Word Search


The emergence of knowledge exchange: an agent-based model of a software market

The emergence of knowledge exchange: an agent-based model of a software market
of 12

Please download to get full document.

View again

All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
  1056 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 38, NO. 5, SEPTEMBER 2008 The Emergence of Knowledge Exchange: AnAgent-Based Model of a Software Market Maria Chli and Philippe De Wilde,  Senior Member, IEEE   Abstract —We investigate knowledge exchange among commer-cial organizations, the rationale behind it, and its effects on themarket. Knowledge exchange is known to be beneficial for in-dustry, but in order to explain it, authors have used high-levelconcepts like network effects, reputation, and trust. We attemptto formalize a plausible and elegant explanation of how and whycompanies adopt information exchange and why it benefits themarket as a whole when this happens. This explanation is based ona multiagent model that simulates a market of software providers.Even though the model does not include any high-level concepts,information exchange naturally emerges during simulations as asuccessful profitable behavior. The conclusions reached by thisagent-based analysis are twofold: 1) a straightforward set of as-sumptions is enough to give rise to exchange in a software market,and 2) knowledge exchange is shown to increase the efficiency of the market.  Index Terms —Adaptive behavior, agent-based modeling, busi-ness economics, cooperative systems, intelligent agents, multiagentsystems. I. I NTRODUCTION T HE GROWTH of the Internet as a medium of knowledgeexchange has stimulated a lot of scientific interest srci-nating from various disciplines. The willingness of individuals,organizations, as well as commercial firms to share informa-tion via the Internet has been remarkable. In some sectorslike scientific research, the communication of newly acquiredknowledge and expertise in a field is considered vital for theiradvancement. On the other hand, in other sectors, the benefitsof such exchanges may not be obvious. For instance, it mighteven be considered damaging for pharmaceutical companiesto make public any innovations generated by their researchand development (R&D) process. In spite of this view, theexchange of intellectual property in some industries occursquite frequently and in various different ways. These includethe forming of strategic partnerships, the participation in opensourcesoftwareprojects,andthepublicationofscientificpapersby research labs that are part of commercial companies. Manuscript received April 16, 2006; revised October 12, 2007. This work was supported by the 6th Framework Programme of the European Commissionunder the project Digital Business Ecosystem (Contract IST–2002–507953).This paper was recommended by Associate Editor R. Popp.M. Chli is with the Department of Computer Science, School of Engineeringand Applied Sciences, Aston University, B4 7ET Birmingham, U.K. (e-mail:m.chli@aston.ac.uk).P. De Wilde is with the Intelligent Systems Laboratory, Department of Computer Science, School of Mathematical and Computer Sciences, Heriot-Watt University, EH14 4AS Edinburgh, U.K. (e-mail: pdw@macs.hw.ac.uk).Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.Digital Object Identifier 10.1109/TSMCA.2008.2001079 We study the knowledge exchange that occurs in the softwareindustry. In particular, we focus on analyzing the rationalebehind this exchange as well as its effect on the industry. Thecomplexity of software requirements is a characteristic thatdistinguishes the software market from others. However, thefindings of this work might be relevant to other industries aswell.Thiseffortfitswithintheframeworkofthedigitalbusinessecosystem (DBE) project. The DBE project is an attempt to de-velop a distributed environment which will interlink Europeansmall and medium enterprises (SMEs) that are softwareproviders and foster collaboration between them.Our broader interest lies in understanding the dynamics of ecosystems[1]–[3].Furthermore,weareinterestedinanalyzingthe global system properties which emerge from the interac-tions that occur in a market ecosystem. We have been usingtechniques from agent-based modeling to simulate the DBEenvironment. The main aspects of the DBE market are capturedin a model where the SMEs are agents with bounded rationality.This model is then studied using simulations of various settings,and a number of observations are made. One of the mostinteresting observations is that exchanges between the agentssimilar to the ones that happen in real life  arise  in the system.This behavior  emerges  in the market even though the modeldoes not explicitly account for social issues of trust, network effects, or managerial strategies.This paper is organized as follows. The following sectiongives an insight to the DBE project and the characteristic of the market that will be developed. In Section III, we sketchthe background of this work, namely, we review the types of exchanges that occur in markets, giving particular attention tothe software market. Section IV details the model used for theinvestigation carried out. Section V analyzes the experimentsperformed and the results produced, and Section VI concludesthis paper.II. DBEIn this section, we give a brief overview of the DBE project,highlighting its aims and motivation. The characteristics of theend-product are identified, and special attention is given to theefficiency of the market that will be formed.  