Implementing continuous improvement using genetic ... - CiteSeerX

5 downloads 3537 Views 95KB Size Report
systems (CAS) for implementing total quality management (TQM) and .... fitness function can be defined by (a) how routines fit with the business model, and ... software process improvement frameworks like the capability maturity model (CMM).
Implementing continuous improvement using genetic algorithms Petter Øgland [email protected] Department of Informatics, University of Oslo P.O.Box 1080 Blindern, 0316 Oslo, Norway ABSTRACT Purpose – On the metaphoric level, much as been written about complex adaptive systems (CAS) for implementing total quality management (TQM) and organizational learning (OL) in turbulent or unpredictable environments. The aim of this paper is to add practical insights on how a specific CAS-technique called genetic algorithms (GA) can be used for designing quality management systems for keeping the organization in a constant state of continuous improvement in unpredictable environments. Methodology/Approach – The paper describes design, implementation and evaluation of a GA method for a TQM program in the context of the climate department of a Scandinavian meteorological institute. The author was at the time working as a quality engineer responsible for designing the genetic algorithm as a change intervention for improving inflow, quality control and computer-generated statistical use of meteorological observations. Findings – In the given organizational context, a genetic algorithm was easy to implement using elementary quality management tools such as statistical process control (SPC), Pareto Analysis and Business Model Assessment. Rather than going through conventional change management steps of unfreeze-change-freeze, the double loop structure of the chosen GA method made it possible to maintain “edge of chaos” stasis of continuous change. As expected from CAS theory, however, the algorithm caused stable improvement in unpredictable environment at the cost of producing redundancy, complexity and less than optimal efficiency. Originality/Value of paper – Although the idea of applying GA to TQM is not new, this paper appears to be the first attempt to go beyond metaphorical ideas and computer simulations in order to actually define, implement and evaluate a GA method for TQM implementation. Keywords Continuous improvement, organizational change, complex adaptive systems, genetic algorithms Type of paper Research paper

1

1. Introduction Continual improvement based on Statistical Process Control (SPC) follows a process of first making sure that the process is stable by eliminating assignable causes of problems, then redesigning the process, and then bringing the improved process into a state of stability (Wheeler & Poling, 1998). In the language of Lewin, the process can be described as following the stages of unfreeze-change-freeze (Gold, 1999), and as illustrated by Deming in his “funnel experiments”, trying to change unstable processes can lead to meaningless plan-do-check-act (PDCA) designs and disastrous results (Deming, 1986; 1994). However, in politically turbulent or otherwise complex organizations, it may not be possible or to bring a process into a state of statistical control, leading some organizational researchers into suggesting the theory of complex adaptive systems (CAS) for understanding and designing organizational change (e.g. Goldstein, 1993). Genetic algorithms (GA) is a CAS theory that has been used for design in robotics and artificial intelligence (Holland, 1998; Brooks, 2002), and it has also been suggested as a strategy for organizational change in complex or unpredictable environments (e.g. Dooley et al, 1995). GA has been used as a metaphor (Dooley, 2000; Axelrod & Cohen, 2000), as a computer method for simulations (Bruderer & Singh, 1996), and, to a lesser extent, for developing computer-based decision support systems (Greer & Ruhe, 2003). The aim of this paper is to contribute to the understanding of how to flowchart the GA in the style of a PDCA cycle as basis for implementing quality management systems (QMS). The research is framed by the following two questions:  

Is it possible to use the GA approach for effective QMS design? If it is possible, why is it not more popular?

The paper is structured by presenting a literature review in section two. In section three there is a presentation of methodology for implementing and testing a GA-based QMS in an empirical setting. The results of the experiment follow in section four, and there is a discussion of results in section five. In section six, the paper concludes by saying that it is possible and relatively easy to implement a GA-based QMS, suggesting that the lack of popularity of this approach may be sociological rather than technical.

