnár, P. (1997). Modeling the evolution of human trail systems. Nature, vol. 388, 47-50.
April 13, 2006
The Percepts and Concepts Laboratory (Directed by Chancellor’s Professor Robert Goldstone, also Director of the Indiana University Cognitive Science Program) applies formal computational and mathematical tools used to study complex systems in biology and physics to understanding human collective behavior. People participate in group-level patterns that they may not understand, or even perceive. Our goals are to conduct experiments that reveal the patterns that groups of people spontaneously create, and to develop computational models that show how these patterns emerge from simple interactions among people.
One common situation that we have formally explored is how groups of people distribute themselves to valuable resources. Morel hunters forage their environment for mushrooms, drivers patrol downtown for convenient parking spaces, web-users surf the internet for desired data, and businesses mine the land for valuable minerals. When an organism forages in an environment that consists, in part, of other organisms that are also foraging, then interesting complexities arise. The resources available to an organism are affected not just by the foraging behavior of the organism itself, but also by the simultaneous foraging behavior of all of the other organisms.
In a series of experiments, we have developed a novel experimental technique for studying human foraging behavior (Goldstone & Ashpole, 2004; Goldstone, Ashpole, & Roberts, 2005). We have created an experimental platform that allows many human participants to interact in real-time within a common virtual environment. Resource pools are created within this environment, participants vie for these resources, and we record the moment-by-moment exploitation of these resources by each participant. The participants’ task is to obtain as many resource tokens as possible during an experiment.
Groups of animals generally distribute themselves well to resource patches. For example, mallard ducks, cichlid fish, and dung beetles all approximately match their numbers to the amount of resource. If twice as much bread is thrown in one pond location than another, then about twice as many ducks will spontaneously go to the more plentiful location. Our groups of humans, recruited from psychology courses, are fairly efficient and about as smart, collectively speaking, as ducks, fish, and dung beetles. However, we also find two important collective inefficiencies in their harvesting.
First, we find that people do not distribute themselves in an extreme enough manner. For example, if one pool produces 80% of the tokens and the other pool produces 20%, people distribute themselves in about a 73%/27% fashion. People who harvest the richer resource patch tend to earn more tokens than those harvesting the poorer patch. If this proves general, our advice is for people to try harvesting the richer patch: fish in pond locations known to be plentiful, study for professions that are hot, and visit bars with attractive people. Even though rich patches will attract more competitors foraging for the same resources, the number of people will not keep up with the patch’s advantage if our experiments generalize.
Second, we find cycles in the harvesting rates over time. In our experiment, these cycles come in 50 second waves of migration into and out of patches. Due to random fluctuations, more people will end up at one patch than another. The people in this over-crowded patch will tend to become dissatisfied with their token earnings, and will decide to leave the patch for hopefully greener pastures elsewhere. However, if they cannot see other people’s movements, they do not realize that what has made them decide to leave is influencing others as well. The result is roughly synchronized waves of migration. Ironically, it is precisely because people share the desire to avoid crowds that migratory crowds emerge! When people can see where other people are in the virtual world, then these waves of crowding do not arise.
We have developed a computational model of foraging behavior that reproduces the results from our experiments (Roberts & Goldstone, 2005). In this model, we create simple rules for each of the agents in a population, and observe the collective patterns that emerge. The assumptions that are critical for getting human-like results are: 1) people are lazy (agents tend to go for tokens that are close), 2) people have inertia (agents tend to keep moving toward a selected token once they have started), 3) people go where the gold is (as the number of tokens in a patch increases, agents will congregate there), 4) people avoid crowds (when agents can see the other agents and all of the tokens in the virtual world, they tend to avoid crowds), and 5) people act like buzzards (when agents can see each other but not the tokens, then they use the presence of other agents to indicate that tokens might be nearby).
Web-citizens can experience these experiments for themselves by visiting http://groups.psych.indiana.edu/. This site offers several ongoing experiments that run continuously 24 hours per day. Participants are automatically grouped together into experiments and play in 4-minute rounds. If there aren’t enough human participants at any given time, then we generate artificially intelligent ‘bots’ to keep the humans company in the virtual worlds. As it turns out, these bots are exactly our computational models of how people forage for resources. Like the ‘human be-in’ events of the 1960s and modern flash mobs, people participating in these experiments can experience what it feels like to be part of a collective mind that adapts to its environment.