A. DBE Economy It is stated in [4] that virtual organizations make dynamiccoalitions of small groups possible. In this way, the companiesinvolved can provide more services and make more profits.Moreover, such coalitions can disband when they are no longer 1083-4427/$25.00 © 2008 IEEE Authorized licensed use limited to: Heriot-Watt University. Downloaded on October 8, 2008 at 12:10 from IEEE Xplore. Restrictions apply.  CHLI AND DE WILDE: EMERGENCE OF KNOWLEDGE EXCHANGE 1057 effective. At present, coalition formation for virtual organiza-tions is limited, with such organizations largely static.The overall goal of the DBE project 1 [5] is to launch anew technology paradigm for the creation of a DBE that willinterlink SMEs and particularly software providers. The projectis encompassed by the European Union’s initiative to becomea leader in the field of software application development andto strengthen its SME industry. An open source distributed en-vironment will support the spontaneous evolution, adaptation,and composition of software components and services, allow-ing SMEs that are solution and e-business service providersto cooperate in the production of components and applicationsadapted to local business needs. This will allow small softwareproviders in Europe to leverage new distribution channels pro-viding niche services in local ecosystems and extending theirmarket reach through the DBE framework. Easy access andlarge availability of applications adapted to local SMEs willfoster adoption of technology and local economic growth. Itwill change the way SMEs and EU software providers use anddistribute their products and services.The main objective of this paper, which was carried out aspart of the DBE project, was to study the properties of this newtype of market. It is clear that the interactions and exchangesbetween the SMEs within the DBE environment will have aneffect on the dissemination of information and subsequently tothe efficiency of the market.  B. Market Efficiency Within the environment of the DBE, business alliances,networks, and supply chains require much less effort to beformed.Thiswillpromotecooperationandeasierdisseminationof information between the member SMEs. On the other hand,competition for a share of the market between SMEs willbecome more direct. It is to be hoped that these factors willraise the levels of efficiency in the DBE market in comparisonto a traditional market. While these aspects of the DBE arevery interesting and the subject of future research, this paperstudies how market efficiency is affected by the exchange of information between SMEs. The experiments carried out on ourmodel confirm that as the agents engage in more informationexchanges between them, with time, the market efficiency of the system rises.Efficient markets theory, as proposed by Fama [6], is a fieldof economics which seeks to explain the operation of an assetmarket. Specifically, it states that at any given time, the price of an asset reflects all available information [7], [8]. The efficientmarket hypothesis implies that it is not generally possible tomake above-average returns in the stock market over the longterm by trading lawfully, except through luck or by obtainingand trading on inside information.The DBE environment is different from an asset market, sothe definition of efficiency needs to be modified, retaining thespirit of the efficient market hypothesis. In the model of theDBE used in this paper, the market is driven by demand which 1 See http://www.digital-ecosystem.org for more information about theproject. is fixed and unaffected by the supplied DBE services. In thiscase, the market is efficient if, at any given time, the supplyof a service reflects all available information. This means thatthe services supplied are such that they satisfy the underlyingmarket needs optimally. In other words, the SMEs are notconcentrating on catering for some needs while others are leftunsatisfied. In an efficient DBE market, all the requests/needswill be satisfied evenly, assuming that there is equal demandfor each of them. To draw a parallel between the traditionaldefinition of an efficient asset market and the proposed de-finition