2. Literature review 2.1 Continuous improvement In their overview on the literature on continuous improvement, Bhuiyan and Baghel (2005) show that the term “continuous improvement” can refer to a wide range of improvement strategies, from small step ongoing improvement to breakthrough projects. Although this seems to contradict the mathematical meaning of “continuous”, which would refer to small step ongoing improvement as continuous and rather use phrases like “discrete improvement” or “discontinuous improvement” for breakthrough projects, the question of whether a process can be described as mathematically continuous can be seen 2

as a question of scale, a continuous curve for approximating discrete changes always being dependent on the size of changes relative to units of measurement (Bender, 2000). Deming (1986, p. 87) and writers in the Deming tradition (e.g. Sherkenbach, 1991; Wheeler & Poling, 1998) often use the word “continual” for describing the SPC-driven unfreeze-change-freeze style of improving processes. Juran (1995) prefers to call it “breakthrough management”. Contrasting continual breakthrough improvement with the “continuous improvement” in “gemba kaizen” (Imai, 1986; 1997), an approach that focuses on management interventions on the factory floor for stimulating a culture of improvement, the continual breakthough improvement may be seen as a scientifically distant and theoretically driven approach. As pointed out by Imai (1997, p. 14), “gemba should be the site of all improvement and the source of all information. … When management does not respect and appreciate gemba, it tends to dump its instructions, designs, and other supporting services – often in complete disregard of actual requirements.” According to Dooley et al (1995), the Deming/Juran approach of continual breakthrough improvement was developed in the manufacturing context and can only be successful in organizations that are sufficiently stable for making this approach work. TQM designs based on the theory of complex adaptive system (CAS) are suggested for organizations where this stability is not attainable. As a theoretical example of organizations that seem to fit with the Deming/Juran approach, Mintzberg (1993), among his five models for organizational design, uses the name “machine bureaucracy” for describing organizations like a national post office, a security agency, a steel company, a custodial prison, an airline, and a gigantic automobile company. On the other side of the spectrum, Mintzberg’s “professional bureaucracy”, exemplified by universities, general hospitals, school systems, public accounting firms, social-work agencies and craft production first, might be an example of a type of organization that is more challenging with the Deming/Juran approach. Concerning standardization and control, Mintzberg says of the professional bureaucracy that “a great deal of the power over the operation work rests … with the professionals” (ibid, p. 195), and “the professional’s power derives from the fact that not only is his work too complex to be supervised by managers or standardized by analysts, but also his services are typically in great demand” (ibid). As a way of addressing the problem of process improvement in a complex environment, Axelrod & Cohen (2000, pp. 113-114) chose a perspective that builds on kaizen, but rather than focusing on individual problems and procedures to be improved, they follow the idea of Nelson & Winter (1982) in conceptualizing the organization as an evolving population of work routines. This is the population of routines in terms of how work is being carried out in practice, regardless of how they are described in the QMS. Contrary to the need for stability in order to improve the organization project-by-project (Juran, 1995), the idea in CAS is to understand the population as being on the edge of order and chaos (Waldrop, 1992). Even though this puts a different perspective on 3

quality management, it does not necessarily mean that well-known models and methods from TQM have to be abandoned. Dooley et al (1995) suggest the CAS method of genetic algorithms (GA) for developing models of parallel learning by running a large number of PDCA improvement cycles simultaneously.

2.2 Genetic algorithms There are many introductory books on the topic of genetic algorithms (e.g. Mitchell, 1996; Goldberg, 1989). The framework to be presented here is based on the simplest and most elementary ideas from GA theory. A typical pseudo-code for genetic algorithms provides a starting point (Wikipedia, 2009): 1. Choose initial population 2. Evaluate the fitness of each individual in the population 3. Repeat until termination: (time limit or sufficient fitness achieved) 3.1. Select best-ranking individuals to reproduce 3.2. Breed new generation through crossover and/or mutation (genetic operations) and give birth to offsping 3.3. Evaluate the individual fitnesses of the offspring 3.4. Replace worst ranked part of population with offspring For implementing this algorithm as a PDCA-like change management process in an organizational context, rather than as an algorithm to be run on a computer, key concepts need to be defined in organizational terms. As mentioned at the end of section 2.1, the general idea is to evolve the population of organizational routines, increasing productiviy by creating good routines and removing bad ones. The starting point is the idea that a work routine (“chromosone”) can be seen as containing work tasks (“genes”), where tasks are the smallest units of standardized work that can be work carried out by a person or executed as a computer program. The population chosen in the first step is a list of all relevant routines, including routines that are being carried out at the moment and routines that seem relevant in the future. A fitness function can be defined by (a) how routines fit with the business model, and (b) by internal errors detected in each task, either as they are explicitly observed or as a result of warnings through use of SPC. In step three, the test criterion can be defined by the organization having reached organizational excellence, which in practice means that the loop will never stop. The four steps inside the loop starts with identifying routines that are both fit with the business model and are not marked by internal problems or immediate needs for quality and efficience improvement. The ranking can be done through the use of Pareto analysis both on short-term and longterm interval basis, for example day by day and month by month. Step 3.2 consists of constructing a new set of routines that consists of new designs plus changes in the descriptions of some of the old ones, like adding, removing or changing individual tasks within a given routine. In step 3.3 each individual in this offspring 4

