-6.1- IS AI IN TEE MODERN CORPORATION IRRELEVANT OR INDISPENSABLE
In spite of promising to alleviate many problems, AI in management is a controversial issue. The controversy revolves around the degree to which management skills can and/or should be replaced by an AI system. Some proponents believe that AI systems will replace management, others believe that it is not relevant to management. Since AI appears to some as a threat, it is surrounded by emotions which may inhibit exploration of the potential of AI as a new management tool. The most appropriate way in which to view AI technology is as a decision support tool.
The strong case for the use of AI techniques in strategic management is made by Holloway [28]:
No part of a Strategic Planning system appears to be beyond the scope and capabilities of an emerging Artificial Intelligence concept... It is not too early for executive management to begin detailed consideration of the impact of Artificial Intelligence and how the corporation should react to it.... In the ultimate.... all decisions are made by AI. Directives would be issued by a figurehead.
The case for the irrelevance of AI technology to management is made by John Dearden [15]:
I am assuming that computers will continue to do what they have.., store and retrieve data... I do not know of one instance of artificial intelligence that has important practical application for top management. Should top managers use personal computers? I believe that the job of the top manager has not been affected significantly by the computer... The greatest impact of the personal computer will be on managers of small businesses. These managers make many line decisions and can use a computer in such applications as inventory control and other logistic operations...(and) tend to be limited in both the quality and quantity of their staff because of financial constraints...(and) use the computer for analysis as well as retrieval of information.
Both positions seem to be overreactions to the potential of AI in strategic management. The issue appears to rest on some assumptions about computers and how they can be used in decision making.
Alter feels that "rather than technical innovation, the key issue for decision support systems in the next decade may simply be the diffusion of the idea that computers can be used to support (rather than replace) decision makers." [3, p.192] Thus, it is not the technology itself that must be developed, but how the technology is to be integrated into the decision making process to best support the decision maker.
AI systems are currently quite limited and may be sufficient for many specific problems, but not for the common sense knowledge that is needed to use and update information. Thus, based on the current size and speed of computers, Dearden is correct; but he does not recognize the similarity between expert systems and the human experts who are frequently consulted by top management. The intention of the expert system is to provide the expert advice, which may be rare and expensive, at a lower cost.
The much talked about fifth generation computer may radically alter the way decisions are made, in the same manner as previous innovations in information technology have affected decisions. The conceptual framework and the technology are available to create computers of the necessary size, but the implementation of the fifth generation computer may take over 10 years to accomplish. We have some time to prepare for the changes. Since managers are the least flexible components of a corporate decision system, this lead time should be used to prepare them for the new technology in some incremental way. A personal decision support system is perhaps a step in that direction.
There is no indication that AI will replace all decision making as Holloway suggests. AI theorists admit that the human being may in fact be the optimal tradeoff between the detection of rules and the acting on those rules. The simple example often cited suggests imagining how to program a computer to cross a busy intersection. The enormous number of instructions required would get in the way of a rapid response to the environmental conditions. Yet even a small child can cross a street without difficulty. Hence it appears most accurate to call AI an aid to, not a replacement for, human judgment. The balance between common sense and specialized knowledge is a dynamic balance in any form of intelligence. Initially, MS/OR professionals (who were in themselves expert systems) were treated with suspicion; but they have hardly replaced top management. Instead, they provide a broad variety of services. A similar evolution for AI can be anticipated.
-6.2- STABILIZING CORPORATE PROCESSES WITH AI
The modern corporation is faced with a rapidly changing environment. The change in the underlying technology continues to accelerate. The corporation is becoming increasingly dependent on long term informal relationships. This dependency increases the productivity loss due to higher employee turnover. The corporation is becoming increasingly process oriented. The process orientation requires focusing on the technological base of the process. The rate of technological change continues to accelerate, which means corporations must either import new expertise from outside the corporation and accept a higher turnover rate or incur the high costs of retraining its own personnel.
The use of the expert system could help stabilize the corporation. An expert system can be used to make highly specialized knowledge available to individuals in the corporation as the need for the specialized knowledge develops. Further, the knowledge can be kept current by a single corporate expert. The same level of expert knowledge would be available to all individuals. The performance of individuals in the corporation would be easier to monitor since the output would be based on the same inputs. The development of the effective use of the expert system could replace some of the more subjective evaluation criteria. This would stabilize the corporation by providing an easy method of importing new technology without causing employee turnover.