for the efficiency of the DBE market, consider thefollowing. In an inefficient asset market, a trading agent canearn excessive returns by buying a particular stock which shebelieves to be undervalued. Similarly, in an inefficient DBEmarket, a company might make excessive profits by satisfyinga need which it knows is not sufficiently satisfied. To invertthe argument, in an efficient asset market, asset prices adjustinstantaneously and in an unbiased fashion to publicly availablenew information, so that no excess returns can be earned bytrading on that information. Similarly, in an efficient DBEmarket, the supply of services will adjust immediately to anyarising information about the underlying needs.Cooperation, symbiosis [9], [10], as well as the efficiency[11], [12] of adaptive multiagent systems have been studied inthe context of the simple games. In [11], no verifiable definitionof efficiency is given, whereas in [12], the system is consideredto be in an efficient market phase when all information thatcan be used by the agents’ strategies is traded away, andno agent can accumulate more points than an agent makingrandom guesses would. In the work presented in this paper,market efficiency, cooperation, and competition are studied inthe context of a more realistic economic market.III. B ACKGROUND In this section, we list a number of ways in which exchangeof knowledge between companies happens in a market and therationale for each of them is briefly reviewed. As this paperfocuses on SMEs that are software providers, we survey thekey characteristics of the software industry and the exchangesin this particular market.  A. Exchange in Economic Markets In an economic market, there are many ways in which thefirms engage in exchanges between them. These include theforming of strategic partnerships, the participation in opensourcesoftwareprojects,andthepublicationofscientificpapersbyresearchcompanieslikeHPLabsandMicrosoftResearch.Inthe paragraphs that follow, we will briefly examine the rationalebehind these different forms of exchange.For a strategic partnership to be formed, the partners mustmutually benefit from the experience, expertise, and talent thatall the parties bring to the partnership. There usually is animmediate worthy goal or objective that the partners concernedwish to achieve. For instance, they may wish to operate in anew market or to bring about a change of leadership in theindustry they operate in. Hagedoorn [13] reports a dramatic rise Authorized licensed use limited to: Heriot-Watt University. Downloaded on October 8, 2008 at 12:10 from IEEE Xplore. Restrictions apply.  1058 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 38, NO. 5, SEPTEMBER 2008 particularly in R&D partnerships over the past 40 years. Thesepartnerships are mostly limited-time project-based collabora-tions as opposed to long-term alliances. The main motivesbehind them are reported to be related to cost-cutting as wellas risk minimization, whereas the partners attempt to enter newtechnological areas.Recent economics and management research has studied thephenomenon of commercial firms contributing to open sourceprojects. The main motive indicated by these analyses is strate-gic [14], as set out in more detail in Section III-B where thespecifics of the software industry are analyzed. This seems to beconsistent with the fact that it is not the leaders in the industrywho engage in open source development but the followers.Another form of exchange, which at first might seem coun-terintuitive, is the publication of scientific papers containingthe findings of the research commercial companies’ perform. Itmay be argued that it would be in the interest of those com-panies to keep their innovative work to themselves. Anotherargument, however, is that, by publicizing their research, theyinvite others to endorse it, add to it, and, in effect, advanceit further. Then, they can use the knowledge acquired by thisprocess to better their products.The model of a software market that we propose as partof this work is simple in the sense that the agents/firms donot have the ability to reason about complex situations. Theycannot make decisions to operate in new markets or formpartnerships in order to change the leadership in the industry.They cannot devise strategies to undercut their competitors.However, they operate in a capitalistic economy where the bestof them succeed while the worst perish. They are thus equippedwith a simplistic mechanism of reinforcement learning, i.e.,being rewarded or punished for choices that prove to be goodor bad, respectively. When given the opportunity to engage inexchange of services between them, they learn with time underwhichcircumstancesthisisbeneficialtothem,andtheyproceedwith it without ever being biased by external factors towardexchanging.  