Measuring Group Personality With Swarm AI
Measuring Group Personality with Swarm AIGregg WillcoxUnanimous AISan Francisco, CA, USAgregg@unanimous.aiDavid AskayCalifornia Polytechnic StateUniversitySan Luis Obispo, CAdaskay@calpoly.eduLouis RosenbergUnanimous AISan Francisco, CA, USAlouis@unanimous.aiBryan KwongUnanimous AISan Francisco, CA, USAbryan.kwong@unanimous.aiRichard LiuUniversity of Califoria, BerkeleySan Francisco, CA, USArichard.y.liu@berkeley.eduAbstract—The aggregation of individual personality tests topredict team performance is widely accepted in managementtheory but has significant limitations: the isolated nature ofindividual personality surveys fails to capture much of the teamdynamics that drive real-world team performance. ArtificialSwarm Intelligence (ASI), a technology that enables networkedteams to think together in real-time and answer questions as aunified system, promises a solution to these limitations by enablingteams to take personality tests together, whereby the team usesASI to converge upon answers that best represent the group’sdisposition. In the present study, the group personality of 94 smallteams was assessed by having teams take a standard Big FiveInventory (BFI) test both as individuals, and as a real-time systemenabled by an ASI technology known as Swarm AI. The predictiveaccuracy of each personality assessment method was assessed bycorrelating the BFI personality traits to a range of real-worldperformance metrics. The results showed that assessments ofpersonality generated using Swarm AI were far more predictiveof team performance than the traditional survey-based method,showing a significant improvement in correlation with at least25% of performance metrics, and in no case showing a significantdecrease in predictive performance. This suggests that Swarm AItechnology may be used as a highly effective team personalityassessment tool that more accurately predicts future teamperformance than traditional survey approaches.Keywords—Group Personality, BFI, Group Performance,Swarm Intelligence, Artificial Swarm Intelligence, CollectiveIntelligence, Group Consensus.I. INTRODUCTIONMost businesses strive to build high performing teamswherein the combination of skills, personality traits, and workhabits of team-members drives effective execution towardsorganizational goals. One commonly used technique forpredicting whether a team will be high performing is toadminister a personality test to each individual member,aggregate the team’s test results, and use those aggregatedresults to forecast whether the combined team is likely to workwell together [1-4]. Prior research has shown a correlationbetween aggregated results on personality tests and resultingteam performance [5]. The current study reviews these priormethods and explores whether improved forecasts of teamThis work was partially funded by NSF Grant #1840937.IEEE Conference on Transdisciplinary AI (Trans AI) 2019 20XX IEEELynn MetcalfCalifornia Polytechnic StateUniversitySan Luis Obispo, CAlmetcalf@calpoly.eduperformance can be attained using Artificial SwarmIntelligence—a unique AI technology that aims to moreaccurately assess team personality.As further background, multilevel approaches toinvestigating organizational phenomena are critical, yetunderstudied [6]. Multilevel research often involves aggregatingindividual-level data (e.g., the personalities of individualscomprising a team) to measure group-level constructs (e.g., teamperformance). Typically, individual-level data are aggregated tomeasure group-level phenomena in one of four ways: bycalculating a mean score of individual measures, by computingthe highest (maximum) individual trait score, by computing thelowest (minimum) individual trait score, or by looking at thevariance of individual trait scores within a team [7, 8]. Each ofthese four methods of aggregation have been found to captureunique information about the group [7]. For example,conscientiousness, an individual-level personality trait, isassociated with high levels of organization and attention todetail. Averaging the conscientiousness scores of individualscomprising a team assumes that the amount of conscientiousnesspossessed by each individual team member contributes to thecollective pool of conscientiousness available to the team,regardless of how the trait is distributed among team conscientiousness brings the rest of the group down on average,making the group minimum score the most appropriate way toaggregate individual scores. While each of these methods ofaggregation provide unique insights, researchers continue toquestion the efficacy of using individual-level measures toassess group-level traits or phenomena.An alternative aggregation method, group consensus, offersthe potential to improve the accuracy of personality assessments.A consensus approach, whereby all members consider eachquestion on an assessment and jointly agree on a collectivescore, has been advocated because it better captures theunderlying and unique group dynamics present in teams [9, 10].For example, a study of MBA students found that measuringteam efficacy through a consensus approach was a betterpredictor of group performance than when measured throughaggregated individual-level constructs [11]. While the