An NL system could also be used to further reduce the level of expertise needed to utilize the information used by other expert systems or stored in databases that the query maker may not know exist. This would reduce the dependence on informal relationships for obtaining information from other individuals in the corporation.
-6.3- AI APPLICATIONS IN CORPORATE MODELING
The introduction of a corporate model to the corporation is an example of the introduction of a new technology to the corporation. The major problems associated with the success of the integrated corporate model (see Figure 2-1) center on the ability of the users to anticipate, assimilate, and actively use the corporate model. The success of these activities depends on the individuals' knowledge and experience of corporate models. The experience and knowledge level of the individuals within the corporation usually does not encompass the technology of the corporate model, and yet the knowledge and experience are needed for it to succeed. Ginzberg [23] indicates that the early expectations of model users may shape the final success or failure of the effort. Clearly the expertise of the modeler must be made available to the individual manager as early as possible. Gathering the managers, who are usually highly paid, is expensive and can result in lower productivity over the short run. Simple training is inadequate because more sophisticated questions will not arise until the simple concepts are absorbed. Giving the individual managers direct access to the modelers is a possibility, but the scarcity of and mobility of the experts makes this impractical.
The use of an NL system and an expert system could serve to aid in the introduction of the new corporate model, by making information available to model users when the users were ready for the information about the model. The NL system would reduce the level of specific knowledge the manager would have to possess about modeling in order to ask questions about the model. The NL system could also be used to tailor the answers about the model to the meanings the individual managers associate with different terms (e.g., "sales" has a different meaning to production, to the legal department, and to marketing). These differences must be taken into consideration by a DBMS which is used by individuals with different views of the corporation.
AI can facilitate interactive planning. Interactive strategic planning was described by Spronk [51] as being the most appropriate planning style in the uncertain climate we now face. When comparing interactive planning to inactive, reactive, preactive, and proactive, Lahr found that:
Most planning modes practiced today do not create kindredness and common purpose that organizations are trying to induce. Interactive planning provides these, as it involves the participation of everyone in the planning process. [34, p.37]
However, even the modest goal of getting executives to use personal computers is not simple to achieve. From the computing community, Jerry Wagner of Execucom [18] stated:
Will microcomputers bring computing to the executive's desk? ...the numbers will never be significant.... When you become an executive, it is not part of the culture to use these kinds of tools... An executive can work more effectively by working with a staff assistant who will do the model building... the key here is that the executive must be able to understand the work his assistant does for him on the computer.
Dysfunctional side effects of the personal computer have been reported by personal computer users in the executive ranks.
Because we become so much more productive with a computer, we often take work away from others without thinking about the consequences. In the same way, we begin turning out work in far greater detail, or at a faster rate, without thinking about the boss's ability to deal with the increased input. [49]
There are side effects of concentrating functions that are seldom discussed. An example of such an effect is the unavailability of information if an executive is not physically available. Other issues may include the security of sensitive data when it is maintained at multiple locations.
These arguments and others point to the tendency to establish a personal computer staff assistant who works closely with the executive to answer questions, but does not have executive status. This amounts to a decentralized and personalized MIS department with the same conflicts between staff and line management as have always existed.
A solution that appears to be feasible, but only in the long term, is for executives to internalize the functions over time. This would be possible because the management process is an open system. The choice is between reducing the complexity of the problem or increasing the complexity of the manager. As the complexity of the environment increases, the complexity of the manager must increase in order that he can make decisions that take into consideration all of the relevant variables. The resistance to using the computer as a decision aid is a matter of cultural identification and will change as the cultural image changes. The popularity of personal computing advertisements which spotlight executives using personal computers can do much to change the cultural stereotype.
-6.4- OTHER EMERGING TECHNOLOGIES AND AI WORKSTATIONS
Natural language interpreters can reduce the complexity with which managers must deal, although with some loss of efficiency in the system. In large scale systems, the overhead costs of natural language systems have deterred their development. The supermicrocomputer-based PDSS would provide sufficient memory for the development of a natural language interface (to facilitate DBMS access, data analysis, and presentation) and an expert system to provide the user with the knowledge of either external experts or predecessors in the position. In a ring network environment with other similar workstations, the expertise of other members of the corporation can be efficiently tapped; but without the support of an expert system, these experts could easily be overcome with requests for information. It appears that the microcomputer network is only a bridge to the new AI technology supported by data processing managers as a means of decreasing the processing demands made on their mainframes.