population of routines is measured against a fitness landscape consisting of user requirements, business model assessment, errors and SPC results etc., thus making it possible to rank all the routines by use of Pareto analysis. Individual routines that are non-obsolete (fitness different from zero) are used as input for step 3.4 where they define the new population of routines. The main difference between the standard PDCA cycle and this interpretation of the genetic algorithm is that the PDCA cycle is usually used for improving individual processes (tasks or routines), while the purpose of the genetic algorithm is to evolve the total population of routines by following a method that focuses on the current situation rather than following an initially developed plan. However, the genetic algorithm does not conflict with the PDCA idea. The PDCAplanning (“plan”) is integrated in the way the offspring is defined, the experiment (“do”) corresponds with how tasks are being monitored through direct observation and SPC, while control (“check”) corresponds with how the fitness fuction works. The return to each cycle with the corresponding evaluation of the cycle criterion corresponds with the decision issue (“act”) from PDCA theory. There have been many applications of GA in operations research and manufacturing (e.g. Chambers, 2000). Less seems to have been written on the application of the ideas from GA in the context of organizational change. Axelrod & Cohen (2000) present a general framework for discussing organizational change through concepts like variation, interaction and selection, but although they state that the approach is based on genetic algorithms (ibid, xvii, 10-11) no explicit algorithms are presented in the book. Goldberg (2000) writes about the fundamental intuition in genetic algorithms, suggesting the equation “selection + mutation = continual improvement” and describing hillclimbing on the fitness landscape towards better and better solutions as something human beings do quite naturally, and “in the literature of total quality management this sort of thing is called continual improvement or as the Japanese call it, kaizen” (ibid, p. 9). Bruderer and Singh (1996) come closer to the aim of this paper as they actually define a genetic algorithm for organizational evolution. Their approach is similar to what is described above, building upon the idea from Nelson & Winter (1982) in using the population of routines as the unit of analysis, and defining an algorithm that is not identical to the one above, but similar. However, rather than implementing the algorithm as a method for doing organizational change, they use it for doing computer simulations.

2.3 Development of complex information systems While TQM had an impact on software engineering in terms of creating TQM-inspired software process improvement frameworks like the capability maturity model (CMM) (Humphrey, 1989), in recent years much of the inspiration for software process improvement has come from Lean Production in creating “agile” methods of software process improvement (e.g. Poppendieck & Poppendieck, 2003).

5

Much of the reasoning behind the agile software development idea is similar to that of kaizen and genetic algorithms. Instead of delivering a monolithic system after a long development time, smaller releases are implemented sequentially. Greer & Ruhe (2003) explain a method based upon genetic algorithms for optimizing the allocation of requirements to increments in the agile development process. While this is not the same as designing the QMS by GA, it is an example of integrating GA as a part of the QMS. CAS and GA have also made an impact on information systems theory. In order to understand the technological and sociological dynamics in developing complex information systems, like the implementation of an ISO 9000 based QMS in a large professional bureaucracy, the theory of information infrastructures offers a design methodology through the use metaphors from CAS and the GA-like “bootstrap algorithm” (Hanseth & Aanestad, 2003; Hanseth & Lyytinen, 2004). Although the ideas behind the bootstrap algorithm are somewhat similar to those of genetic algorithms, the bootstrap algorithm is a method that consists of improvisation and political maneuvering, thus not an algorithm in a technical sense.