B. Software Industry Complexity is a key characteristic of software which dis-tinguishes the software industry from others. Typical softwareproducts carry a large number of features, with innumerable[15] interactions between them. For a program to be successfulin the market, it is necessary that it has the right set of featuresto satisfy the customer base and that these features operatesuccessfully together.The market of proprietary software providers/publishers isdominated by large companies, not SMEs. Microsoft Corpo-ration holds the lion’s share in the software market with com-panies like Oracle, IBM, Hewlett-Packard, and Sun followingwith smaller shares. 2 2 The information reflects the year 2002–2003 and was obtained from IBISWorld, a strategic business information provider. http:www.ibisworld.com/ snapshot/industry/default.asp?page = industry \ &industry \ _id = 1239 accessedon 27/05/2005. At the same time, the open source 3 movement has been quitesuccessful in developing relatively complex software productslike Linux, Apache, or sendmail that are serious competitors of well-established proprietary software [16]. Networks of thou-sands of volunteers have contributed to these highly complexproducts. This appears, asitispointed out in [15],tocounter theeconomic intuition that private agents, without property rights,will not invest sufficient effort in the development of publicgoods because of free-rider externalities.Lerner and Tirole in [17] justify the volunteers’ motivationfor contribution to the open source movement as an opportunityto “signal their quality.” In other words, the volunteers believethat it will enhance their career prospects, as the names of thecontributors are always listed in open source projects. Otherindividual motivations, like altruism or opportunity, to expresscreativity are also mentioned.It is important to point out that in recent years, open sourceprojects have not only received contributions by individuals.There have been organized efforts by firms like Sun, IBM,and others that have endorsed such projects. The survey [18]conducted among firms, as well as the account of Gabriel andGoldman [19] of Sun Microsystems and that of [14], listsstrategic reasons behind the motivation of firms to contributeto open source projects. These reasons include efforts to un-dercut rival products, gaining a wider tester base for their ownproducts, initiating a gift economy culture between the firmand the open source developer community (where the firmprovides the software for free and the community providesdebugging or more source code in return), and giving out thesoftware to clients in order to charge for its maintenance andsupport.Previous work in this area includes that of Johnson [20] andBessen [15] who have used mathematical models to explainthe emergence of the open source initiative. Johnson focusesmore on analyzing the individual motives and establishing therelationship between the size of the developer base and whetherthe development goes on. On the other hand, Bessen con-centrates on the firm motives for participation in open sourceinitiatives.Bessenmodelssoftwareasabitstring,eachbitbeinga certain feature of the software. In this way, the notion that thenumber of combinations of features grows exponentially withthe number of features is captured, depicting the complexitythat the software can have. In his paper, he compares opensource development with proprietary prepackaged provision of software and concludes that the two complement each other,recognizing that they serve different groups of customers.The latter suits customers with standard noncomplex softwareneeds, whereas the former serves customers who have softwaredevelopment capabilities and who need more complex soft-ware products.Bonaccorsi and Rossi [21] have designed a multiagent sys-tem simulation with which they explore the circumstances for 3 In open source software, the source code for a program is made open andavailable for anyone to screen. There are different open source licenses whichprescribewhatoneisallowedtodowiththesourcecode,e.g.,screenit,interpretit, make changes, etc. This is in contrast to proprietary software licenses wherethe source code is protected by property rights against modification. Authorized licensed use limited to: Heriot-Watt University. Downloaded on October 8, 2008 at 12:10 from IEEE Xplore. Restrictions apply.  