Goldstone, R. L., Ashpole, B. C., & Roberts, M. E., (2005). Knowledge of resources and competitors in human foraging. Psychonomic Bulletin & Review, 12, 81-87.
Goldstone, R. L., & Ashpole, B. C. (2004). Human foraging behavior in a virtual environment. Psychonomic Bulletin & Review, 11, 508-514.
Roberts, M. E., & Goldstone, R. L. (2005). Explaining resource undermatching with agent-based models. Proceedings of the Twenty-seventh Annual Conference of the Cognitive Science Society. Hillsdale, New Jersey: Lawrence Erlbaum Associates.
‘Trailblazing’ Video Game Offers Model For Human Behavior’ IU Press Release (Sept, 2006) , also appearing inHouston Chronicle, TX – Sep 11, 2006, Myrtle Beach Sun News, SC – Sep 10, 2006, Macon Telegraph, GA – Sep 10, 2006, Belleville News-Democrat, IL – Sep 10, 2006, Contra Costa Times, CA – Sep 10, 2006, Biloxi Sun Herald, MS – Sep 10, 2006, Kentucky.com, KY – Sep 10, 2006, Duluth News Tribune, MN – Sep 10, 2006, Monterey County Herald, CA – Sep 10, 2006, Kansas City Star, MO – Sep 10, 2006, San Luis Obispo Tribune, CA – Sep 10, 2006, Charlotte Observer, NC – Sep 10, 2006
Goldstone, R. L., & Leydesdorff, L. (2006). The import and export of Cognitive Science.Cognitive Science, 30, 983-993.
From its inception, a large part of the motivation for Cognitive Science has been the need for an interdisciplinary journal for the study of minds and intelligent systems. In the inaugural editorial for the journal, Allan Collins (1977) wrote “Current journals are fragmented along old disciplinary lines, so there is no common place for workers who approach these problems from different disciplines to talk to each other” (p. 1). The interdisciplinarity of the journal has served a valuable cross-fertilization function for those who read the journal to discover articles written for and by practitioners across a wide range of fields. The challenges of building and understanding intelligent systems are sufficiently large that they will most likely require the skills of psychologists, computer scientists, philosophers, educators, neuroscientists, and linguists collaborating and coordinating their efforts.
Janssen, M. A., & Goldstone, R. L. (2006). Dynamic-persistence of cooperation in public good games when group size is dynamic. Journal of Theoretical Biology, 234, 134-142.
The evolution of cooperation is possible with a simple model of a population of agents that can move between groups. The agents play public good games within their group. The relative fitness of individuals within the whole population affects their number of offspring. Groups of cooperators evolve but over time are invaded by defectors which eventually results in the group’s extinction. However, for small levels of migration and mutation, high levels of cooperation evolve at the population level. Thus, evolution of cooperation based on individual fitness without kin selection, indirect or direct reciprocity is possible. We provide an analysis of the parameters that affect cooperation, and describe the dynamics and distribution of population sizes over time.
Gureckis, T. M., & Goldstone, R. L. (2006). Thinking in groups. Pragmatics and Cognition, 14, 293-311
Is cognition an exclusive property of the individual or can groups have a mind of their own? We explore this question from the perspective of complex adaptive systems. One of the principle insights from this line of work is that rules that govern behavior at one level of analysis (the individual) can cause qualitatively different behavior at higher levels (the group). We review a number of behavioral studies from our lab that demonstrate how groups of people interacting in real-time can self-organize into adaptive, problem-solving group structures. A number of principles are derived concerning the critical features of such “distributed” information processing systems. We suggest that while cognitive science has traditionally focused on the individual, cognitive processes may manifest at many levels including the emergent group-level behavior that results from the interaction of multiple agents and their environment.
Roberts, M. E., & Goldstone, R. L. (2006). EPICURE: Spatial and Knowledge Limitations in Group Foraging. Adaptive Behavior, 14, 291-313.
We propose an agent-based model of group foraging, EPICURE, for patchily distributed resources. Each agent makes probabilistic movement decisions in a gridworld through a linear combination of current perceptual information and a reinforcement history. EPICURE captures the empirical results from several foraging conditions in Goldstone and Ashpole (2004) and Goldstone, Ashpole, and Roberts (2005), and it leads to a re-evaluation of findings from those papers. In particular, human foragers show contingent usage of information, initially using social information to discover resource pools before private sampling information has been established. We describe a series of simulations that test the sources of resource undermatching often found in group foraging experiments. After testing the effects of foragers’ starting locations, travel costs, the number of foragers, and the size of uniform food distributions, we discuss a novel hypothesis for undermatching. Spatial constraints lead to inadequate individual and group information sampling and cause group undermatching. The foraging group size, food rate, spatial distribution of food, and resulting forager reinforcement histories interact to produce undermatching, and occasionally overmatching, to resources.