3. Methodology As the main method in this research consists of reflecting upon work done by myself and colleagues at the Norwegian Meteorological Institute (DNMI) about ten years ago, I have tried to align the research approach with guidelines and suggestions on reflection offered by Atkins (1993). The starting point for developing a new climate database KLIBAS was in 1989. As I joined the organization in 1991, my understanding of the first years of the development cycle is based on conversations at the time and the final project report (Moe, 1995). The KLIBAS project followed a conventional waterfall lifecycle project management until 1995 when it was put into production. The project team consisted of six research scientists and two software engineers, and the project was extensively documented in a series of more than 60 technical reports, most of them written by myself. The period 1996-99 was a period of maintenance, continuous improvement and further development through prototyping. An even longer series of technical documents were written during the period (see: Øgland, 1999), and there were several research papers presented at conferences. In writing this paper, I have revisited much of the documentation, but only to the extent to be able to make sure that my memory and current understanding of the evolution of the GA-based QMS fits with what was recorded and written at the time of implementation. In 2008 and 2009 I returned to interview some of the people who are now running the organization, trying to get an impression of the current state and different versions of the KLIBAS development story.

6

4. Case study 4.1 Design and development of the KLIBAS system The first period 1991-95 was a period of conventional project management. Even though it was important for the project and the department to work scientifically, and explain to others that they worked scientifically, I would in retrospect argue that the period was non-rational in several respects. The main source of problems was a mix of distrust and lack of competence. While the research scientists of the KLIBAS development team were all highly educated, none of them had any experience in the design of relational database systems. The people with the most practical experience were the software engineers, but perhaps due to their lack of formal education a sense of distrust developed, followed by various problems related to prestige, envy, jealousy, anger, fights and so on. The mixture of a politically turbulent climate with technical inexperience made the development process unpredictable and complex. Although the project leader had practical experience with software development, the development method ended up as a waterfall approach, running through a long period of requirements specifications and design before any coding was done. When implementation started, it was soon discovered that specifications were missing or were wrong, resulting in implementation being done in prototype mode with minimal use of documents developed during requirements specification and design. After six years of development, it became difficult to explain why there was so little progress, thus resulting in the project being pragmatically defined as successfully completed, although probably less than 10% of the project scope had been done.

4.2 Redesigning KLIBAS by use of genetic algorithms By terminating the project and handing it over to maintenance, those who had put their prestige in the project could now safely move to other projects while the remains of KLIBAS could be implemented in a politically less turbulent climate. However, with less management focus, work had to be done with fewer resources and the development team was reduced to two research scientists and one engineer. The second period 96-99 was concerned with finding efficient ways of redesigning KLIBAS in a way that was iterative, user-driven, and resulted in stable and efficient solutions. By accident I happened to stumble upon an overview book on complex adaptive systems (Kelly, 1994), a book that drew parallels between the evolutionary processes in biology and experiments in robotics, artificial intelligence and other areas that were specially designed to produce results in unpredictable and complex environments.

7

AWS: Automatic weather stations

SYNOP PRECIP: Manual precipitation stations

e-mail METAR: Airport weather stations

e-mail

e-mail e-mail

Pareto analysis

UASS: upper air sounding stations

e-mail e-mail Monitoring of system outputs and users (customer satisfaction)

e-mail

e-mail

HIRLAM: quality control by use of forecast data

System monitoring

Figure 1: Topological sketch of the QMS as a population of loosely coupled routines From this seed, the quality management system based on genetic algorithms gradually started to grow. In addition to the book by Kelly, I also studied an article in Scientific American (Holland, 1992), a book on artificial intelligence, a couple of books on quality management and quality control, plus some biology and philosophy. Here are some of the ideas that evolved: 

  

Each computer program should (more or less) automatically produce its own systems documentation, thus designing system documentation that included SPC diagrams and Pareto charts of production monitoring within a textual context that would be useful for error diagnosis. All error handling should be communicated via the e-mail system, thus linking the programs and routines together in a loosely coupled way. A program applying Pareto analysis should make checks in the mailbox each night, sorting programs with errors and helping identifying those with the most problems. Monthly status reports should be written (as automatically as possible) in order to inform management, produce statistical feedback and document continual improvement

The diagram in figure 1 illustrates the loosely coupled topology of the QMS as it was implemented. Figure 2 illustrates the GA design for process improvement of the population of routines as a whole, rather sequential improvement process by process. Compared with the algorithm in section 2.2, the first step of choosing initial population followed directly 8

from the outcome of the KLIBAS project. Evaluation of the fitness of each individual routine, step 2 and 3.3 in the GA, was done by the computer programs automatically reporting problems through the e-mail system, including user comments (customer complaints etc) that were manually recorded on lists which were read by a computer program that mailed the problems forward into the centre of figure 2. Non-compliance with requirement specification and design specifications were done in the same way. Enter office on the morning of day i. Evaluate population: Real-time and nightly automatic data collection for total system by use of e-mail.

