Perspectives on team dynamics: Meta learning and systems intelligence

SystemsResearch andBehavioral ScienceSyst. Res.25, 757^767 (2008)Published online 27 August 2008 inWiley InterScience(www.interscience.wiley.com)DOI:10.1002/sres.905

& ResearchPaper

Perspectives on Team Dynamics: MetaLearning and Systems Intelligence

Jukka Luoma*, Raimo P. Hamalainen and Esa Saarinen

Systems Analysis Laboratory, Helsinki University of Technology, Finland

Losada observed management teams develop their annual strategic plans in a labdesigned for studying team behaviour. Based on these findings he developed a dynamicalmodel of team interaction and introduced the concept of meta learning (ML) whichrepresents the ability of a team to avoid undesirable attractors. This paper analyses thedynamic model in more detail and discusses the relationship between ML and the newconcept of systems intelligence (SI) introduced by Saarinen and Hamalainen. We proposethat the ML ability of a team clearly represents a systems intelligent competence. Losada’smathematical model predicts interesting dynamic phenomena in team interaction. How-ever, our analysis shows how themodel also produces strange and previously unreportedbehaviour under certain conditions. Thus, the predictive validity of the model alsobecomes problematic. It remains unclear whether the model behaviour can be said tobe in satisfactory accordance with the observations of team interaction. Copyright# 2008John Wiley & Sons, Ltd.

Keywords team interaction; meta learning; systems intelligence; chaotic dynamics; Lorenzattractor

INTRODUCTION

The perspective taken in systems intelligence1

(SI) (Saarinen and Hamalainen, 2004; Hamalai-nen and Saarinen, 2006, 2007) is the acknowl-edgement that we are always embedded in and apart of systems involving interaction and feed-back which we cannot escape but in which wecan take intelligent action. The study of SI deals

with the behavioural intelligence of humanagents in systemic environments. SI looks forefficient ways for an agent to change his/her ownbehaviour in order to influence the behaviour of asystem in different environments.

In this paper, a related concept of metalearning (ML) is discussed. The concept, asintroduced by Losada (1999), refers to theconversational, or micro-behavioural compe-tence of teams. Teams with ML ability are ableto avoid and dissolve undesirable lock-ins intheir conversational interactions. Correlates forteam performance are sought in such teams’micro-behavioural competencies and, thus,

*Correspondence to: Jukka Luoma, Systems Analysis Laboratory,Helsinki University of Technology, P.O. Box 1100, 02150 HUT,Finland.E-mail: [email protected]://www.systemsintelligence.hut.fi [accessed 17 April 2007].

Copyright # 2008 John Wiley & Sons, Ltd.

viewed as potential contributors to team per-formance. We see such competence as a mani-festation of what we frame as SI. SI highlights theongoing opportunity to generate actions, some-times micro-behavioural, that can contribute tosome systemic improvement and can thus beregarded as intelligent, that is, as ‘embedded inaction’ intelligence.

We view the concept of a system as a useful onein understanding human action in dynamicsettings. Today, many organizational studiesdescribe organizational phenomena by drawinganalogies to concepts which originate in the fieldof mathematical modelling of dynamical sys-tems. The use of systems studies ranges fromquantitative to qualitative modelling; fromnumerical modelling of organizational phenom-ena (e.g. Sterman, 2000, 2002), to metapho-rical and analogical use of such models (e.g.Senge, 1990; Senge et al., 1994) to drawinganalogies from simulation studies of complexsystems (e.g. Stacey, 1995, 2001, 2003; Morel andRamanujam, 1999) to thinking about organiz-ations as systems (see e.g. Jackson, 2000, pp.108–127; Stacey, 2007). Our emphasis is on anagent’s own interdependence, that is, his/herinfluence on and influence by, systems. Webelieve that a systemic perspective on humanaction can bring valuable contributions to thestudy of organizational behaviour as it movestowards understanding human interaction,such as team meetings, as something thatpeople together as a system generate but—at thesame time—people’s choices are with respect toand constituent of those systems. One convenientway of approaching social interaction fromsuch a systemic perspective is mathematicalmodelling.

