Search

Can human decision-makers learn from AI feedback? Using data on 52,000 decision-makers from a large online chess platform, we investigate how their AI use affects three interrelated long-term outcomes: Learning, skill gap, and diversity of decision strategies. First, we show that individuals are far more likely to seek AI feedback in situations in which they experienced success rather than failure. This AI feedback seeking strategy turns out to be detrimental to learning: Feedback on successes decreases future performance, while feedback on failures increases it. Second, higher-skilled decision-makers seek AI feedback more often and are far more likely to seek AI feedback after a failure, and benefit more from AI feedback than lower-skilled individuals. As a result, access to AI feedback increases, rather than decreases, the skill gap between high- and low-skilled individuals. Finally, we leverage 42 major platform updates as natural experiments to show that access to AI feedback causes a decrease in intellectual diversity of the population as individuals tend to specialize in the same areas. Together, those results indicate that learning from AI feedback is not automatic and using AI correctly seems to be a skill itself. Furthermore, despite its individual-level benefits, access to AI feedback can have significant population-level downsides including loss of intellectual diversity and an increasing skill gap.

How does the presence of an AI assistant affect the collective attention of a team? We study 20 human teams of 3-4 individuals paired with one voice-only AI assistant during a challenging puzzle task. Teams are randomly assigned to an AI assistant with a human- or robotic-sounding voice that provides either helpful or misleading information about the task. Treating each individual AI interjection as a treatment intervention, we identify the causal effects of the AI on dynamic group processes involving language use. Our findings demonstrate that the AI significantly affects what teams discuss, how they discuss it, and the alignment of their mental models. Teams adopt AI-introduced language for both terms directly related to the task and for peripheral terms, even when they (a) recognize the unhelpful nature of the AI, (b) do not consider the AI a genuine team member, and (c) do not trust the AI. The process of language adaptation appears to be automatic, despite doubts about the AI's competence. The presence of an AI assistant significantly impacts team collective attention by modulating various aspects of shared cognition. This study contributes to human-AI teaming research by highlighting collective attention as a central mechanism through which AI systems in team settings influence team performance. Understanding this mechanism will help CSCW researchers design AI systems that enhance team collective intelligence by optimizing collective attention.

Large language models like GPT have proven widely successful on natural language understanding tasks based on written text documents. In this paper, we investigate an LLM's performance on recordings of a group oral communication task in which utterances are often truncated or not well-formed. We propose a new group task experiment involving a puzzle with several milestones that can be achieved in any order. We investigate methods for processing transcripts to detect if, when, and by whom a milestone has been completed. We demonstrate that iteratively prompting GPT with transcription chunks outperforms semantic similarity search methods using text embeddings, and further discuss the quality and randomness of GPT responses under different context window sizes.

Crowdsourcing has evolved as an organizational approach to distributed problem solving and innovation. As contests are embedded in online communities and evaluation rights are assigned to the crowd, community members face a tension: they find themselves exposed to both competitive motives to win the contest prize and collaborative participation motives in the community. The competitive motive suggests they may evaluate rivals strategically according to their self-interest, the collaborative motive suggests they may evaluate their peers truthfully according to mutual interest. Using field data from Threadless on 38 million peer evaluations of more than 150,000 submissions across 75,000 individuals over 10 years and two natural experiments to rule out alternative explanations, we answer the question of how community members resolve this tension. We show that as their skill level increases, they become increasingly competitive and shift from using self-promotion to sabotaging their closest competitors. However, we also find signs of collaborative behavior when high-skilled members show leniency toward those community members who do not directly threaten their chance of winning. We explain how the individual-level use of strategic evaluations translates into important organizational-level outcomes by affecting the community structure through individuals' long-term participation. While low-skill targets of sabotage are less likely to participate in future contests, high-skill targets are more likely. This suggests a feedback loop between competitive evaluation behavior and future participation. These findings have important implications for the literature on crowdsourcing design, and the evolution and sustainability of crowdsourcing communities., Comment: Currently in press

Even though research has repeatedly shown that non-cash incentives can be effective, cash incentives are the de facto standard in crowdsourcing contests. In this multi-study research, we quantify ideators' preferences for non-cash incentives and investigate how allowing ideators to self-select their preferred incentive -- offering ideators a choice between cash and non-cash incentives -- affects their creative performance. We further explore whether the market context of the organization hosting the contest -- social (non-profit) or monetary (for-profit) -- moderates incentive preferences and their effectiveness. We find that individuals exhibit heterogeneous incentive preferences and often prefer non-cash incentives, even in for-profit contexts. Offering ideators a choice of incentives can enhance creative performance. Market context moderates the effect of incentives, such that ideators who receive non-cash incentives in for-profit contexts tend to exert less effort. We show that heterogeneity of ideators' preferences (and the ability to satisfy diverse preferences with suitably diverse incentive options) is a critical boundary condition to realizing benefits from offering ideators a choice of incentives. We provide managers with guidance to design effective incentives by improving incentive-preference fit for ideators., Comment: Journal of Management Information Systems, forthcoming 2024

Team interactions are often multisensory, requiring members to pick up on verbal, visual, spatial and body language cues. Multimodal research, research that captures multiple modes of communication such as audio and visual signals, is therefore integral to understanding these multisensory group communication processes. This type of research has gained traction in biomedical engineering and neuroscience, but it is unclear the extent to which communication and management researchers conduct multimodal research. Our study finds that despite its' utility, multimodal research is underutilized in the communication and management literature's. This paper then covers introductory guidelines for creating new multimodal research including considerations for sensors, data integration and ethical considerations., Comment: 27 pages, 3 figures

Voice assistants are increasingly prevalent, from personal devices to team environments. This study explores how voice type and contribution quality influence human-agent team performance and perceptions of anthropomorphism, animacy, intelligence, and trustworthiness. By manipulating both, we reveal mechanisms of perception and clarify ambiguity in previous work. Our results show that the human resemblance of a voice assistant's voice negatively interacts with the helpfulness of an agent's contribution to flip its effect on perceived anthropomorphism and perceived animacy. This means human teammates interpret the agent's contributions differently depending on its voice. Our study found no significant effect of voice on perceived intelligence, trustworthiness, or team performance. We find differences in these measures are caused by manipulating the helpfulness of an agent. These findings suggest that function matters more than form when designing agents for high-performing human-agent teams, but controlling perceptions of anthropomorphism and animacy can be unpredictable even with high human resemblance., Comment: 12 pages, 4 figures

Online innovation communities allow for a search for novel solutions within a design space bounded by constraints. Past research has focused on the effect of creative constraints on individual projects, but less is known about how constraints affect learning from repeated design submissions and the effect of the technical constraints that are integral to online platforms. How do creative versus technical constraints affect individual learning in exploring a design space in online communities? We analyzed ten years of data from an online innovation community that crowdsourced 136,989 design submissions from 33,813 individuals. We leveraged data from two types of design contests-creatively constrained and unconstrained-running in parallel on the platform, and we evaluated a natural experiment where a platform change reduced technical constraints. We find that creative constraints lead to high rates of learning only if technical constraints are sufficiently relaxed. Our findings have implications for the management of creative design work and the downstream effects of the technical constraints of the information systems that support online innovation communities.

Existing literature predominantly focuses on how freelancers individually complete tasks and projects. Our study examines freelancers' willingness to work collaboratively. We report results from a survey of 122 freelancers on a leading online labor market platform (Upwork) and examine freelancers' preferences for collaboration and explore several antecedents of cooperative behaviors. We then test if actual cooperative behavior matches with freelancers' stated preferences through an incentivized social dilemma experiment. We find that respondents cooperate at a higher rate (85%) than reported in previous comparable studies (between 50-75%). This high rate of cooperation may be explained by an ingroup bias. Using a sequential mediation model we demonstrate the importance of a sense of shared expectations and accountability for cooperation. We contribute to a better understanding of the potential for collaborative work on online labor market platforms by assessing if and what social factors and collective culture exist among freelancers. We discuss the implications of our results for platform designers by highlighting the importance of platform features that promote shared expectations and improve accountability. Overall, contrary to existing literature and predictions, our results suggest that freelancers in our sample display traits that are more consistent with belonging to a coherent group with a shared collective culture, rather than being anonymous actors in a transaction-based market.

We study the coevolution of network structure and signaling behavior. We model agents who can preferentially associate with others in a dynamic network while they also learn to play a simple sender-receiver game. We have four major findings. First, signaling interactions in dynamic networks are sufficient to cause the endogenous formation of distinct signaling groups, even in an initially homogeneous population. Second, dynamic networks allow the emergence of novel {\em hybrid} signaling groups that do not converge on a single common signaling system but are instead composed of different yet complementary signaling strategies. We show that the presence of these hybrid groups promotes stable diversity in signaling among other groups in the population. Third, we find important distinctions in information processing capacity of different groups: hybrid groups diffuse information more quickly initially but at the cost of taking longer to reach all group members. Fourth, our findings pertain to all common interest signaling games, are robust across many parameters, and mitigate known problems of inefficient communication.

We develop a network of Bayesian agents that collectively model the mental states of teammates from the observed communication. Using a generative computational approach to cognition, we make two contributions. First, we show that our agent could generate interventions that improve the collective intelligence of a human-AI team beyond what humans alone would achieve. Second, we develop a real-time measure of human's theory of mind ability and test theories about human cognition. We use data collected from an online experiment in which 145 individuals in 29 human-only teams of five communicate through a chat-based system to solve a cognitive task. We find that humans (a) struggle to fully integrate information from teammates into their decisions, especially when communication load is high, and (b) have cognitive biases which lead them to underweight certain useful, but ambiguous, information. Our theory of mind ability measure predicts both individual- and team-level performance. Observing teams' first 25% of messages explains about 8% of the variation in final team performance, a 170% improvement compared to the current state of the art., Comment: 9 pages, Accepted at AAAI 2023

Can egalitarian norms or conventions survive the presence of dominant individuals who are ensured of victory in conflicts? We investigate the interaction of power asymmetry and partner choice in games of conflict over a contested resource. We introduce three models to study the emergence and resilience of cooperation among unequals when interaction is random, when individuals can choose their partners, and where power asymmetries dynamically depend on accumulated payoffs. We find that the ability to avoid bullies with higher competitive ability afforded by partner choice mostly restores cooperative conventions and that the competitive hierarchy never forms. Partner choice counteracts the hyper dominance of bullies who are isolated in the network and eliminates the need for others to coordinate in a coalition. When competitive ability dynamically depends on cumulative payoffs, complex cycles of coupled network-strategy-rank changes emerge. Effective collaborators gain popularity (and thus power), adopt aggressive behavior, get isolated, and ultimately lose power. Neither the network nor behavior converge to a stable equilibrium. Despite the instability of power dynamics, the cooperative convention in the population remains stable overall and long-term inequality is completely eliminated. The interaction between partner choice and dynamic power asymmetry is crucial for these results: without partner choice, bullies cannot be isolated, and without dynamic power asymmetry, bullies do not lose their power even when isolated. We analytically identify a single critical point that marks a phase transition in all three iterations of our models. This critical point is where the first individual breaks from the convention and cycles start to emerge.

Bayesian optimal experiments that maximize the information gained from collected data are critical to efficiently identify behavioral models. We extend a seminal method for designing Bayesian optimal experiments by introducing two computational improvements that make the procedure tractable: (1) a search algorithm from artificial intelligence that efficiently explores the space of possible design parameters, and (2) a sampling procedure which evaluates each design parameter combination more efficiently. We apply our procedure to a game of imperfect information to evaluate and quantify the computational improvements. We then collect data across five different experimental designs to compare the ability of the optimal experimental design to discriminate among competing behavioral models against the experimental designs chosen by a "wisdom of experts" prediction experiment. We find that data from the experiment suggested by the optimal design approach requires significantly less data to distinguish behavioral models (i.e., test hypotheses) than data from the experiment suggested by experts. Substantively, we find that reinforcement learning best explains human decision-making in the imperfect information game and that behavior is not adequately described by the Bayesian Nash equilibrium. Our procedure is general and computationally efficient and can be applied to dynamically optimize online experiments.

Novelty is a key ingredient of innovation but quantifying it is difficult. This is especially true for visual work like graphic design. Using designs shared on an online social network of professional digital designers, we measure visual novelty using statistical learning methods to compare an images features with those of images that have been created before. We then relate social network position to the novelty of the designers images. We find that on this professional platform, users with dense local networks tend to produce more novel but generally less successful images, with important exceptions. Namely, users making novel images while embedded in cohesive local networks are more successful., Comment: accepted to 10th International ACM Web Science Conference, 2018, May 27-30, Amsterdam, The Netherlands, 11 pages, 6 figures, 60 references

An important way to resolve games of conflict (snowdrift, hawk-dove, chicken) involves adopting a convention: a correlated equilibrium that avoids any conflict between aggressive strategies. Dynamic networks allow individuals to resolve conflict via their network connections rather than changing their strategy. Exploring how behavioral strategies coevolve with social networks reveals new dynamics that can help explain the origins and robustness of conventions. Here we model the emergence of conventions as correlated equilibria in dynamic networks. Our results show that networks have the tendency to break the symmetry between the two conventional solutions in a strongly biased way. Rather than the correlated equilibrium associated with ownership norms (play aggressive at home, not away), we usually see the opposite host-guest norm (play aggressive away, not at home) evolve on dynamic networks, a phenomenon common to human interaction. We also show that learning to avoid conflict can produce realistic network structures in a way different than preferential attachment models.

We report the findings of a month-long online competition in which participants developed algorithms for augmenting the digital version of patent documents published by the United States Patent and Trademark Office (USPTO). The goal was to detect figures and part labels in U.S. patent drawing pages. The challenge drew 232 teams of two, of which 70 teams (30%) submitted solutions. Collectively, teams submitted 1,797 solutions that were compiled on the competition servers. Participants reported spending an average of 63 hours developing their solutions, resulting in a total of 5,591 hours of development time. A manually labeled dataset of 306 patents was used for training, online system tests, and evaluation. The design and performance of the top-5 systems are presented, along with a system developed after the competition which illustrates that winning teams produced near state-of-the-art results under strict time and computation constraints. For the 1st place system, the harmonic mean of recall and precision (f-measure) was 88.57% for figure region detection, 78.81% for figure regions with correctly recognized figure titles, and 70.98% for part label detection and character recognition. Data and software from the competition are available through the online UCI Machine Learning repository to inspire follow-on work by the image processing community.

We employ a 2x3 factorial experiment to study two central factors in the design of prediction markets (PMs) for idea evaluation: the overall design of the PM, and the elasticity of market prices set by a market maker. The results show that 'multi-market designs' on which each contract is traded on a separate PM lead to significantly higher trading performance than 'single-markets' that handle all contracts one on PM. Price elasticity has no direct effect on trading performance, but a significant interaction effect with market design implies that the performance difference between the market designs is highest in settings of moderate price elasticity. We contribute to the emerging research stream of PM design through an unprecedented experiment which compares current market designs., Comment: Presented at Collective Intelligence conference, 2012 (arXiv:1204.2991)