Select solutions for next population: Run a Pareto analysis for setting the agenda for the day. This defines the population of processes to be improved.

Perform crossover and mutation: Read, write, discuss; design and implement etc.; the daily practical work of process improvement.

Exit office in the afternoon of day i.

i: = i + 1

Figure 2: Continuous improvement by use of genetic algorithm The model worked nicely for software development in a chaotic environment, but it also produced a system that was (probably) difficult for others to understand, despite a long trail of scientific publications on quality control algorithms and technical documents explaining the computer programs, the routines and how they interrelated. There was no general index for summarizing the total quality of the system, but as this was a professional bureaucracy of climate research, the annual number of scientific, technical and administrative publications could be a good indicator of productivity and improvement for the system. The SPC chart on the left-hand side of figure 3 shows three levels of continual improvement, with the process being out of control in the final measurement, perhaps indicating productivity rise to yet another level. The diagram on the right-hand side shows productivity during the final three years broken down by type of publication.

9

80 60 40 20

Productivity UCL = 112

1999

1998

1997

1996

1995

1994

1993

1992

0

AVG = 91 LCL = 70

Scientific

100

Administ

120

Technical

100 90 80 70 60 50 40 30 20 10 0

140

1997

1998

1999

Figure 3: Continual improvement in productivity as measured in publication volumes As this was a scientific community, only the scientific publications gave merit. The scientific publications were often about development of new algorithms for quality control, usually written for the Nordic co-operation within climate activities (NORKLIM) or other research communities. While the development of the QMS resulted in an average of 2-3 annual scientific publications during the first seven years, the final year resulted in 15 scientific publications.

4.3 KLIBAS ten years after On current visits to the organization I was told that very little, or probably nothing at all of the CAS/GA system remained. Despite the eroding of software, however, some of the ideas continued to live. I noticed that predictions concerning further development of the system that I made at an EU conference (Øgland, 1999) have come true, e.g. the total automation of the QC of climate data. The theoretical work on algorithms for quality control continued to be discussed within the NORKLIM community after my departure in 1999 (Vejen, 2002), and presentation of software during the interviews showed that many of the theoretical ideas develop during the period 1991-99 were now implemented in different ways. Some of the theoretical work on quality control algorithms was also included in recent overview of quality control methods in meteorology, as part of research conducted at the University of Costa Rica (Lopez, 2007).

5. Discussion The case study showed how a quality management system was designed as a genetic algorithm for bringing the organization into a stasis of continuous improvement despite the organization being unpredictable and complex.

5.1 Supportive evidence of the GA approach being effective The evolution of the GA model started with problems in making traditional project management work in a professional bureaucracy, fitting with theories on organizational design and the organizational complexity theories (Mintzberg, 1993; Dooley et al, 1995, Axelrod & Cohen, 2000) presented in section 2.1. The GA model evolved over several 10

years, and the statistics in figure 3 shows that the GA design correlated positively with continual improvement. As explained in the case story, only limited knowledge about GA was used for developing the system (Øgland, 2000). When comparing the evolution of the system with the theories on GA, the data structure (figure 1) and algorithm (figure 2) of the case study fit with the Wikipedia algorithm in section 2.2. The long trail of scientific, technical and administrative documents produced by the system shows not only was the QMS relatively easy to implement along the principles of GA, but it was also highly successful (Øgland, 1999; 2008; Vejen, 2002). As pointed out in section 2.3, although it is difficult to find literature on genetic algorithms used as an overall design principle for agile software development, genetic algorithms have been used for developing tools to be used as a part of the agile development process. The GA-based QMS in the case study was also developed as a decision support system (DSS) fully integrated in the development and maintenance environment. A major difference, however, was that in the case of the KLIBAS system, the idea was not optimizing allocation of requirements to increments, but it was to eliminate decision processes by integrating Pareto analysis into the maintenance system, having the work agenda for each day automatically updated.

5.2 Why others do not use this approach As the basic ideas of genetic algorithms are old and have been debated within the community of organizational research for the past 10-15 years, it was surprising that it was so difficult to find related research on how to use GA for designing a method of continuous improvement. Pondering on explanations for this curious finding, the following list comes to mind: 1. There has been done a lot of research on this, although not under the name of “genetic algorithms”. 2. People are not willing to have their lives run by a computer algorithm. They want to make their own decisions. 3. Using GA generates complexity thus making it difficult to hand over the responsibility for the QMS from one person to another. 4. Quality management professionals with a statistical or industrial engineering background have adopted the beliefs of unfreeze-change-freeze that is the underlying assumption of quality management gurus like Deming and Juran, and have little knowledge of CAS theory. 5. Quality management professionals with a human relations background may find the metaphorical aspects of CAS and GA interesting, but are skeptical of the literal interpretation of such theory in terms of defining and implementing algorithms in a strict sense. 6. There is limited prestige in applying the GA approach. The way of the method is to implement TQM with minimal management commitment. There are few quick wins, and no particular focus on creating “breakthrough improvement”. There is

11

little in the method that would predict that people responsible for designing and running the system would feel valued or acknowledge for their work. The first point is a difficult one. The way Imai (1986) describes kaizen as ongoing improvement by developing standards, improving standards, focusing on details, applying the “logic of the crowd” through the use of quality circles etc., this looks very much like genetic algorithms put in action. Much literature on Japanese quality management seems highly consistent with understanding organizations and society from a CAS perspective, including the formulating of solutions (management systems) as GA (e.g. Fujimoto, 1999). However, there seems to be no mention of genetic algorithms in this literature, something that seems to be reflected in the fact that quality management methods that work well in Japan and other Asian countries do not export well onto cultures based on different values and beliefs (Ishikawa, 1985). Even though there should be literature explaining the GA approach through different terms and words, such literature is of limited value until the practices are described in a universal language, such as the mathematical language of GA, in order to make it possible to analyze and understand this type of quality management in a precise and scientific manner. Point two is interesting from the perspective that the case study has been done in a Scandinavian context where organizational structures are relatively flat, authorities are often questioned, rules are often created but not always followed etc. (Hofstede, 1991), meaning that “management by algorithm” does not seem to fit particularly well with the cultural context of “Scandinavian pragmatism”. Nevertheless, in this particular case it worked well. During the past decade there has been a world-wide growing interest in self-management methods that are described through flowcharts and algorithms (e.g. Allen, 2001; 2004). Some have tried to explain this phenomenon through the theory that self-control leads to “flow” and happiness (Csikszentmihalyi, 1991). In my personal experience, I would say that getting rid of trivial administration by applying a simple system had a positive mental effect. On the other hand, in interviews I’ve done with quality practitioners, few or close to none of them seem to share the idea that the quality management consultant company or the quality department should apply the same sort of quality control and process improvement tools that want others to use (Øgland, 2008). Point three corresponds with my observations in the third part of the case study. When returning to the organization where the GA-based QMS had been implemented, little or none of the system remained despite the fact that it had been working well, it was extensively documented, and there was still need for maintenance and development of more software. While there may be many reasons for why the QMS had been redesigned in different ways, it did not come as a surprise to me that the GA approach was no longer in use. No matter how well the GA-based system had been documented and no matter how well it was working, the main mechanism in the GA approach is adaptation and adaptation builds complexity. In the process of one person taking over the responsibility for a system developed by another person, there is always the question of whether one should invest time in figuring out what the other person has done and continue along that path or rather redesign and follow ones own experiences and beliefs in how things should be done. 12

Concerning point four, while there may be researchers that have studied the implications of CAS in the context of TQM (e.g. Dooley, 1995; 2000), I do not believe these insights have yet been fully adopted by the quality management community in a way that leads to practical implications like designing TQM by GA. As can be observed from the online discussion forum on the American Quality Society, even the moderator of the discussion thread on complex adaptive systems seems to have limited knowledge of CAS beyond the metaphorical level (ASQ, 2009). Point five fits with my personal experience of social isolation and a certain level of alienation as a consequence of formalizing and computerizing management issues, thus reducing the need for meetings and discussions as the machine was more or less fully capable or doing the administrative work of planning, monitoring and evaluating work. The fact that concepts like CAS and GA had little to do with meteorological or climate issues did not help much either. On the other hand, although technical skill was necessary for implementing the system, the idea was to eliminate the “waste” of trivial administrative work in order to be able to focus on creative work. Creative work in this particular case was to a large degree focused on software development, but there was also a monthly discussion with the technical department for discussing automatic weather stations. If properly understood and implemented in a simple fashion, I believe there is no reason why the GA approach should not be used to TQM experts with a human relations focus. The final point about power, pay, prestige and social issues seems relevant on a first approach. As none of my co-workers at DNMI understood anything about CAS and GA, and had little interest in such issues, I had to depend on my peer network consisting of a handful of TQM researchers in other Scandinavian geophysical institutes for making meaningful conversation. Implementing TQM as a long-term, noiseless and invisible process does not correlate well with aspirations of power, prestige, pay or social acknowledgement within the organization, but, on the other hand, I believe these social implications have more to do with the nature of working in the field of quality management than anything particular to do with the GA approach. What makes the GA approach different from other ways of cultivating TQM is that it is designed to handle an unpredictable organizational environment by trading off on efficiency and visibility, or at least that was what could be seen in this case study.

6. Conclusion Classical ideas about organizational change and process improvement are based upon the idea of freeze or stability. In turbulent organizations such stability may not be possible to achieve, so it has been suggested that theory of complex adaptive systems (CAS) and genetic algorithms (GA) could be used instead. Although the GA approach seems so far to only have resulted in computer simulations and metaphors about change, the research question in this paper consisted of first asking whether it would be possible to use the GA approach more literally for effective QMS

13

design, and if this is possible, why is the GA approach not more popular? The conclusion is similarly twofold. 1. It is relatively easy to design an effective quality management system based on the logic of GA, as argued through theory and example from case study. 2. Despite being a simple and effective method for implementing continuous improvement, the GA approach may represent some social challenges. Summarizing from the discussion section, (1) those being fascinated by CAS and GA as management metaphors may not necessarily propagate literal applications of GA, (2) people in general may not be willing to have their lives run by a computer algorithm, and (3) there are few social rewards to be expected from applying GA. Despite challenges, the GA approach proposes to be a simple and effective method that works in environments where traditional TQM methods are predicted not to work. Further research needs to be done on the GA approach in different empirical settings, clarifying what can be done when the technical environment is not as supportive as it was in this case, and more research should be done on social challenges of the GA approach, searching for ways to have the challenges met.

References Allen, D. (2001), Getting Things Done: How to Achieve Stress-Free Productivity, Piatkus, London. Allen, D. (2004), Ready for Anything: 52 Productivity Principles for Work and Life, Piatkus, London. ASQ (2009), Discussion Board: Complexity Theory in Quality, http://www.asq.org/discussionBoards/forum.jspa?forumID=50, accessed on April 29th, 2009. Atkins, S. (1993), “Reflection: a review of the literature,” Journal of Advanced Nursing, Vol. 18, No. 8, pp. 1188-1192. Axelrod, R. and Cohen, M.D. (2000), Harnessing Complexity: Organizational Implications of a Scientific Frontier, Basic Books, New York. Bender, E.A. (2000), Introduction to Mathematical Modelling, Dover, London. Bhuiyan, N. and Baghel, A. (2005), “An overview of continuous improvement: from the past to the present,” Management Decision, Vol. 43, No. 5, pp. 761-771. Brooks, R. (2002), Flesh and the Machines: How Robots Will Change Us, Pantheon Books, London. Bruderer, E. and Singh, J.V. (1996), “Organizational evolution, learning, and selection: A genetic-algorithm-based model”, Academy of Management Journal, Vol. 39, No. 5, pp. 1322-1349. Chambers, L.D. (2000), The Practical Handbook of Genetic Algorithm Applications, 2nd Edition, Chapman & Hall, London. Csikszentmihalyi, M. (1991), Flow: The Psychology of Optimal Experience, Harper Perennial, London. Deming, W.E. (1986), Out of the Crisis, MIT Press, Cambridge, Massachusettes.

14

Deming, W.E. (1994), The New Economics, 2nd Edition, MIT Press, Cambridge, Massachusettes. Dooley, K., Johnson, K., and Anderson, J. (1995), “TQM, chaos, and complexity,” Human Systems Management, Vol. 14, No. 4, pp. 1-16. Dooley, K. (2000), “The paradigms of quality: evolution and revolution in the history of the discipline,” Advances in the Management of Organizational Quality, Vol. 5, JAI Press, pp. 1-12. Fujimoto, T. (1999), The Evolution of a Manufacturing System at Toyota, Oxford University Press, Oxford. Gold, M. (1999), The Complete Social Scientist: A Kurt Lewin Reader, American Sociological Association, Washington, DC. Goldberg, D.E. (1989), Genetic Algorithms in Search, Optimization, and Machine Learning, Addison-Wesley Professional, New York. Goldberg, D.E. (2000), “The Design of Innovation: Lessons from Genetic Algorithms, Lessons for the Real World,” Technological Forecasting and Social Change, Vol. 64, pp. 7-12. Goldstein, J. (1993), The Unshackled Organization: Facing the Challenge of Unpredictability Through Spontaneous Reorganization, Productivity Press, Portland, Oregon. Greer, D. and Ruhe, G. (2004), “Software release planning: an evolutionary and iterative approach,” Information and Software Technology, Vol. 46, No. 4, pp. 243-253. Hanseth, O. and Aanestad, M. (2003), “Bootstrapping networks, communities and infrastructures: On the evolution of ICT solutions in health care,” Methods of Information in Medicine, Vol. 42, No. 4, pp. 385-391. Hanseth, O. and Lyytinen, K. (2004), “Theorizing About the Design of Information Infrastructures: Design Kernal Theories and Principles,” Sprouts: Working Papers on Information Environments, Systems and Organizations, Vol. 4, No. 12, pp. 1-35. Hofstede, G. (1991), Cultures and Organizations: Intercultural Cooperation and Its Importance for Survival, Profile Books, London. Holland, J.H. (1992), “Genetic Algorithms,” Scientific American, Vol. 267, No. 1, pp. 66.72. Holland, J.H. (1998), Hidden Order: How Adaptation Builds Complexity, AddisonWesley, New York. Humphrey, W. (1989), Managing the Software Process, Addison-Wesley, New York. Imai, M. (1986), Kaizen: The Key to Japan’s Competitive Success, Random House, New York, NY. Imai, M. (1997), Gemba Kaizen: A common sense, low-cost approach to management, McGraw-Hill, New York. Ishikawa, K. (1985), What is Total Quality Control? – The Japanese Way, Prentice Hall, Englewood Cliffs, N.J. Juran, J.M. (1995), Managerial Breakthrough: The Classic Book on Improving Management Performance, McGraw-Hill, New York. Kelly, K. (1994), Out of Control: The New Biology of Machines, Social Systems, and the Economic World, Addison-Wesley, New York. Lopez, J.L.A. (2007), Algoritmos de control de calidad de datos en estaciones meteorologicas automaticas, Tesis de grado, Universidad de Costa Rica. 15

Mintzberg, H. (1993), Structues in Fives: Designing Effective Organizations, Prentice Hall, Upper Saddle River, N.J. Mitchell, M. (1998), An Introduction to Genetic Algorithms, MIT Press, Cambridge, Massachusettes. Moe, M. (1995), KLIBAS – The DNMI Climatological Database System, DNMI Report no. 22/95 KLIMA, Oslo. Nelson, R.R. and Winter, S.G. (1982), An Evolutionary Theory of Economic Change, Belknap Press, New York. Poppendieck, M. and Poppendieck, T. (2003), Lean Software Development: An Agile Toolkit, Addison-Wesley, New York. Sherkenbach, W.W. (1991), Deming’s Road to Continual Improvement, SPC Press, Knoxville, Tennessee. Vejen, F. (ed.) (2002), Quality Control of Meteorological Observations: Automatic Methods used in the Nordic Countries, DNMI Report no. 8/2002 KLIMA, Oslo. Waldrop, M.M. (1992), Complexity: The emerging science at the edge of order and chaos, Simon & Schuster, New York. Wheeler, D.J. and Poling, S.R. (1998), Building Continuous Improvement: A Guide for Business, SPC Press, Knoxville, Tennessee. Wikipedia (2009), Genetic algorithm, http://en.wikipedia.org/wiki/Genetic_algorithm, accessed on January 29th 2009. Øgland, P. (1999), A guide to the KLIBAS document trail 1991-1999, DNMI Report no. 09/99 KLIMA, Oslo. Øgland, P. (2000), “Quality Assurance of Data Flow and System Management at DNMI”, in: Moe, M. (ed.) Proceedings from Workshop of the European Climate Support Network (EUMETNET – ECSN), Oslo, 11.-12. October, 1999, DNMI 05/2000 KLIMA, pp. 56-62. Øgland, P. (2008), “Designing quality management systems as complex adaptive systems,” Systemist, Vol. 30, No. 3, pp. 468-491.

16