CHLI AND DE WILDE: EMERGENCE OF KNOWLEDGE EXCHANGE 1059 adoption of open source software. They also conclude thatproprietary and open source software will coexist in the future.Their model of the diffusion of the two competing streams of software production takes into account issues like the effect of advertising, network externalities, and achievement of criticalmass as in [22].Thestylizedmodelpresentedinthispapersimulatesamarketin which the companies try to satisfy a set of underlyingsoftware needs with the services that they develop. The com-panies follow simple high-level rules imposed by a capitalisticeconomy. Interestingly, exchanges between the agents similarto the ones that happen in real software markets arise in thesystem. This behavior emerges in the system even though wehave avoided modeling issues like social or strategic motives of the contributors or network effects.IV. A GENT -B ASED  M ODEL OF THE  DBE  A. Agent-Based Modeling Agent-based modeling has been recently used in economicsresearch work to study models of markets, e.g., the Santa Feartificial stock market [23], [24], and their characteristics[25]; in computing-economics interdisciplinary work to studyinformation economies of autonomous agents [26]–[30] andbusiness processes [31]; in social sciences to study emergentbehavior [32] and issues of trust [33] and to perform syndromicbehavior surveillance [34]; and in other disciplines.Much research in multiagent systems explores how refine-ments to one agent’s reasoning can affect the performance of the system [35]. Significant effort has been directed towardformally defining emergence in agent-based systems. A strongemergent property is a property of the system that cannot befound in the properties of the system’s parts or in the interac-tions between the parts [36]. In addition, in [37], the notion of universality is studied: systems whose elements differ widelymay have common emergent features.Agent-based modeling according to Tesfatsion [38] “is amethod for studying systems exhibiting the following twoproperties.1) The system is composed of interacting agents.2) The system exhibits emergent properties, i.e., propertiesarising from the interactions of the agents that cannotbe deduced simply by aggregating the properties of theagents.”In models like the one proposed next, where the interactionof the agents is determined by past experience and the agentscontinually adapt to that experience, mathematical analysisis typically very limited in its ability to derive the dynamicconsequences. In this case, agent-based modeling might be theonly practical method of analysis.We follow a “bottom-up” approach; in Sections IV-B and C,we describe the first principles of agent behavior, and inSection V, we analyze the macroproperties emerging from theagent interactions. A brief overview of the methods used can befound in the Appendix.  B. Setting In this section, the model used for the simulation of the DBEenvironment is set out.SMEs are modeled as agents in a multiagent system. Theservices that the SMEs provide are modeled as bit strings in thesame manner that software services are modeled in [15], eachbit symbolizing a feature of the service. Finally, the underlyingmarket is modeled by a set of requests (market needs) which areexogenous and are generated randomly. The set is fixed duringthe simulation. A request is a bit string of the same size as aservice bit string.Each SME has a population (or portfolio) of services. Thispopulation is not static throughout the lifetime of the SME. If aservice is successful, the SME tends to add similar services tothe portfolio while an unsuccessful service is usually discarded.The whole process is modeled quite elegantly by a genetic al-gorithm (GA) within the portfolio which involves mutation andcrossover with survival of the fittest. Through this population,each SME can choose which request it will try to satisfy. TheGA represents the R&D businesses perform in order to improvetheir services. An overview of GAs is given in the Appendix.The use of GAs is a natural and simple way to model R&D,with minimal assumptions. The GA captures the followingcharacteristics:1) trying to find a solution to a particular problem;2) using a population of possible solutions.Any other method that can capture the aforementioned twocharacteristics may be used in place of the GA.The objective of an SME is to increase its fitness. Each SMEmaintains a portfolio of candidate services, only one of whichwill be submitted to the market. Each candidate service receivesa rating according to how profitable it would be for the SME if it was submitted to the market. This calculation is performedusing the services submitted by all other SMEs in the previousround. The rating of each candidate service within the SMEportfolio is used as follows: 1) to decide on which service tosubmit to the market and 2) to evolve the best services in theportfolio (with mutation and crossover) and eliminate the worstservices.The fitness of a service measures how profitable it is to itsowner. The profitability of a service depends on the following:1) how close the service is to the market needs (service-request similarity);2) how many other services satisfy those needs (limiteddemand).The fitness of an SME equals the fitness of the service itoffers.In the section that follows, we discuss the factors that affectthe fitness (or profitability) of a service. 