Goldstone, R. L., Jones, A., & Roberts, M. E. (2006). Group path formation. IEEE Transactions on System, Man, and Cybernetics, Part A, 36, 611-620.
When people make choices within a group, they are frequently influenced by the choices made by others. We have experimentally explored the general phenomenon of group behavior where an early action facilitates subsequent actions. Our concrete instantiation of this problem is group path formation where people travel between destinations with the travel cost for moving onto a location inversely related to the frequency with which others have visited the location. We compare the resulting paths to optimal solutions [Minimal Steiner Trees (MSTs)] and the “Active Walker” model of pedestrian motion from biophysics. There were systematic deviations from beeline pathways in the direction of MST. These deviations showed asymmetries (people took different paths from A to B than they did from B to A) and varied as a function of the topology of the destinations, the duration of travel, and the absolute scale of the world. The Active Walker model accounted for many of these results, in addition to correctly predicting the approximate spatial distribution of steps.
Goldstone, R. L., & Roberts, M. E. (2006). Self-organized trail systems in groups of humans. Complexity, 11, 43-50.
We have developed an experimental platform for studying the trail systems that spontaneously emerge when people are motivated to take advantage of the trails left by others. In this virtual environment, the participants’ task is to reach randomly selected destinations whileminimizing travel costs. The travel cost of every patch in the environment is inversely related to the number of times the patch was visited by others. The resulting trail systems are a compromise between people going to their destinations and going where many people have previously traveled. We compare the results from our group experiments to the Active Walker model of pedestrian motion from biophysics. The ActiveWalker model accounted for deviations of trails from the beeline paths, the gradual merging of trails over time, and the influences of scale and configuration of destinations on trail systems, as well as correctly predicting the approximate spatial distribution of people’s steps. Two deviations of the model from empirically obtained results were corrected by (1) incorporating a distance metric sensitive to canonical horizontal and vertical axes, and (2) increasing the influence of a trail’s travel cost on an agent’s route as the agent approaches its destination.
Roberts, M., & Goldstone, R. L. (2006). EPICURE: An agent-based foraging model. Artificial Life X: Proceedings of the Tenth International Conference on the Simulation and Synthesis of Living Systems. Cambridge, MA: MIT Press (379-385).
We present an agent-based foraging model, EPICURE, which captures the results from recent human group foraging experiments (Goldstone and Ashpole, 2004; Goldstone et al., 2005), provides a novel explanation for those results and previous animal foraging results, and makes predictions for future foraging experiments. We describe a series of simulations that test the sources of resource undermatching often found in group foraging experiments. We conclude that foraging group size, food rate, and spatial distribution of food interact to produce undermatching, and occasionally, overmatching, to resources. Furthermore, we present wealth distribution results from the aforementioned empirical studies and EPICURE simulations.
Mason, W. A., Jones, A., & Goldstone, R. L. (2005). Propagation of innovations in networked groups. Proceedings of the Twenty-seventh Annual Conference of the Cognitive Science Society. Hillsdale, New Jersey: Lawrence Erlbaum Associates. (pp. 1419-1424)
A novel paradigm was developed to study the behavior of groups of networked humans searching a problem space. We examined how different network structures affect the diffusion of information about good solutions. Participants made numerical guesses and received scores that were also made available to their neighbors in the network. When the problem space was monotonic and had only one optimal solution, groups were fastest at finding the solution when all of the groups’ information was presented to them. However, when there were good but suboptimal solutions (i.e., local maxima), the group connected via a small-world network (Watts & Strogatz, 1998) was faster at finding the best solution than all other network structures.
Roberts, M. E., & Goldstone, R. L. (2005). Explaining resource undermatching with agent-based models. Proceedings of the Twenty-seventh Annual Conference of the Cognitive Science Society. Hillsdale, New Jersey: Lawrence Erlbaum Associates. (pp. 1872-1877)
We propose two agent-based models of group foraging for two perceptual conditions. These models exhibit complex group-level behavior using only a simple rule set with a homogeneous group of agents. The models are shown to replicate results from Goldstone and Ashpole (2004), and we describe a series of simulations that test the sources of the resource undermatching often found in group foraging experiments. After testing the effects of travel costs, the number of agents, and uniform variance food distributions, we conclude that many group foraging studies have overlooked the interplay of spatial constraints with food rates in causing undermatching.