In addition to the relatively low cost, the personal computer can easily be expanded in a communications network with other microcomputers. Networks of personal computers can be expanded in modules which allow individuals and departments to be connected or disconnected as necessary. Figures 6-1 to 6-4 indicate how the general corporate model is linked together. Figure 6-1 indicates how the information in the general corporate model is generated. The variables may not exist in the accounting database of the corporation, as indicated by the different types of computations of the D matrix. These matrices, which represent different departments of a corporation, can be linked with other models as shown in Figure 6-2 to form the corporate planning system. An example of how the data is linked together is shown in Figure 6-3. In this figure, the example uses historical accounting data for the sake of familiarity, but forecasted values not based on accounting data can be substituted for planning purposes. Figure 6-4 shows how the D matrix is constructed, maintained and moved between modules. This representation of the general corporate model forms the basis of the AI workstation introduced in section 7.
The advantage of the network over the individual workstation which makes it a likely course of development in the corporation is that the communications connection is almost 1000 times faster between the network and a mainframe computer than between individual workstations and the mainframe. Since the accounting data in the mainframe is useful primarily for historical performance evaluation, the processing of information in the mainframe is likely to be of little importance to the members of the network and frequent access to the mainframe would be likely to degrade the efficiency with which the mainframe can carry out clerical tasks, such as payroll, inventory, or invoicing. The AI networkstation is likely to have locally maintained databases and accessed printers. Since the customization of the workstation depends on the adaptability of the workstation software, we examine this area in this paper.
As the true parallel processing (fifth generation) machines are developed and the amount of memory located at each workstation is expanded, the general data processing functions will be relegated to the AI workstation network. The greater reliability of the ring network over a stand alone machine lies in the fact that if there is a machine failure in the stand alone processor the entire system stops, while failure in an individual processing station (node) of a network results in only a slight change in efficiency. The AI workstation is a key element in the transitional technology towards the full scale distributed network which has no centralized computer facility and dynamically allocates the processing functions among the available processing nodes.
The functions for data analysis in the PDSS could be written in a programming language which would allow the individual manager to rename these functions so that it has some meaning to him. The tailored functions should be resident in both the PDSS and in the OEM so that they can be called from a utility library if they are not available in the individual manager's PDSS. As the manager's knowledge of the capability of the system increases, his demands increase. As the complexity increases, the PDSS can be reopened by the systems analyst so that he can add the specialized functions and compact the utility functions that the manager is now beginning to use. Thus, the evolution of the PDSS can follow the more natural development of the individual manager's needs as they grow over time, while not compromising the full capacity of the system. This is an important benefit of the modularity of the microcomputer network.
An expert system can reduce some of the implementation problems experienced in implementing OEM projects. Grinyer notes that sufficient documentation is a problem in corporate modeling because the developers, programmers, and maintainers are frequently promoted. (Alter also lists this as one of his eight factors affecting DSS success.) Without sufficient documentation, the OEM will fall into disuse. The essential functions in the manager's PDSS must be transportable and capable of being integrated with a new (or vacated) station. The functions renamed by the manager can be maintained on a cross-reference list that matches them with the functions in the CBM utility library. By comparing the cross references, one can keep duplication of functions in the PDSS to a minimum. Knowledge systems specific to the task could retain some of the expert experience of the previous holder of the position and hence eliminate the loss of the special knowledge of the CBM system when an incumbent manager leaves.
The following section is an example of a personal decision support system in LISP (LISt Processing language), designed to run on a personal computer. LISP is the traditional language used by the AI community and so was chosen for this project. This personal decision support system incorporates some natural language handling ability and provides a consulting expert system on methods of problem analysis. This expert system is based on the taxonomy of Macrimmon. Although this taxonomy does not directly address the problems of successful DSSs suggested by Keen, it does provide a support system that allows the user to chose a method which he has not used before and it provides an illustration of how the new technology might be used.
In general, DSS should be a personalized, flexible system that includes some natural language interface which is capable of producing computations and, at the same time, can act as a consultant to aid in analysis with which the manager may be unfamiliar. The degree of natural language capability and the scope of the expert system depend on the hardware and software available.
The intention of the implementation is to demonstrate the utility of the approach. Much more sophisticated natural language and expert systems exist, but they are not available on personal computers. To make the AI technology effective in the working environment, some method of transition is needed to the more sophisticated systems. Thus, the following is an example of what currently can be provided to managers to help them understand how to effectively utilize the new technology without the threat of a totally unfamiliar and complex technology. This is the human factor which determined the choice of the personal computer for this implementation.