consensus method offers a potentially superior way ofaggregating individual-level constructs, it suffers fromdrawbacks. Specifically, the context of a group discussionallows for social influence to silence some members or toencourage conformity. Additionally, achieving consensus iscostly in terms of time and logistical organization ofparticipants. For these reasons, and despite the potential ofgroup-level consensus personality measurement and calls tomove away from the aggregation of individual-level data [12],researchers seldom use group-level consensus ratings.Advances in networking technology and artificialintelligence have led to the development of Artificial SwarmIntelligence (ASI) systems that provides a way for groups ofhumans to quickly reach a consensus in a way that overcomesthese limitations. ASI has been found to significantly amplifydecision-making accuracy in human groups [13 - 19]. Indeed,groups can achieve consensus in less than 60 seconds, while alsolimiting social influence from group members throughanonymous deliberation that capture group dynamics. ASIpresents a promising method that answers the call for researchusing consensus-based aggregation approaches. Specifically, wefocus on the potential of using ASI as a method of administeringand composing group-level personality assessments, and inpredicting team performance based on these personalityassessments.II. FOUNDATIONS OF SWARM INTELLIGENCEIn the natural world, Swarm Intelligence (SI) enables socialorganisms to aggregate their collective insights rapidly and toconverge in synchrony on optimal decisions by forming realtime closed-loop systems. Swarm Intelligence has been deeplystudied across many social species, from schools of fish andflocks of birds to swarms of honey bees and even slime molds.Unlike birds, bees and fish, humans have not evolved the naturalability to form real-time swarms, as we lack the innatemechanisms used by other species to form closed-loop systems.Schooling fish detect vibrations in the water around them.Flocking birds detect high-speed motions propagating throughthe group formation. Swarming bees generate complex bodyvibrations called a “waggle dance” that encode assessmentinformation. To enable networked human groups to form similarclosed-loop systems, a cloud-based platform called “swarm.ai”was developed. It enables human groups, connected fromremote locations, to make collective predictions, decisions, andassessments by working together as closed-loop swarms.When using the swarm.ai platform, networked human teamsanswer questions by collaboratively moving a graphical pointerto select from a set of answer options. Each participant providestheir individual input by manipulating a graphical magnet witha mouse, touchpad, or touchscreen. By adjusting the positionand orientation of their magnet with respect to the moving puck,participants express their real-time intent. The input from eachuser is not a discrete vote, but a stream of vectors that variesfreely over time. Because all members of the group can adjusttheir intent continuously in real-time, the swarm explores thedecision-space, not based on the input of any individualmember, but based on the emergent dynamics of the full system.This enables a complex behavioral interaction among allmembers of the population, empowering the group tocollectively consider the options and synchronously convergeon the most agreeable solution.Fig. 1. Architecture of the swarm.ai platform with graphical client andcloud-based AI engineIt is important to note that participants not only vary thedirection of their intent but also modulate the magnitude of theirintent by adjusting the distance between their magnets and thepointer, which is commonly represented as a graphical puck.Because the graphical puck is in continuous motion across thedecision-space, users need to move their magnets continually sothat they stay close to the puck’s rim. This is significant for itrequires that all participants, regardless of group size orcomposition, be engaged continuously throughout the decisionprocess, evaluating and re-evaluating their intent in real-time. Ifa participant stops adjusting their magnet with respect to thechanging position of the puck, the distance grows and theparticipant’s influence on the group’s decision wanes.Thus, like bees vibrate their bodies to express sentiment in abiological swarm or neurons fire to express conviction levelswithin a biological neural-network, the participants in anartificial swarm must continuously update and express theirchanging preferences during the decision process or lose theirinfluence over the collective outcome. This is generally referredto as a “leaky integrator” structure and common to both swarmbased and neuron-based systems. In addition, intelligencealgorithms monitor the behaviors of swarm members in realtime, inferring their relative conviction based on their actionsand interactions over time. This reveals a range of behavioralcharacteristics within the population and weights theircontributions accordingly.Just as ASI provides an effective way for groups to reach aconsensus around decision-making, it is a promising method forreaching a consensus around responses to psychometricassessments like a personality test. Through ASI, a question canbe answered in less than 60 seconds, participants are anonymousand less subject to dysfunctional social influence, and consensusis achieved through interactions as participants deliberatevisually through the interface.III. METHODTo assess the ability of ASI technology to function as anaccurate assessment tool of team personality, a large study wasconducted across a set of 94 working groups (i.e. teams), eachcomprising 3 to 6 members. Each of these teams were engagedin a 10-week group project. In total, 384 human subjects