1) Service-RequestSimilarityandLimitedDemand:  Assumethat there are  m  SMEs in the market, each one offeringa single service. Consider a service  S   and a request  R ,each represented by a bit string of fixed length. Similarityis measured by the percentage of shared bit values between S   and  R , denoted by  d ( R i ,S  j ) ,  0 ≤ d ≤ 1 . If the market Authorized licensed use limited to: Heriot-Watt University. Downloaded on October 8, 2008 at 12:10 from IEEE Xplore. Restrictions apply.  1060 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 38, NO. 5, SEPTEMBER 2008 Fig. 1. Relationship of   φ  with the service-request similarity  d  for  a  = 0 . 2 .The variable  φ  is used to parameterize the fitness landscape (make maximamore or less pronounced). requests are  R 1 ,R 2 ,...,R n  and services in the market are S  1 ( t ) ,...,S  m ( t ) , the fitness of a service  S  j ( t )  is U  j ( t ) = n  i =0 ( φ ( R i ,S  j ( t )) × ρ i ( t ))  (1)where φ ( R i ,S  j ( t )) =  e − 1 − d ( Ri,Sj ( t ) ) α 2 .  (2)The variable  φ  is used to parameterize the fitness landscape(make maxima more or less pronounced), with  α  being a shapeparameter. Fig. 1 shows the relationship of  φ with the similarity d . The weight/discounting factor  ρ  is given by ρ i ( t ) = min  1 ,  1  j =1 φ ( R i ,S  j ( t ))  .  (3)The variable  ρ  models the fact that the demand in the market islimited. When a request is saturated (i.e., too many services tryto satisfy it), then ρ <  1 . Subsequently, the fitness of the serviceis discounted. Otherwise, when  ρ  = 1 , the fitness of the serviceequals  φ .The fitness of an SME is equal to the fitness of the service itsubmits to the market. 2) Satisfaction of Requests and Market Efficiency:  An ad-ditional useful measure is the degree to which a request issatisfied. This is a metric of how saturated it is, in terms of howmany services try to satisfy it and how similar their features areto those of the request. The degree of satisfaction  Q i ( t )  of arequest  R i  at round  t  is given by Q i ( t ) = m  j =1 φ ( R i ,S  j ( t )) .  (4)This measure is necessary for assessing the efficiency of theDBE market. As discussed in Section II-B, in an efficientDBE market, all the market requests will be equally saturated, TABLE IF EW  E XAMPLE  R ULES OF THE  C LASSIFIER  W HICH AN  SME U SES TO D ECIDE ON  W HAT  T YPE OF  P ARTNER TO  C HOOSE FOR AN  E XCHANGE assuming there is the same demand for all of them. Thus, wecalculate the standard deviation  σ ( t )  of the satisfaction valuesof all the requests in the market at round  t . The smaller itis, the more similar to each other the saturation levels of therequests are σ ( t ) =  stdev { Q 1 ( t ) ,...,Q n ( t ) } .  (5)The mean of the saturation values will be constant due to thedemand in the model being fixed. C. Exchange of Services As outlined in Section III-A, exchange of services mayencompass many real-life situations that occur in a market.These include the forming of strategic partnerships of com-panies, participation in free/open source projects, and others.The setting described here is a loose model of such situationswhich aims to identify the basic factors that lead to this generalbehavior of exchanging.In our model, the exchange involves selecting a set of ser-vices from one SME’s portfolio and swapping them with thecorresponding set of services of the other SME’s portfolio.When a company chooses to swap a set of services, this meansthat after the exchange has taken place, it won’t have theseservices in its portfolio any more. The services in a portfolioof a company are sorted according to their fitness (i.e., howprofitable they are to the SME that owns them). The model inits current state supports exchange of services that are in thesame rank in the two portfolios, e.g., the fifth service in theportfolio of one SME with the fifth service in the portfolio of the other. 4 At each time tick, the SMEs need to decide whether theywant to exchange some of their services with one of the otherSMEs. A statistical classification algorithm is used to model thedecision problems that an individual agent faces. An overviewof statistical classification is given in the Appendix. 1) Exchange Decisions:  Every SME has a classifier systemwhich it uses to decide on whether they want to exchange someof their services with one of the other SMEs. The rules of theclassifier are shown in Table I. The objective of an SME at alltimes is to increase its fitness. 4 Experiments have shown that the rank of the services being exchanged isnot of much significance, assuming that services of the same rank are beingexchanged, but we plan to investigate this further in the future. Authorized licensed use limited to: Heriot-Watt University. Downloaded on October 8, 2008 at 12:10 from IEEE Xplore. Restrictions apply.
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks