Future: Books and Managing Complexity (with computers) Date: , Page ©Copyright 1996, Rainer von Königslöw. All rights reserved.
15 York Valley Cr., North York, Ont., M2P 1A8Tel.: (416) 410-2202, Fax: 489-4688
Books and Managing Complexity Draft: for review only Rainer von Königslöw Version 0.1: June 14, 1997 Copyright 1997, All rights reserved. The future of books: trends Integrate money flows and credit reporting more elegantly - improve reporting to individuals, households All income, expenses directly reported into individual accounts, but with better privacy and secrecy protection trends: books less static, more immediately connected to symbolic reality, physical reality through sensors, faster changing, more integration and connection to other information New paradigms Interactive books video games electronic banking computer aided learning Integrating different types of information multi media - MacroDirector animation - PowerPoint presentations How can we re-design and re-engineer user's tasks? (CSALT '94) Four examples illustrate how some user tasks could be re-engineered: Electronically supported telephone and video conferencing Let us assume a lawyer is on the phone with a client and would like to investigate a point of law. In such a context speed is of the essence. One cannot hold a client on the phone while paging through books. Ideally one should be able to look up relevant information with very few keystrokes and display it on a convenient monitor. There should also be confidence that the risk of missing relevant information is as low as desired. If several disparate chunks of information are necessary, one should be able to display them in separate windows, possibly on multiple monitors. Library alerts for work where legislation or a case has changed It would be useful to receive an alert from the electronic library that a document needs to be revisited because there was a change in the environment such as a change in the legislation or a reversal in a case on appeal which materially affects the matter at hand. Let us take a tax plan as illustration. Let us assume the plan is expressed in an electronic book with explicit references to narrow chunks of information which are time stamped. Let us assume that the references go through the central catalogue. Let us further assume that an update is acquired and integrated into the catalogue, the update has newer time stamps for some of the relevant information (the information has changed). Let us further assume that there is a program which sweeps through all internal documents to identify references to changed information (on the update disk). Authors of such documents can be notified automatically to decide whether they need to review the document; want to keep the reference to the previous chunk of information; or want to change the reference to the now current information. Work in progress A deal team is working on a very large and complex document. The essential logic is captured in a skeleton which functions somewhat like a catalogue or index to the document. As the members work independently on different segments of the document, the cross reference to key assumptions can be verified (and time stamped) to ensure that another member of the team has not introduced changes in key elements of the skeleton. Tedious and expensive rework is prevented. Taking work home or to a client's site One can have remote access to an electronic library and to one's files through modem communication (with call-back modems high levels of security can be maintained). Alternatively one can copy the relevant information onto a recordable CD-ROM. The latter approach is secure and does not require the firm to maintain a bank of high speed modems. One CD can store the word processing documents as well as a number of source texts (if they are available as electronic books). How do we get there from here? All the above examples can be accomplished with present technologies. Some of them will be clumsy until more standards and more mature technologies are in place. To develop in-house publications or an in-house electronic library an organization needs both library management skills and project development by people that have the skills to deal with software packages such as Folio Views. It may develop skills in-house by hiring in or with training courses. It may have the publications produced externally by companies specializing in the field. A hybrid approach is often the fastest and most cost effective, where the first publications are done outside, but with the active participation of in-house staff in apprenticeship mode. Major consideration: Information representation and mapping measurement theory numbers descriptive text emotive, evocative instructional text, how to equations programs interpretive speech acts ambiguous Major consideration: Cognitive Ergonomics Introduction Much of ergonomics deals with the physical relationship of people to environments, including computers. In the relationship of people to computers, the main emphasis has been on the user interface. The physical component of the user interface involves objects such as keyboards, monitors, and a mouse. The physical aspects may deal with lighting, keyboard layout, relationships of hands and eyes of a seated person to the keyboard and monitor, etc. There is an important cognitive component to the user interface, dealing with ease of understanding and the relationship between the requirements of the computer and the skills and mental models of the user. The approach I am developing here might be called cognitive ergonomics. As part of this approach I will to broaden the concept of user interface to include not only the relationship between an end-user and an information processing application, but generalize it to the relationship between management and other people and abstract objects which may include MIS departments and source files for computer programs. All of this will be developed below. Let me start with a brief perspective on information processing. Computing has evolved over the years, and is continuing its evolution. My own involvement only dates back to the late fifties and early sixties, so I shall start there. In that period I recall a fair bit of diverse computing, where problems were solved with a combination of approaches that might involve card sorting as well as digital computing, and even include analog computing. Later, with the strengthening of digital approaches, computing seemed to get relatively more homogeneous with the development of large applications. In the last ten years or so, with the advent of relatively inexpensive minis, workstations and PCs, problem solving again became more diversified. It might involve databases on a mainframe, communications to a smaller system such as a PC, and then use of spreadsheets, wordprocessing, and other diverse application programs. Two trends may signal an increase both in the rate of change and in the eventual complexity of the process: the increased use of distributed computing, and the increased involved of non-computing experts in application development such as with the use of spreadsheet macros and expert systems. I believe that the limiting factor to the use of computers is neither a limit in scope of applicability nor a cost or access factor. I believe that the limit is due to limits in our ability to understand and to manage both information processing and change in information processing. It is this general issue that I wish to address. Ergonomics applied to abstract objects Ergonomics traditionally is the study of the relationship between people and objects in the environment. I would like to extend the concept to consider the relationship between people and abstract objects, including macroscopic objects such as "information processing" in an organization. In other words I suggest treating information processing as an object so that ergonomics can be applied to it. "Information processing" is an entity, an abstract object to upper management. Ergonomics might then apply "to improve the user interface" to information processing, so that it becomes more manageable.. They have to understand what they hear from and about information processing in the organization, and they have to know which "buttons to push" to make it work properly for the organization. To treat information processing as an object we need a model. We shall start with an analogy. Information processing "processes" information. Let us extend the analogy by comparing information to a packaged food, and information processing to a chemical processing plant. Later on in the discussion we shall look at information processing at a lower level of abstraction. We shall extend the analogy to identify people and roles. We shall look at some alternative approaches and explore what ergonomics might have to contribute. The role of ergonomics Cognitive ergonomics suggests that people have to have a mental model that they can understand clearly. Above we have suggested the use of analogies to provide such a model. Upper management is primarily involved in monitoring and in providing objectives for project management. Use of an analogy suggests that reports to upper management, and decision options be structured and presented in a form to fit the model. If the model is seen to have inadequacies, it is suggested to construct a second model, based on a different analogy. The models can then be compared and contrasted. It is suggested that the models allow people to place specific issues in context. Analogies for information packaging and for information processing A packaged food analogy for information Before getting to the main analogy, I would like to briefly introduce the use of an analogy to raise questions and to act as basis for a model. For fun and entertainment, let me start with a small analogy that treats information as analogous to a food product. Information is like food, some occurs naturally but most is manufactured Usually we only see the end-product in its package. We can evaluate it by its packaging and by its taste, but it is difficult to know the contents. There is no "truth in packaging" law to label the ingredients of information There is no "Food & Drug Admin." (U.S.) to test for safety and risk of repeated exposure to particular information small exception: TV violence Information goes stale, but it is rarely dated for expiry Information is trendy, more is available on current topics: collapse of communism ... recession ... the war ... Like groceries, most information is not consumed directly, but combined with other information for further processing. The analogy raises questions. We are invited to visualize information as a nicely packaged sausage in the grocery store. Using the analogy we can explore attributes of information that might be overlooked in the normal context. A chemical plant analogy for information processing The main analogy we are proposing below is somewhat more complex, but has more points of comparison. In fact, an analogy of information processing with a conveyor belt manufacturing operation such as a car plant might have even greater similarity, but then the manufacturing operation becomes too complex to provide a clear enough model for our purposes. 1 Information processing as manufacturing process 1.1 Manufacturing information in a large organization might be seen as analogous to a chemical plant, with information flowing through pipes, being combined and modified, and being stored in vessels. 1.2 Information has to be fed in, from suppliers. Some is entered "just-in-time", while other information is stockpiled with delayed entry. 1.3 There needs to be quality assurance (receiving inspection) on information entered into the process. 1.4 The processing can be automated either in part or completely 1.5 There is a risk of things going wrong, and there is risk of human error. 1.6 If programs are like reaction vessels, pipes are I/O streams connecting programs, and databases are storage vessels. 1.6.1 Data flow diagrams play a similar role to PFDs (Process Flow Diagrams), and to PIDs (Piping and Instrumentation) Diagrams. 1.6.2 There is an assumed temporal ordering, where information flows into the inputs and flows through subsequent processing stages until it flows out the output. 1.6.3 Traditionally the processing is viewed as unchanging, just like the reaction vessels, and the information flows through the programs, from input to output. 1.6.4 In reality, even in chemical plants the processing is changed over time - engineers are always improving the processes. 1.6.5 By changing the process, the output may be changed, which in turn affects the next process -- so that changes can propagate through the plant, with no definite scope of how far the change will propagate. This means that "subsequent" or "downstream" processes have to be adjusted. 1.6.6 Large software systems are at least as complex as large chemical plants. 1.7 Large plants are usually divided into smaller plants that act as suppliers and customers to each other 1.7.1 one plant delivers feedstock to another plant, i.e., produces intermediate products 1.7.2 Intermediate products may be held in large holding tanks between plants 1.7.3 In information processing one can usually identify intermediate information products, and subdivide on that basis. 2 Process control for manufacturing information 2.1 In continuous chemical processing, process control is concerned mainly with adapting the processing to variations in the feedstock and to variations in the processing environment. 2.1.1 Variations in the feedstock include variations in the chemicals that enter the process, including impurities, relative concentrations, and including physical characteristics such as temperature and particle size. 2.1.2 Variations in the processing environment can include external factors such as temperature, humidity which affect heating and cooling. It can also include internal factors such as clogged filters and the efficiency of pumps. 2.1.3 Temporal order in the processing is "built in", i.e. it is determined by traversal of the chemicals through the piping & vessels. 2.1.4 Most process control systems depend on local, negative feedback controls. Feedforward controls and large plant integrations have been difficult to archieve. 2.2 The main process control for information manufacturing is temporal ordering. Much of this is automated in a secondary layer of information processing. Handling exception events is where the human operators become involved, much like in chemical plants. 3 Changing the manufacturing process for information 3.1 Updating information processes in an organization is analogous to updating the process in a chemical plant 3.2 The process in plants is not static - process improvements are invented and implemented regularly. 3.2.1 Typically this involves a sequence of invention & development, pilot testing and modifications (possibly in a pilot plant), then integration in the target plant. 3.3 Piping, vessels, and instrumentation such as actuators and sensors are changed regularly in a continuing effort to improve productivity and quality. 3.4 Specialty engineering facilities and plants are set up to research improvements and to produce the piping, vessels, and instrumentation to implement it. These may be part of the plant or act as suppliers to the plant. 3.5 The rate of change is limited by how fast changes can be invented, how fast the piping and instrumentation can be built, and how long it takes to implement the change in the plant. 3.6 To implement the change, the process has to be shut down, old piping and instrumentation removed, new piping and instrumentation inserted, and the process has to be started up again. 3.7 The change in process is different from the manufacturing process: building and installing piping and instrumentation is a mechanical process while the manufacturing process is chemical. 3.7.1 In information processing, on the other hand, changing the process still involves processing information: a new program is developed and put into place. 3.7.2 Although not usually done, it is possible to change the process without shutting down information processing activities 3.7.3 It is possible to design self-modifying systems - such as Neural Networks and other forms of machine learning. 3.7.3.1 In some cases these systems return to the same state with equivalent input. One might think of the learning as adaptive and reversible. In this case they act like systems with integrated process controls. 3.7.3.2 In some cases these systems do not return to the same state, even with long sequences of equivalent input. In this case the "learning" may be irreversible. If the nature of the change is not fully predictable and not easy to monitor than these systems are hard to manage. 4 Project management for the change in manufacturing process 4.1 Project management for change projects plays the same role as process control for processing: it orders and controls the process. 4.2 The invention and development component usually involves repeated operations such as experiments and successive approximations. There is reasonable project management and process control for each of the components, but much less control for the overall project. 4.3 Sometimes the invention generates major challenges for process control, such as when new sensors have to be developed, and new ways of controlling a process have to be invented. In some cases the invention focusses less on the novelty of the process and more on the novelty of the process control. 4.4 The pilot testing and modification cycle again involves repeated operations. It is somewhat more predictable and controllable than the invention process, but there are still uncertainties how many cycles will be required until all the problems are solved. 4.5 The implementation phase is under good project management - with methods developed from years of experience and tools from the building industry etc. 4.6 Software development follows a similar cycle, involving invention, pilot development and scale up implementation. 4.6.1 Prototyping is similar to pilot plant experimentation. The challenge for cognitive ergonomics There is an important cognitive component to the user interface, dealing with ease of understanding and the relationship between the requirements of the computer and the skills and mental models of the user. Information to people & information from people In an earlier section we identified end-users as playing two roles according to our analogy: information suppliers and information receivers. We shall focus most of the subsequent discussion on people as information receiver. This is an area where much can be done with current technology. When people are the information suppliers, they depend on the capabilities of the computer as information receivers. One relevant research topic is natural language understanding - both in written and spoken form. A second is computer understanding of graphs and pictures. In both areas progress has been slow and difficult. The net result is that people still have to adapt to computer capabilities and limitations to function as information suppliers. In the discussion below we shall therefore focus mainly on people as information receivers. In this section we shall focus primarily on identifying the challenges. General topics Form & content Presenting information requires both form and content. A sequence of numbers is a good example. The information can be presented as a column of numbers in a spreadsheet, or it can be presented as a graph. The content is the same, but the form of presentation is quite different. Meaning & ambiguity There is a variety of ways of extracting meaning from information. For example, a message to the operator of a chemical plant may indicate to him that the process needs tuning, possibly by raising a set point. The message may well be extracted from a large number of tags values over several minutes. In other words, the message represents an information reduction. However, the message may only indicate the problem, and several control strategies may be available. In other words, the present message may well be ambiguous. In an extended system, a more explicit message may suggest a control strategy as well as an action sequence to implement the control strategy. In this case there would have been information expansion after the information reduction to reduce the ambiguity in the meaning of the message to the operator. Speech acts & other analogies Role related topics Running the process: End-users The requirements are distinct for different end-users in distinct roles. In general one has to analyse what they have to do with the information. For example, the operator of a chemical plant has to evaluate the information as part of a decision to implement a different control strategy. The eventual action may be: to do nothing and to ignore the message to "tune" the process by making adjustments to request some filed maintenance to shut down or partially shut down the process Running the process: Operators Improving the process: Engineers and others Improving the process: project managers Ergonomics in the large: upper management overview To restate the perspective suggested by cognitive ergonomics, it is important that people in upper management have a clear overall perspective, and that this perspective is most easily gained with a model. To present a clear model of "information processing" to upper management, we need to identify the core components of the analogy. The main components of a chemical processing plant are as follows: 1. feedstock input from suppliers 2. product output to customers 3. a facility such as a plant 4. chemical processing in pipes and vessels 5. chemical storage in storage vessels 6. process control with operators 7. process maintenance and improvement projects - with engineering resources 8. project control over engineering projects 9. project input: inventions, research & development, pipes, vessels, instruments - from project suppliers The corresponding elements in information processing are: 1. data and information input from suppliers (end-users) 2. information output to customers (end-users) 3. a facility, including computing resources 4. information processing in application programs 5. information storage in files and databases 6. job control & task control (process control) with operators 7. process change projects - with engineering resources 8. project control over engineering projects 9. project input: inventions, research & development, programs & tools - from project suppliers Information processing inside an organization may be separated into fully independent operations, or it may be partially separated. If partially separated we can list the connections. Understanding the connections is important for understanding how the operations may impact on each other. external connections: 1. common suppliers 2. common receivers 3. common process change suppliers internal connections: 1. shared processing components 2. shared facilities & resources 3. shared operators - common process control 4. shared process change projects - shared process change resources 5. shared project control - facilities & resources We can now define high level objectives and methods for information processing by extending the analogy, using the components of the model. We shall only illustrate. Objectives: 1. increase product range: more products 2. increase market responsiveness & competitiveness a) quality b) customization c) availability d) more product features 3. lower cost 4. reduce time from conception to availability Methods: reduce costs 1. reduce costs of facilities 2. automate: a) automate processing (mostly done?) b) automate process control - reduce cost of manned operations c) automate process change projects - 4GL, ? d) automate project control - testing, release, ? 3. reduce process change project costs Methods: speed up process change projects 1. reduce testing time 2. reduce development time a) reduce size & complexity of software relative to complexity of information product produced better design better methodology better tools b) increase reuse of previously developed software Ergonomics for the end-user There has been much discussion and development about user interfaces. One question that needs emphasizing is what the user does with the information. The presentation or packaging of the information should be approapriate to the use of the information. For instance, for a process control operator in a chemical plant, there are a number of questions that must be addressed for any message from the process control system. 1. Should the message be ignored - or should there be some action a) can the message be trusted b) what is the logic behind the message c) what is the data behind the message d) what is the probable cause behind the data 2. Should there be a process control action (- or a field maintenance action) a) are there alternative control strategies b) what is the nature of the process tuning action e.g., increase the setpoint by 3% Mid-scale ergonomic analysis This level of analysis is intended to be partway between the "ergonomics in the large" for upper management, dealing with information processing as a whole, and normal ergonomics, or ergonomics in the small which deals with individual people and specific systems and environments. Mid-scale ergonomics identifies people in terms of their roles and it identifies objects in terms of general groups of programs and systems. It then discusses the relationship between the people and the objects. Again, we build a model of the information processing system. We then identify objects from the model and assess their ergonomic characteristics. In most cases the objects have a rough correspondence with system architecture. The main focus of this analysis is on the process change projects. It is this components that tends to present problems for most organizations. Most of the development effort tend to be spent on software. Software development tends to be slow, expensive, and not always produce the desired results. There will also be a brief look at regular information processing, to see what contributions this kind of approach might make to more traditional questions of user interface design. We shall start by identifying the types of people involved in the relationship and then focus on the objects. The people There are four classes of users defined by the four streams identified in the analogy, end-users, operators, engineers, and project managers. Running the process: End-users I would like to differentiate between two general subclasses of end-users, casual or incidental end-users and part or full-time end-users. Casual and incidental end-users includes all those to whom the computer usage is only a small part of the job description, such as: customers, e.g., bank ATM users (automated teller machines) sales clerks on computerized sales stations managers, accountants, and other jobs & professions Part and full-time end-users include all those for whom computer usage is an important part of the job they do: data entry operators bank tellers operators in chemical plants secretaries many authors (e.g., writers of novels) To follow the analogy to a chemical plant, the end-user often has two roles: he is the supplier of information. He provides the feedstock, the input into the information process. he is the receiver of information, the customer of the information processing. Quite often in information processing the same end-user is both the supplier and the receiver. Running the process: Operators In chemical plants the sequences of processes and most of the detailed nature of the processes is predetermined by the engineers who designed the plant, and is not under control of operators who run the plant. The operators just deal with process tuning and exception events, such as equipment failure and routine maintenance. In information processing applications one cannot always find the euivalent of operators. Sometimes the processing is fully automatic - the sequence of operations and the detailed nature of the processing is predetermined by the software engineers who designed and implemented the information processing applications. In other cases the end-user may also monitor and control the sequence and the nature of the information processing. In large operations one is most likely to find operators who have an analogous role to operators in chemical plants. This is especially true if both central processing and distributed processing are involved, and where there are dedicated computer centres. Improving the process: Engineers, information specialists, and end-users Engineers: computer programmers Some information is done with standard commercial packages on standard equipment in standard configurations. However, most even modestly complex information processing operations are custom designed and configured. Also, most of them are maintained and improved on regular schedules. To do this engineering, most medium sized and larger organizations maintain a specialized engineering group to maintain and improve their information processing. Information specialists: business wise but computer illiterate Improvements in the information processing is also based on the expertise of other information specialists who may or may not be computer literate. Many of these are specialists in the various business domains that are covered by the information processing application. This includes people from marketing, manufacturing, distribution, administration, and technical specialists. End-users End-users also play a role in program development, to adapt the application to their needs. Traditionally they played a more passive role, unless the end-user was an engineer or scientist. Lately, however, with more experience and the development of better support, end-users have played a much larger role. Improving the process: project managers Program development needs to be managed. This management role typically falls to computer specialists who have some experience in development. Sometimes the role is played by more traditional managers. Objects for a mid-scale ergonomic analysis The main focus of this analysis is on the process change projects. It is this components that tends to present problems for most organizations. Most of the development effort tend to be spent on software. Software development tends to be slow, expensive, and not always produce the desired results. According to our chemical plant model there are four major components: 1. There are external suppliers for such projects 2. There is a project for producing the new processing components (usually software) 3. There is project control (includes management and quality assurance) 4. There is an event, the eventual change-over to the new process Process change projects based on C or Cobol It is assumed that managers, information specialists, and end-users understand and express the requirements and objectives for the process change project in documents and meetings, and that programmers then work from the understanding they develop from the document and the meetings. Model 1: Development in a stardard language, e.g., C or Cobol We have 5 major types of objects 1. Documentation of requirements, including specification of interfaces to other programs and to data 2. Text files with source code for the program modules 3. A compiler for the language 4. Executable objects and other intermediate objects object files, libraries, etc. 5. Data files for testing the executable objects in a testbed Model 2 - development in a standard language with reusable source code We add 1 type of object 6. A collection of text files with source code for "generic" modules which can be adapted Model 3 - development in a standard language with extended system and other services We add 1 type of object: 7. Interface specifications for various system and other services. MS Windows and HP New Wave illustrate this type of model for development aimed at the PC. Ergonomic considerations for models 1 - 3 From an ergonomic perspective, all three models have major common elements and constraints. 1. The major ergonomic constraint is based on object 2. Only a trained programmer can work with object 2 because the language is quite different from English in syntax, paragraph structure, vocabulary, etc. This prevents most managers, end-users and information specialists from participation in specifying this crucial component in the project, and mandates a dependence on trained programmers. 2. Given the constraint above, object 1 now becomes crucial in the project. Object 1 must be understandable to all parties involved in the process change project, and it must carry most of the burden in generating a common and complete understanding of all aspects of the change in process that is required. The problem is that most documents are either incomplete and ambiguous, or they are so verbose that they become confusing. 3. Objects 3, 4 & 5 are involved in fairly mechanical processes and therefore are not quite as crucial to the project, assuming adequate documentation and adequate skill of the programmer. 4. The ergonomics of objects type 6 can be crucial for determining whether software reuse adds to programming efficiency. If the source code is well structured and well documented then trained programmers can adapt them quickly and thus save time over building from scratch. 5. Judging by the difficulties programmers have had in moving applications to Windows, even with development kits, it appears that the ergonomics of APIs and similar system services are not yet adequate. However, it is often difficult to separate ergonomic factors from basic complexity. I suspect that the problem is in the lack of clarity of the underlying model - which is an ergonomic problem. Process change projects incorporating expert systems In this approach it is assumed that managers, information specialists, and end-users can express some of their objectives directly in a form that can be interpreted by computers. Model 4: Development using an expert system with a knowledge base The concept developed here should be differentiated from the use of expert systems to solve very complex problems. What is suggested here is the use of expert systems technology as a 4GL which allows a more natural representation of the logic underlying the information processing. Since we shall later suggest integrating this with models 1 - 3, we shall continue the numbering of objects. We have 4 major types of objects 8. Text files with knowledge base modules. These knowledge base modules represent the logic behind the information processing in rules and other near-English expressions. 9. An interpreter/compiler for the expert system shell 4. (as above) Executable objects and other intermediate objects object files, libraries, etc. 5. (as above) Data files for testing the executable objects in a testbed Model 5 - development using a spreadsheet Conceptually this is very similar to model 4, the expert system. Again there are two components, a spreadsheet (which is similar to a text file) containing the application logic in the form of functions and macros, and the spreadsheet program. This model is particularly useful for simple simulations involving numerical data, and for displaying standard graphs. 10. A spreadsheet containing the logic of the program, along with macros to help execute the program. 11. The spreadsheet application program 5. (as above) Data files for testing the spreadsheet & its macros Model 6 - development using wordprocessing with macros Again, this is very similar to model 4 and 5 above. This model is useful for automating the production of individualized boilerplate letters. 12. A wordprocessing document containing the logic for the application expressed as boilerplate macros. 13. A collection of paragraphs in an appropriate wordprocessing document or documents containing the boilerplate content. 14. A wordprocessing system 5. (as above) Data files for testing: producing sample boilerplate documents with the wordprocessing system. Model 7 - development using other application programs with macros or scripts There are other application packages that have similar features in allowing the user to execute a text file which lists commands to the package. For instance, there are communication packages that support script languages which allow the user to automate the communication. 15. A text or similar file which contains the commands to the application package. 16. Possibly a collection of other files used to control the system. 17. The application system 5. (as above) Data files for testing Ergonomic considerations for models 4 - 7 From an ergonomic perspective, all four models have major common elements and constraints. 1. Objects 8, 10, 12 & 13, and 15 & 16 allow information specialists and end-users to contribute directly to the development of the information processing system without having to learn about programming. The ergonomic qualities of the object depend on the particular application package, but it could normally be assumed that only those applications would be chosen where the information specialist or end-user has prior training and experience. 2. The main benefit of the four models is that they remove the dependency on the programmer and on object 1. Even though a programmer may set up the model, it is assumed that the information specialist, manager or end-user can a) inspect it directly, and understand what it does b) make minor changes without programmer assistance c) test it directly with pertinent data 3. Objects 9, 11, 14, and 17 are assumed to be commercial systems, but they might be developed or adapted in-house under models 1 through 3. The ergonomics depend on the package, but competition allows the selection of an adequate system. 4. Development and acceptance can usually be focussed on one person, i.e., there is no requirement for a large team effort and for team coordination. That improves the ergonomics of decision making and speeds convergence to the desired system. 5. There seems to be a limit in the complexity of information processing that can be handled with this type of model. For expert systems the limit seems to be around 450 rules. There seem to be similar limits to the other applications, reflecting the capabilities and the scope of control that can be exerted by a single person. 5. A complete application package will normally have to be composed out of a number of modules developed with some combination from models 4 through 7. This requires application configuration (to be discussed below) which adds to the complexity of the application.