participated in this study. All were college students enrolled inbusiness, communication studies, or engineering courses, forwhich a team project was a significant component. Participantsfirst completed the personality assessment individually bythemselves, then they completed a personality test collectivelyas a group using ASI. The individual results were used tocompose group-level team personality through typicalaggregation approaches (mean, max, min, and variance). Theresults from the ASI represent a consensus-based teampersonality. Finally, at the conclusion of the group project, anoutcome survey was administered individually to participants tomeasure group outcomes (e.g., performance).The Big Five Inventory (BFI) assessment [20] wasused to measure personality for both individual and ASIconditions. Qualitics was used to administer the assessment toindividuals and the Swarm software platform was used formeasuring ASI consensus. The BFI test is commonly used inliterature and industry as a personality assessment tool, and awide body of research has validated that individual and groupscores on this test are correlated with performance on real-worldtasks [21-28]. The questions that were included in Individualand Swarm versions of the BFI test are listed in Appendix A.When answering the BFI individually, participants were askedabout their own personalities (e.g., Are you talkative?). Whengroup were asked questions through ASI, the referent shifted tothe group-level (e.g., Is this group talkative?).The swarms were attended by 297 (77.3%) participants, andany group in which fewer than 2 individuals participated in theswarm was eliminated from the dataset. The swarms had oneminute to answer each question, and if they failed to reach aconsensus in that time (referred to as a Brain Freeze), thequestion was repeated only once. No swarm experienced a brainfreeze during the second round.The individual personality assessments were aggregated inpost-processing into a group personality assessment using eachof four different methods: (1) average score, (2) minimum score,(3) maximum score, and (4) the variance of individual scores. Inthis way, the traditional method for assessing group personality(i.e. statistically aggregating individual BFI scores) and a newmethod for assessing group personality (i.e. enabling teams totake the BFI test together as a unified swarm intelligence) couldbe directly compared. Satisfaction--the degree to which group members arepleased with group members and the team [33] Viability--the degree to which the group desires to worktogether again in the future [34] Transactive memory--the degree to which groupmembers know about the skills, emotions, and tasks ofother group members [35] Team Effectiveness—a self-rating of how well the groupaccomplished it’s task [36]Prior studies have established connections between grouplevel personality and these performance outcome variables. Foreach group, the aggregated scores (average, min, max, variance)and the swarm scores for the BFI were correlated with the sixperformance indicators with Pearson’s correlation coefficient.The resulting R2 values were compared and used for statisticaltests in analysis.IV. ANALYSISThe correlation between each personality assessmentmethod and the performance of each team was calculated usinga linear regression. The Pearson coefficient of determination (R 2value) between each BFI Dimension and performance metricwas calculated for each of the five group personalitymeasurement methods. The study originally measured 17performance metrics, which have been averaged by categorydown to 9 metrics for ease of viewing.The R2 values for each personality measurement method areshown in Appendix A, and the Survey Average vs SwarmCorrelations with the performance metrics are shown in figure 2below. Immediately, these plots show that, on average, swarmbased assessments of group personality have a higher correlationwith team performance than the survey-based assessments ofgroup personality.Several team outcome variables were measured at theconclusion of the group project, which occurred several days orweeks after the swarm assessment. Several performance relatedself-assessments were administered to each team member: Cohesiveness--degree of bonding towards the team, teammembers, and the task [29]. Conflict--the degree of relational, task, and processbased conflict experienced in the group [30] Psychological Safety--the degree to which groupmembers feel like they can be vulnerable and speak upwith other group members [31] Potency--general perception of the group’s confidenceand capability [32]Fig. 2. Heat map of Pearson R2 values between Swarm or Survey AveragePersonality Measurement and Performance Metrics
A bootstrapped significance test was performed to measurewhether the swarm could have outperformed the survey methodsin this test due to random chance alone. In this process, theobserved groups (including the personality assessment by eachmethod and performance metrics) were randomly resampledwith replacement 1000 times, and the 90% confidence intervalof the difference in R2 values between the survey and swarmassessments of group personality was calculated. This processwas repeated for each group performance metric and eachsurveying method.A table of confidence intervals generated using this approachis shown in Appendix B, with the cells in which the swarm’sassessment was found to correlate with the performance metricsignificantly more than the survey’s assessment highlighted inyellow, and the cells in which the reverse is true highlighted ingreen. Table 1 below gives an overview of this statisticalsignificance test: out of the 85 comparisons made between eachsurvey assessment method and the swarm, the swarmsignificantly outperformed the survey in at least 25.9% of cases,while the survey never significantly outperformed the swarm.TABLE I.SUMMARY OF BOOTSTRAPPED CORRELATION DIFFERENCESBETWEEN SWARM AND SURVEY ASSESSMENTS OF TEAM PERSONALITYPercentage of Comparisons Where s