The literature on the modelling of socialinteraction is extensive. It dates back to the earlywork of Simon (1952) who illustrated howmathematical modelling could be used in‘clarifying of concepts’ of a theory, and in‘derivation of new propositions’. Here, we onlylist some recent studies, which in our view haveconnections to the topic of this paper. Axelrod(1984) used game theoretical experiments toshow that under suitable conditions cooperationcan emerge without a central authority. Young

(2001) built a dynamical model of conformitybased on Blume’s (1993) work on strategicinteraction. Gintis et al. (2005) used gametheoretical models and experimental gametheory to study decision-making in social inter-action and the evolution of cooperation. Collinsand Hanneman (1998) suggested a commonframework for modelling the theory of inter-action rituals (Collins, 1981, 2004). Gottman et al.(2002a, 2002b) developed nonlinear differenceequations to model marital interaction as well asto design of counselling interventions forunhappy couples. Losada (1999; Losada andHeaphy, 2004) and his associates used nonlineardifferential equations to describe team beha-viour. Losada (1999); Losada and Heaphy, 2004)pay particular attention to the potentially chaoticbehaviour of the model, and to the interpretationof that behaviour. The papers discuss the modelbehaviour in connection with the concept of ML,the ability to avoid and dissolve undesiredlock-ins in teams’ verbal communication. Com-mon to all of these studies is that they all usemodelling as a tool to understand phenomenarelated to the dynamics of human interaction.This paper focuses on Losada’s research on teaminteraction and in particular, on the character-istics of the dynamical model proposed by him.

META LEARNING AND SYSTEMSINTELLIGENCE

Losada (1999, p. 190) defines ML ‘as the ability ofa team to dissolve attractors that close possibi-lities and evolve attractors that open possibilitiesfor effective action’. Attractors of the first type aresomething that ‘trap individuals and organiz-ations into rigid patterns of thinking thatinevitably lead to limiting behaviour’. The‘attractor’ that Losada refers to as closingpossibilities is used as a metaphor for beha-vioural patterns that ‘teams get stuck’ with.

The concept of ML is presented in connectionwith a dynamical model of team interaction. Thedevelopment of the model is said to reflectobservations Losada (1999) made of team inter-action and team performance. Losada collectedthe data by observing 60 strategic business unit

Copyright � 2008 JohnWiley & Sons,Ltd. Syst. Res.25, 757^767 (2008)DOI:10.1002/sres

758 Jukka Luoma et al.

RESEARCHPAPER Syst. Res.

management teams of a large informationprocessing corporation in sessions where theywere developing their annual strategic plans.These sessions were held in a lab designed forteam research. The measurements were codedusing bipolar scales which were positivity–negativity, inquiry–advocacy and other–self.The coding was based on the observed verbalcommunication of the teams. The interactionpatterns and, in particular, the interrelatedness ofthe participants’ behaviours were studied byanalysing the time series of the observations.

The team interaction model (Losada, 1999) isused to analyse the dynamics of the three teaminteraction variables, Losada recorded. It pre-dicts what types of dynamics are possible. Thesecan be of three types: ‘point attractor, limit cycleand complexor’ dynamics (Losada and Heaphy,2004) The authors argue that point attractordynamics correspond to low performance, limitcycle dynamics to medium performance andcomplexor (chaotic) dynamics to high perform-ance of teams, respectively. In terms of the MLability, the model presents one way of describingsome behavioural patterns a team might getstuck with.

Losada (1999) also reports that his research hashad practical implications for his personalconsulting work. His strategy for organizationalinterventions is based on identifying whichattractors trap teams into ‘low performancepatterns’ and designing interventions that dis-solve these attractors and allow them to evolvenew ones that open possibilities. The approach issimilar to the way Gottman et al. (2002a, 2002b)design and implement martial counselling inter-ventions. There are also other interestingsystemic approaches linking micro-behaviourswith system functioning and systemic improve-ment. These include the research on parent–infant interactions by Beebe and her collaborators(Beebe and Lachmann, 1998; Beebe et al., 2000)and Fogel and Garvey (2007) A similar view ofpsychoanalytic treatment has been proposed bythe BostonChangeGroup (2002). The SI perspect-ive shares a similar motivation for looking intothe micro-behaviours of individuals in socialencounters. SI seeks for ways in which people arealready connecting to and interacting with one

another on a micro-behavioural level that mayalso contribute to macro-level systemic improve-ments. Presumably, the track record of a team isreflected in the situations in which team mem-bers encounter each other—how much theyexpress encouragement and show interest, forexample. We could also hypothesize that teammembers’ experience of the quality of suchinteractions feeds back to team performance.Team interactions, then, might be viewed as anopportunity for betterment through payingattention to one’s own part in the quality andcharacteristics of such interactions.

Losada’s setting is of interest to SI research, asit views conversation among team members as acontext wherein systemic improvement canpotentially be evoked. If there is a valid modelof team interaction, it could serve as a valuabletool that, with the use of simulations and analysisof its structures, can contribute to understandingthe conversational and micro-behavioural pat-terns of teams and identifying possible ways ofimproving such dynamics.

BACKGROUND OF THE MODEL

The verbal communication of the teams wascoded using the three bipolar scales, positivity–negativity, inquiry–advocacy and other–self asfollows.

A speech act was coded as inquiry if it involved aquestion aimed at exploring and examining aposition and as advocacy if it involved arguing infavour of the speaker’s viewpoint. A speech act wascoded as self if it referred to the person speaking orto the group present at the lab or to the company theperson speaking belonged, and it was coded as otherif the reference was to a person or group outside thelab and not part of the company to which the personspeaking belonged. A speech act was coded aspositive if the person speaking showed support,encouragement or appreciation, and it was coded asnegative if the person speaking showed disapproval,sarcasm or cynicism (Losada 1999).

The coded speech acts ‘were aggregated in1-min intervals’ to generate the time series ofobservations. The teams’ time series were


PerspectivesonTeamDynamics 759

Syst. Res. RESEARCHPAPER

analysed and, for each team, a parameter calledthe degree of connectivity was estimated bycounting ‘the number of cross-correlations[between the participants’ time series] significantat the 0.001 level or better’ (Losada, 1999). It issaid that this measure is indicative of ‘a processof mutual influence’ (Losada, 1999; Losada andHeaphy, 2004). These cross-correlations can beunderstood to reflect the influence that theinteracting parties have on each other. However,because the degree of connectivity as reported inthe original papers is the sum of all significantcross-correlations, some interesting informationregarding the degree of influence betweenindividual team members is unavailable to thereader of the original papers.

The observed teams were classified accordingto three performance measures as high, mediumand low performance teams. The performancelevel was indicated by profitability, customersatisfaction and assessment of the team bysuperiors, peers and subordinates.

Losada found that the average ratios of thethree bipolar scales and the estimated connec-tivity parameter correlated with the performancelevel of a team. On the average, high performingteams had high positivity/negativity ratios, andinquiry/advocacy and other/self ratios near one.Low performing teams had low positivity/negativity ratios, and low inquiry/advocacyand other/self ratios, that is there is moreadvocacy than inquiry and more self-orientationthan other-orientation. High performance teamshad a high level of connectivity, whereas lowperformance teams had a much lower level ofconnectivity. Medium performance teams werefound to be somewhere in themiddle. For details,see Table 1.

The amplitudes of the time series of theobservations were high and nondecreasing for

the high performance teams, whereas the lowperformance teams’ time series ‘showed adramatic decrease in amplitude. . .about the firstfourth of the meeting and stayed locked. . .’(Losada, 1999). Again medium performanceteams were found to be somewhere in the middle(Losada 1999; Losada and Heaphy, 2004).

In search of a model that would produce a timeseries similar to the time series of coded obser-vations, Losada ends up suggesting a set ofnonlinear differential equations identical to thoseLorenz (1963) presented in his seminal paperentitled Deterministic Nonperiodic Flow. Lorenzused these equations to describe the dynamicsof heat convection in a fluid. This is alsomentionedin (Losada, 1999) and (Losada and Heaphy, 2004),but there is no further discussion why theseequations should describe the dynamics of thecoded observations of verbal communication.Only very limited explanations are given aboutthe modelling process and the meaning andinterpretation of its parameters (see Losada,1999, p. 182). Thus, the reasoning behind themodel equations remains unclear to the reader.

THE MODEL

Losada’s model of team interaction has threevariables: inquiry–advocacy (denoted by X), other–self (denoted by Y) and emotional space (denotedby Z). We provide our interpretation of thevariables and equations of the model, in order toclarify the possible reasoning behind it.

The model variables X and Y refer to thedifference between the number of inquiry andadvocacy speech acts and the number ofother-referring and self-referring speech acts,respectively. At time t, X and Y are thought torepresent the subtraction of the amount of

Table 1. Average positivity/negativity, inquiry/advocacy and other/self ratios, and rounded averages of the connectivityparameter (Losada and Heaphy, 2004, p. 747)

Positivity/negativity Inquiry/advocacy Other/self Connectivity

High performance teams 5.614 1.143 0.935 32Medium performance teams 1.855 0.667 0.633 22Low performance teams 0.363 0.052 0.034 18




advocacy speech acts from the amount of inquiryspeech acts and the amount of other-referringspeech acts from the amount of self-referr-ing speech acts, respectively. Thus, a positiveX(t) means that there is more advocacy thaninquiry at time t, whereas a negative X(t) meansthat there is more inquiry. Similarly, a positiveY(t) means that there are more self-referring thanother-referring speech acts at time t, whereas anegative Y(t) means that there are more other-referring speech acts. Themodel does not includea variable which would indicate the difference ofthe number of positive and negative speech acts.Instead, it is considered that the emotional spacevariable indicates the ratio of positive andnegative speech acts (Losada and Heaphy,2004, p. 757). The ratio is computed using theequation:

P=N ¼ ZðtÞ � Zð0Þb

(1)

where Z(t) is the level of emotional space at timet,Z(0) the level emotional space at time t¼ 0 and bis a dissipation coefficient of the emotional spacevariable, see Equation (2) below (Losada andHeaphy, 2004, p. 757). Thus, high values ofemotional space correspond to high positivity/negativity ratios, whereas low values ofemotional space correspond to low positivity/negativity ratios (Losada, 1999).

The rate of change of emotional space is assumedto depend on the interaction between inquir-y–advocacy and other–self and on the level ofemotional space itself, so that

dZ

dt¼ XY� bZ (2)

where b is the proportionality coefficient deter-mining the dissipation rate of emotional space. Inthe model2 b is fixed to 8/3. The growth rate isdetermined by the product of inquiry–advocacyand other–self variables. The dissipation term,

�bZ, is directly proportional to the level ofemotional space itself.

Emotional space increases when inquir-y–advocacy and other–self are both eithernegative or positive and the product of thesevariables is greater than the absolute value of thedissipation term. Thus, whether emotionalspace either decreases or increases, depends onthe quadrant, on the (X, Y) plane, in which theteam operates (see Figure 1). The justification ofthis assumption is, however, unclear. There canbe problematic situations, for example, whenspeech acts refer to a person or a group in the labor within the company. The model predicts thatin this case emotional space, and thus positivity,increases only if speech acts involve arguing infavour of the speaker’s viewpoint rather thanexploring and examining another teammember’sposition.

The rate of change of other–self is assumed todepend on all the three variables and, in addition,on the connectivity of the team, so that

dY

dt¼ cX � XZ� Y (3)

where c is the connectivity parameter. The rate ofchange of Y increases with the level of X, theinquiry–advocacy variables, with the connec-tivity parameter as a multiplier. The secondterm �XZ represents interaction between the

Figure 1. The sign of the term XY on the (X, Y) plane

2One should note that there is a typographical error in Losada (1999):parameters b in Equation (2) and a in Equation (4) have been inter-changed in the paper. In both papers that present the model equations,Losada (1999) and Fredrickson and Losada (2005), a¼ 10 and b¼ 8/3,but in Losada (1999), a is in Equation (2) and b in Equation (4). Theequations are correct in Fredrickson and Losada (2005).




inquiry–advocacy and emotional space variables.The impact of the can be positive or negative. Thesign of the term depends solely on the sign of Xprovided that Z is nonnegative. It is assumed thatother–self dissipates at a rate proportional to thelevel of other–self itself which is represented bythe �Y term. Rewriting the Equation (3) gives

dY

dt¼ X c� Zð Þ � Y (4)

Note that the term X (c�Z) acts as a positivefeedback mechanism, that is it increases the rateof change of Y, the other–self variables, as long as(c�Z) remains positive. As Z exceeds the level ofconnectivity (Z> c), the term becomes negativeand starts restraining the rate of change.

Finally, the rate of change of inquiry–advocacy isassumed to be proportional to the differencebetween the other–self and the inquiry–advocacyvariables, so that

dX

dt¼ a Y� Xð Þ (5)

The related assumption made here is that thelevel of inquiry–advocacy is connected to thelevel of the other–self variables. If other–self issmaller or greater than inquiry–advocacy, inquir-y–advocacy decreases or increases, respectively.Similarly, in Losada’s (1999) time series ofobservations, self-orientation typically precededadvocacy and other-orientation precededinquiry. Parameter a is a proportionality coeffi-cient determining the rate at which inquir-y–advocacy follows other–self. In the model, ais fixed to 10.

By looking at the set of equations, one is able tosee the impact of the parameter c in the system. InEquation (4) one sees how c, through the (c�Z)term, introduces a negative and a positiveboundary for Y, the other–self variables. Therate of change of other–self, dY/dt, becomes zerowhen Z approaches c. Consequently, it intro-duces a negative and a positive boundary for X,the inquiry–advocacy variables, see Equation (5).Furthermore it sets an upper limit to Z, theemotional space variable, see Equation (2). Thus,it determines how much positivity (see Equation(1)) the system can sustain. It remains unclear,however, whether the estimation method of the

parameter can be said to capture such charac-teristic of team interaction.

It is not explained why all parameters but cshould be considered to be fixed.Why should, forexample, all teams have an equal dissipation rateof the emotional space variable, that is value ofthe parameter b in Equation (2). Losada uses thesame parameter values that are commonly usedin models that apply the Lorenz equations. Thisapplies to the parameters a, b and c as well as toparameters that are implicitly considered unity(e.g. the term XY in Equation (2)). In the originalpapers, it is said that the value 8/3 for theparameter b is used by scholars in manydisciplines who use Lorenz attractors (Losadaand Heaphy, 2004). Besides this note, it is notexplained why the chosen parameter values arethe same ones to those that Lorenz used.

ANALYSIS OF THE MODEL

Model Validation

The model was originally validated by compar-ing the model behaviour with coded obser-vations. The match between theoretical andempirical data sets was ‘indicated by the cross-correlation function at p< 0.01’ (Fredrickson andLosada, 2005). This implies that the model canproduce data similar to the original data. Whenrunning the model, Losada (1999, pp. 183–188;Losada and Heaphy, 2004, pp. 754–755) used oneset of initial values, that is (X0, Y0, Z0)¼ (1, 1, 16)and three values of c, that is 18, 22 and 32.According to Losada and Heaphy (2004), theinitial values ‘eliminate transient, whichrepresents features of the model that are neitheressential nor lasting’. It is, however, notexplained why all teams should start with moreadvocacy than inquiry and more self-orientationthan other-orientation. The validation by simu-lation that is presented in the original papersdoes not, in general, guarantee that the modelcould be used to predict behaviour in differentenvironmental conditions. The simulations pre-sented here were run with MATLAB3 and,

3http://www.mathworks.com [accessed 17 April 2007].




following Losada, a fourth-order Runge–Kuttaalgorithm with a time step of 0.02 was used forthe numerical integration.

Qualitative Behaviour of the Model

In the model, parameters a¼ 10 and b¼ 8/3 arefixed and the connectivity parameter, c, is anadjustable one. The lowest value Losada used forthe adjustable parameter, c, was 18 and thehighest value was 32 (Losada, 1999). In ouranalysis, the parameter values are allowed torange from 16 to 35, as this range reflects therange of parameter estimates Losada reported.

Since the model equations are identical to theones in Lorenz’s model of fluid convection,related research (see e.g. Lorenz, 1963, 1993;Yorke and Yorke, 1979; Sparrow 1983; Csernakand Stepan, 2000) can be referred to in studyingthe model behaviour. The Lorenz system is alsodealt with in books on dynamical systems (seee.g. Seydel, 1988; Hilborn, 1994; Alligood et al.,1996; Verhulst, 1996). These studies are of interesthere, as they show how the behaviour of themodel changes as the model parameter c isvaried.

The Lorenz system—characterized byEquations (2), (4) and (5)—has one to threesteady states which are obtained when thetime derivatives are set to zero. This givesthe following three steady state solutions:

X ¼ Y ¼ Z ¼ 0 (6)

X ¼ Y ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffib c� 1ð Þp

Z ¼ c� 1

�(7)

X ¼ Y ¼ � ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffib c� 1ð Þp

Z ¼ c� 1

�(8)

of which the latter two steady states only existwhen c> 1.

With the chosen range of parameter values, thesystem exhibits two types of dynamics: meta-stable chaotic behaviour and chaotic behaviour.In general, deterministic chaos refers to dynamicbehaviour that is bounded, nonperiodic andsensitively dependent on initial conditions (seee.g. Alligood et al., 1996). The concept of sensitivedependence to initial conditions refers to

the phenomenon that a small change in theinitial conditions may lead to significant changein behaviour of the system. This is often called the‘butterfly effect’, referring to the idea that thebeating wings of a butterfly can cause a majoratmospheric phenomenon somewhere far away.Metastable chaos refers to dynamic behaviourthat is similar to chaotic behaviour, with theexception that the chaotic behaviour is transient,that is after some time the system no longerbehaves chaotically and thereafter evolvestowards an attractor. Accordingly, this initialphase of chaotic behaviour is referred to asa chaotic transient.

For parameter values a¼ 10, b¼ 8/3 and16 < c < c1 � 24:06 there is a region of meta-sable chaos. Depending on the initial state thesystem either converges to one of the steadystates, Equation (7) or (8), or exhibits a chaotictransient before eventually ending up in one ofthe steady states. For c1 < c < ccr � 24:74, thereis also a chaotic attractor. Depending on theinitial state, the system will converge to one ofthe two stable steady states or it will behavechaotically indefinitely. When c passes ccr, whichis here the last relevant critical value of theparameter c, the two steady states lose theirstability and the system evolves towards thechaotic attractor from all initial states. For detailsof the Lorenz model, see for example (Sparrow,1983; Dykstra et al., 1997).

Model Dynamics for the Three TeamPerformance Categories

Themodel thus predicts metastable chaos for low(c¼ 18) and medium (c¼ 22) performance teamsand chaotic dynamics for high performanceteams (c¼ 32). Low and medium performanceteams end up in either steady state, whereashigh performance teams end up in neither.The interpretation of the steady state solutionEquation (7) is that there is more advocacythan inquiry and more self-referring speech actsthan other-referring speech acts, whereas inEquation (8) there is more inquiry and other-referring speech acts. Chaotic solutions do notend up in either steady state, but oscillate around




both, thus predicting roughly an equal amountof inquiry and advocacy, and other- and self-referring speech acts.

The time series of observations of the lowperformance teams ‘showed a dramatic decreasein amplitude for all three dimensions about thefirst fourth of the meeting and stayed locked [. . .]for the rest of the meeting’ (Losada, 1999, p. 182).The time series of observations of the mediumperformance teams in all three dimensions‘tended to have patterns of decreasing ampli-tude’ (Losada, 1999, p. 182). Themodel behaviouris in agreement—at least, intuitively—with whatis reported of the time series of observations.This is because for those values of parameterc that correspond to low and medium perform-ance teams, the system potentially displays achaotic transient after which it convergestowards either of the steady states. The reporteddecrease of amplitude—presumably happeningsooner for low than for medium performanceteams—is similar to the decreasing amplitude ofoscillations observed in the model behaviourafter the potential chaotic transient. On theaverage, this decrease of amplitude is observedsooner for smaller values of the parameter c,when the initial state is chosen randomly. Thus, itcan be said that the model predicts the decreaseof amplitude sooner for low performance thanfor medium performance teams.

‘High performance teams had time series thatshowed high amplitudes over thewhole durationof the meeting in all three dimensions’ (Losada1999, p. 182). The model displays chaoticbehaviour for those values of the parameter cthat correspond to high performance teams.Thus, the model predicts persistently highamplitudes for time series projections of themodel behaviour. All in all, our simulations ofLosada’s model are in accordance with Losada’sdescription of the general characteristics of thetime series of observations.

Model Predictions

The initial values of the inquiry–advocacy andother–self variables affect the behaviour of themodel. Since for sufficiently small c, the two

steady states outside the origin are stable theteams with relatively low connectivity can endup in either of these two steady states.

When using the initial values presented in theoriginal papers, that is (X0, Y0, Z0)¼ (1, 1, 16), andwhen connectivity is set to 18, the behaviour ofthe model is in accordance with the observation,that low performance teams have low inquiry/advocacy and other/self ratios. By changing theinitial values to (X0, Y0, Z0)¼ (�1, �1, 16), themodel shows qualitatively the same dynamics,that is spiralling towards a steady state. How-ever, now the system lies all the time in thequadrant inwhich bothX andY are negative, thatis inquiry and other are dominant. This seems tobe in disagreementwith the observed findings, asillustrated in Figure 2.

Teams with connectivity of 20 are typicallyconsidered to be of low or medium performance(see Table 1). In both cases, according to theobservations, there should be more advocacyand self-orientation than inquiry and other-orientation. When connectivity is set to 20 andthe model is run with the same initial state asin the original papers, the system displays achaotic transient thereafter ending up in thequadrant inwhich bothX andY are negative, thatis there is more inquiry than advocacy andmore other-orientation than self-orientation. Thisbehaviour seems to be in disagreement with thereported observations (see Figure 3).

Due to the characteristics of the transientchaotic behaviour of the model, the stable steadystate to which the system eventually convergesto, can change due to a small shift in the initialstate or the value of the connectivity parameter,or due to a change in the numerical integrationalgorithm. This is why the model displays thebehaviour described above and illustrated inFigure 3. However, no reference to this type ofbehaviour is given in the original papers.

DISCUSSION

One particularly interesting observation reportedby Losada is that high performance teams’ timeseries showed high amplitudes in all threedimensions. Because roughly an equal amount




of inquiry and advocacy was observed in highperformance teams, these teams are high ininquiry but also high in advocacy. The sameapplies to other and self variables and possibly topositivity and negativity as well. While theamount of negativity, advocacy and self-orientation may not be distinctively lower forhigh performance teams as such, the degree ofconnectivity is significantly higher for highperformance teams. From the SI perspective, thisfinding is interesting as the bipolar variables areindicative of observable ‘outcomes’ of inter-

action, whereas the degree of connectivityreflects the characteristics of the dynamics ofthe interaction process.

What seems to be differentiating high per-formance teams from low performance teams isthat high performance teams do not get ‘locked’with negativity, advocacy or self modes, they areable to dissolve these ‘attractors’, that is they areable to meta learn. This also implies thatthe average positivity/negativity, inquiry/advo-cacy and other/self ratios do not tell the wholestory of team interaction and its relation to team

Figure 3. The development of (a) inquiry–advocacy, X,and (b) other–self, Y, over time, when c¼ 20 from the initial

state (X0, Y0, Z0)¼ (1, 1, 16)Figure 2. The development of inquiry–advocacy, X, other–self, Y, and emotional space, Z, in the (a) (X, Z) and (b) in the(Y, Z) planes when c¼ 18. The projected trajectories refer to

two different initial states




performance. The SI approach is interested indeveloping such capabilities of human agentswhere one avoids myopic behavioural schemaswhich might result in undesired lock-ins, forexample, by finding ways to avoid getting lockedinto temporary negativity or advocacy.

Saarinen and Hamalainen (2004, see also(Hamalainen and Saarinen, 2006)) describeanother behavioural pattern resulting in alock-in. The mechanism is called the system of‘holding back in return’ which refers to reactivebehaviour commonly observed in everyday life.Human agents reacting to their social environ-ment by reciprocally ‘holding back’ co-produce alock-in system. For example, an individual may‘hold back’ inquiry as others seem to be alwaysresponding with advocacy, thus evokingadvocacy in the individual and, furthermore,advocacy in others. This pattern may give riseto a perception that this lock-in cannot bedissolved, furthermore reinforcing the patternwhich is beneficial to no one. We see that thisbehavioural pattern is similar to the undesiredlock-ins in the Losada setting. We are interestedin how an individual, with his/her SI, is able toavoid and dissolve such lock-ins. One could saythat ML as a micro-behavioural competence ofteams reflects the same human competencewhich we call SI.

SI research strives to search anddevelopmodelsof social interaction which help reveal the hiddenpotential that is inherent in most human systems.These modelling approaches can include theanalysis of the ‘attractors’ that Losada observed.Such models are interesting, we believe, as theycan bring fresh insight into what people aredoing when they manage to facilitate success intheir organizations. They can contribute to betterunderstanding of the characteristics and quality,that is the process, of interaction between peopleto complement perspectives focusing on thecontent of team interaction.

REFERENCES

Alligood KT, Sauer TD, Yorke JA. 1996. Chaos—AnIntroduction to Dynamical Systems. Springer: NewYork.

Axelrod R. 1984. The Evolution of Cooperation. BasicBooks: New York.

Beebe B, Lachmann FM. 1998. Co-constructing innerand relational processes: self- and mutual regulationin infant research and adult treatment. PsychoanalyticPsychology 15: 480–516.

Beebe B, Jaffe J, Lachmann FM, Feldstein S, Crown C,Jasnow M. 2000. Systems models in developmentand psychoanalysis: the case of vocal rhythm coordi-nation and attachment. Infant Mental Health Journal21: 99–122.

Blume L. 1993. The statistical mechanisms of strategicinteraction. Games and Economic Behavior 4: 387–424.

Boston Change Process Study Group. Bruschweiler-Stern N, Harrison AM, Lyons-Ruth K, Morgan AC,Nahum JP, Sander LW, Stern DN, Tronick EZ, 2002.Explicating the implicit: the local level and themicroprocess of change in the analytic situation.International Journal of Psychoanalysis 83: 1051–1062.

Collins R. 1981. On the microfoundations of macro-sociology. The American Journal of Sociology 86:984–1014.

Collins R. 2004. Interaction Ritual Chains. PrincetonUniversity Press: Princeton.

Collins R, Hanneman R. 1998. Modelling the inter-action ritual theory of solidarity. In The Problemof Solidarity: Theories and Models, Doreian P,Fararo T (eds). Gordon Breach: Amsterdam; 213–237.

Csernak G, Stepan G. 2000. Life expectancy calcu-lations of transient chaotic behavior in the Lorenzmodel. Periodica Polytechnica—Mechanical Engineer-ing 44: 9–22.

Dykstra R, Malos JT, Heckenberg NR. 1997. Metastablechaos in the ammonia ring laser. Physical Review A56: 3180–3186.

Fogel A, Garvey A. 2007. Alive communication. InfantBehavior & Development 30: 251–257.

Fredrickson B, Losada M. 2005. Positive affect and thecomplex dynamics of human flourishing. AmericanPsychologist 60: 678–686.

Gintis H, Bowles S, Boyd RT, Fehr E. 2005. MoralSentiments and Material Interests: The Foundations ofCooperation and Economic Life. The MIT Press: Cam-bridge.

Gottman J, Murray J, Swanson C, Tyson R, Swanson K.2002a. The Mathematics of Marriage—Dynamic Non-linear Models. The MIT Press: Cambridge.

Gottman J, Swanson C, Swanson K. 2002b. A generalsystems theory of marriage: nonlinear differenceequation modeling of marital interaction. Personalityand Social Psychology Review 6: 326–340.

Hamalainen RP, Saarinen E. 2006. Systems Intelli-gence: a key competence in organizational life.Reflections: The SoL Journal on Knowledge, Learningand Change 4: 17–28.




Hamalainen RP, Saarinen E. 2007. Systems intelligentleadership. In Systems Intelligence in Leadershipand Everyday Life, Hamalainen RP, Saarinen E(eds). Helsinki University of Technology, SystemsAnalysis Laboratory Research Reports: Espoo;3–38.

Hilborn R. 1994. Chaos and Nonlinear Dynamics: AnIntroduction to Scientists and Engineers. Oxford Uni-versity Press: New York.

Jackson MC. 2000. Systems Approaches to Management.Kluwer Academic/Plenum Press: New York.

Kobda S, Perc M, Marhl M. 2005. Detecting chaos froma time series. European Journal of Physics 26: 205–215.

Lorenz EN. 1963. Deterministic nonperiodic flow. Jour-nal of the Atmospheric Sciences 20: 130–141.

Lorenz EN. 1993. The Essence of Chaos. University ofWashington Press: Seattle.

Losada M. 1999. The complex dynamics of high per-formance teams. Mathematical and Computer Model-ling 30: 179–192.

Losada M, Heaphy E. 2004. The role of positivity andnegativity in the performance of business teams. TheAmerican Behavioral Scientist 47: 740–765.

Morel B, Ramanujam R. 1999. Through looking glass ofcomplexity: the dynamics of organizations as adap-tive and evolving systems. Organization Science 10:278–293.

Saarinen E,Hamalainen RP. 2004. Systems intelligence:connecting engineering thinking with human sensi-tivity. In Systems Intelligence—Discovering a HiddenCompetence in Human Action and OrganizationalLife, Hamalainen RP, Saarinen E (eds). HelsinkiUniversity of Technology, Systems AnalysisLaboratory Research Reports, A88, October 2004,9–37.

Senge P. 1990. The Fifth Discipline: The Art and Practice ofLearning Organizations. Doubleday Currency: NewYork.

Senge P, Ross R, Smith B, Roberts C, Kleiner A. 1994.The Fifth Discipline Fieldbook: Strategies and Tools forBuilding a Learning Organization. Doubleday Cur-rency: New York.

Seydel R. 1988. From Equilibrium to Chaos: PracticalBifurcation and Stability Analysis. Elsevier: New York.

Simon HA. 1952. A formal theory of interaction insocial groups. American Sociological Review 17: 202–211.

Sparrow C. 1983. An introduction to the Lorenzequations. IEEE Transactions on Circuits and Systems30: 533–543.

Stacey R. 1995. The science of complexity: an alterna-tive perspective for strategic change management.Strategic Management Journal 16: 477–495.

Stacey R. 2001. Complex Responsive Processes in Organ-izations: Learning and Knowledge Creation. Routledge:London.

Stacey R. 2003. Learning as an activity of interdependentpeople. Learning Organization 10: 325– 331.

Stacey R. 2007. Strategic Management and OrganizationalDynamics: The Challenge of Complexity to Ways ofThinking About Organizations (5th edn). FT PrenticeHall/Pearson Education: Harlow.

Verhulst F. 1996. Nonlinear Differential Equations andDynamical Systems (2nd edn). Springer: Berlin.

Yorke JA, Yorke ED. 1979. Metastable chaos: thetransition to sustained chaotic behavior in the Lor-enz model. Journal of Statistical Physics 21: 263–277.

Young PH. 2001. The dynamics of conformity. In SocialDynamics, Durlauf S, Peyton PH (eds). The MITPress: Cambridge; 133–153.




Perspectives on team dynamics: Meta learning and systems intelligence

Documents

Transcript of Perspectives on team dynamics: Meta learning and systems intelligence