Goldstone, R. L., & Janssen, M. A. (2005). Computational models of collective behavior. Trends in Cognitive Science, 9, 424-430.
Computational models of human collective behavior offer promise in providing quantitative and empirically verifiable accounts of how individual decisions lead to the emergence of group-level organizations. Agent-based models (ABMs) describe interactions among individual agents and their environment, and provide a process-oriented alternative to descriptive mathematical models. Recent ABMs provide compelling accounts of group pattern formation, contagion, and cooperation, and can be used to predict, manipulate, and improve upon collective behavior. ABMs overcome an assumption underlying much of cognitive science – that the individual is the critical unit of cognition. The advocated alternative is that individuals participate in collective organizations that they may not understand or even perceive, and that these organizations affect and are affected by individual behavior.
The allocation of human participants to resources was studied by observing the population dynamics of people interacting in real-time within a common virtual world. Resources were distributed in two spatially separated pools with varying relative reinforcement rates (50-50, 65- 35, or 80-20). We manipulated whether participants could see each other and the distribution of resources. When participants could see each other but not the resources, the richer pool was underutilized. When participants could see the resources but not each other, the richer pool was overutilized. In conjunction with prior experiments that correlated the visibility of agents and resources (Goldstone & Ashpole, in press), these results indicate that participants’ foraging decisions are influenced by both forager and resource information. The results suggest that the presence of a crowd at a resource is a deterring rather than attractive factor. Both fast and slow oscillations in the harvesting rates of the pools across time were revealed by Fourier analyses. The slow waves of crowd migration are most prevalent when the resources are invisible, whereas the fast cycles are most prevalent when the resources are visible and participants are invisible.
There has been a long history of research into the structure and evolution of mankind’s scientific endeavor. However, recent progress in applying the tools of science to understand science itself has been unprecedented because only recently has there been access to high-volume and high-quality data sets of scientific output (e.g., publications, patents, grants), as well as computers and algorithms capable of handling this enormous stream of data. This paper reviews major work on models that aim to capture and recreate the structure and dynamics of scientific evolution. We then introduce a general process model that simultaneously grows co-author and paper-citation networks. The statistical and dynamic properties of the networks generated by this model are validated against a 20-year data set of articles published in the Proceedings of the National Academy of Science. Systematic deviations from a power law distribution of citations to papers are well fit by a model that incorporates a partitioning of authors and papers into topics, a bias for authors to cite recent papers, and a tendency for authors to cite papers cited by papers that they have read. In this TARL model (for Topics, Aging, and Recursive Linking), the number of topics is linearly related to the clustering coefficient of the simulated paper citation network.
Our goal in this research is to collect a large volume of time-evolving data from a system composed of human agents vying for resources in a common environment, with the eventual aim of guiding the development of computational models of human resource allocation. We have developed an experimental platform that allows a large number of human participants to interact in real-time within a common virtual world. Two resource pools were created with different rates of replenishment. The participants’ task was to obtain as many resource tokens as possible during an experiment. In addition to varying the relative replenishment rate for the two resources (50-50, 65-35, 80-20), we manipulated whether agents could see each other and the entire food distribution, or had their vision restricted to food in their own location. As a collective, the agents would optimally harvest the resources if they distribute themselves proportionally to the distribution of resources. Empirical violations of global optimality were found. First, there was a systematic underutilization of the more preponderant resource. For example, agents distributed themselves approximately 75% and 25% to resources pools that had relative replenishment rates of 80% and 20%, respectively. The expected pay-off per agent was larger for pools with relatively high replenishment rates. Second, there were oscillations in the harvesting rates of the resources across time, particularly when agents’ vision was restricted. Perceived underutilization of a resource resulted in an influx of agents to that resource. This sudden influx, in turn, resulted in a glut of agents, which then led to a trend for agents to depart from the resource region. This cyclic activity in the collective data was revealed by a Fourier analysis showing prominent power in the range of about 50 seconds per cycle.
Goldstone, R. L., Ashpole, B. C. (2003). The distribution of people to resources in a networked multi-player environment. Proceedings of the 25th Annual Conference of the Cognitive Science Society.
This is an abridged version of Goldstone & Ashpole (2004).