theSurveySurvey SignificantlyOutperforms theSwarmAverage R2IncreaseAverage30 (35.3%)0 (0%)0.0654Maximum25 (29.4%)0 (0%)0.0687Minimum22 (25.9%)0 (0%)0.0484Variance24 (28.2%)0 (0%)0.0684V. CONCLUSIONThe group personality of 94 small teams was assessed byasking the teams to respond to a standard set of 45 Big FiveInventory questions using both traditional surveys of individualpersonality and a real-time collaboration interface (Swarm AI)to establish a group consensus of the team’s own personality.Four different multilevel approaches to aggregating the teammember’s answers to the survey BFI questions were studied: theaverage, variance, minimum, and maximum of the team’sanswers.The performance of the surveying methods wascompared to the swarming methods by correlating the BFIdimensions, as calculated by each method, to various metrics ofthe team’s self-reported performance. The swarming methodssignificantly outperformed each of the survey aggregationmethods at predicting a wide range of performance metrics (atleast 25.9%, n 85), and were never significantly outperformedby the survey aggregation methods.This result suggests that ASI can be used to evaluate teampersonality, and predict team performance, more accurately thantraditional individual surveying methods. There are severaladvantages to this approach. First, it overcomes concerns aboutboth time and social influence of the consensus-based approachto aggregation. The average time to reach a consensus was 18.8seconds. The anonymity provided by the platform enablesparticipants to interact and deliberate visually, while protectingthe identities of team members. Second, the analysis reveals thatthe BFI results of the ASI-based group consensus was a strongerpredictor of important group outcomes, such as performance,viability, and cohesion. In doing so, it provides a response tocalls for consensus-based aggregation and support for consensusbeing a superior method of aggregating group-level variables[9]. Future research is needed to replicate and extend thesefindings to new contexts and different group-level variables.This research was limited by the availability andparticipation rate of participants, as 72.9% of participants did nottake the pre-swarm survey, and 77.3% did not participate in theswarm. This research also did not investigate whether thepresentation of the question itself contributed to the highersuccess rate of the swarm in predicting team performance, sinceparticipants were asked directly about the team’s personality inthe swarm, but were asked about their own personality in thesurveys.ACKNOWLEDGMENTThanks to Unanimous AI for the use of the Swarm platformfor this ongoing work. Thanks also to Erick Harris for his effortsin coordinating
performance metrics. The results showed that assessments of personality generated using Swarm AI were far more predictive of team performance than the traditional survey-based method, showing a significant improvement in correlation with at least 25% of performance metrics, and in no case showing a significant
By default, Docker Swarm is disabled, so to run Docker in swarm mode, you will need to either join an existing cluster or create a new swarm. To create a new swarm and activate it in your system, you use the swarm init command shown here: docker swarm init This will create a new single-node swarm cluster on the node you are currently working on.
to other personality disorders such as antisocial personality disorder. Ogloff (2005) distinguishes psychopathy from antisocial personality disorder due to the emphasis on affective and personality rather than mostly behavioral elements of antisocial personality disorder. Besides antisocial personality disorder, there are other DSM-IV personality
Swarm intelligence Swarm intelligence collection of intelligent techniques inspired by the collective behavior of some self -organizing systems The name was coined in 1989 by Gerardo Beni and Jing Wang in the context of control systems for robots The swarm intelligence techniques use sets of agents characterized by:
First, install Docker CE on the machine (refer to the prior sections on installing Docker CE). Initialize the swarm: Note: Set --advertise-addr to an address that other nodes in the swarm will see this node as. docker swarm init --advertise-addr advertise address We can find information about the current state of the swarm using docker info.
2.5 Quick start tutorial This short tutorial is designed to get the user using SWARM productively in just a few minutes. Note that SWARM is set to open in a window that will fit on a 1024x768 screen. SWARM is significantly more usuable at higher resolutions. Helicorders display a large
Understanding The Supporter Personality Chapter 5: Understanding The Promoting/Supporter Personality Chapter 6: Understanding The Promoter/Controller Personality Chapter 7: Understanding The Controller/Analyzer Personality Chapter 8 : Understanding The Analyzer/Supporter Personality Chapter 9: Understanding The Centric Personality Wrapping Up
the light of sound theoretical systems of personality. 3. To acquaint the students with the applications of personality theories in different walks of life. Unit Contents No. of Lectures Unit-I INTRODUCTION TO PERSONALITY 1.1.Definitions and nature of personality 1.2.Characteristics of good personality theory and Evaluation of personality theory
www.2id.korea.army.mil 2 Indianhead August 13, 2010 “Jeju Island, it takes about a half day to travel, so on a long weekend you can spend three full days touring, exploring, and enjoying yourself.” Pfc. Reginald Garnett HHC, 1-72th Armor OpiniOn “Gangneung is a great place to visit. It has a beautiful beach, Gyeongpo Beach, where many .
