EXPERT SYSTEMS - CONCEPT, PROMISES AND REALITIES FOR MANAGEMENT OF CONSTRUCTION

HUMAN EXPERTS AND NON-EXPERTS

In the context of any topic including managing construction and put simply, an individual can be thought of as being (a) ineffective because he does not have the knowledge, information, thought processes, etc., to get the job done or (b) effec­tive, i.e., getting the job done somehow or (c) effective and efficient i.e. doing it effectively plus using only a minimum consumption of resources by use of appropriate knowledge, thought processes and requisite information, etc. The above (c) effective and efficient exponent will probably provide a cost/benefit result well above that expected of the normal exponent of that process. The crucial ingredient in producing something better than the others is his additional increment of expertise along with the fundamental common knowledge of that specialized topic.

An expert system implies provision of expertise at level (c) above but it will be very difficult to repLuduce that level of professional expertise using a computer program [11. It appears that current examples of expert systems may really present expertise at levels below that of (c) above. This raises either a profound question of their usefulness or only a question of improper nomenclature.

In a paraphrased format, experts are capable of employing plausible inference and reasoning from incomplete or uncertain data and determining relevance or relative value of information in a situation, as well as restructuring and reorganizing that knowledge, breaking the letter of rules, because they understand the spirit of the rules, and explaining and justifying what they do as well as doing all of the above [2]. Against the above definition of an expert, expert systems are just beginning to model same of the above characteristics [2].

In artificially representing or modelling knowledge based on a topic, the model can be deep or shallow [2]. Deep implies capability with the knowledge as in the above human expert. This can further imply a flexible network made up of pieces of information on the topic and the thought processes of the human expert. Under a particular circumstance or situa­tion, the network is shaped to form a hierarchy representing the confluence between the situation or domain information and the knowledge base and thought processes of the human expert. From this situational hierarchy, an experts solution can be produced.

A very early, crude effort in this direction in management of construction, was to analyze "the building" as cells of an information matrix such that each information cell was analogous to a mosaic chip. Then each managerial process in design, contracting and construction of a generic construction project (or of a specific construction project if it used the information shell of the generic project) could be serviced by a whole or part permutation of all the information mosaic chips in the matrix [3].

THE PROMISE OF KNOWLEDGE BASED EXPERT SYSTEMS

Four Promises

The future promise from knowledge based expert systems can be presented as the dream of computers evaluating situations and making decisions and choices which produce more realistic and accurate results that those currently and frequently made by the average exponent of a specific skill related to that situation and/or at lower cost than that of a human expert in the topic [1] [4]. Also, our con­sideration of appropriateness of knowledge to a situation or problem will becomme more rational after the evolution of expert systems [4], [5].

The computer executing a faster, less expensive but still top quality selection of the cause of a human illness and selection of the most appropriate cure or medication compared to the benefit/cost of hav­ing specialist medical doctors make the choice could be one example of this benefit [2]. Alterna­tively, from the keeping of very careful and multi­faceted statistics on the performance of baseball pitchers and hitters in all major league teams may increase the expertise of a manager of a specific team on how to better minimize the strengths of the opposition and maximize the strengths of his own team and hence increase his potential to win one game [6] .

Evolutions are taking place today towards both of these objectives. Both seek the advantage to the user in a competitive situation by programing generic human expertise in that topic.

Alternatively, the promise of knowledge based expert systems lies in the greater use of computers to move beyond the "numbers crunching" work that most do today and utilize their potential to operate as "logic machines" rather than high speed calculators [4]. The sense of being a logic machine for the use of man leads to developing the use of computers carrying expert systems as fast processing of a massive volume of relevant informa­tion greater than can be achieved by a human being on such a topic in the time needed to make a high quality decision in a dynamic situation. Of crucial importance for success in this direction is that the computer can be programmed according to the ways of thinking of one or more experts in a topic. Thus the computer program could simulate expert thinking and produce artificial intelligence comparable in quality to that of the expert or experts whose thought processes were analyzed, captured and programmed.

A third major promise from knowledge based expert systems is for education of neophytes and non experts. The expert system could be the master craftsmen and the neophyte is the learning appren­tice. The learner could study the expert system as the process by which to carry out the task. He would practice and develop his skills using the expert system just as an airline pilot practices on a flight simulator. While each expert system may not compare to the wholistic learning of the flight simulator there is at least one attempt to create a wholistic construction project management simulation system [7].

A fourth major promise of expert systems is that each can be a dynamic library of expertise. in the future, contemporary experts will retire and die their expertise will die with them after each of them had to slowly and painfully accumulate their expertise over time. Without the guidance of agent such as another expert or expert systems will future experts have to slowly and painfully create their own expertise. Thus, expert systems could be a dynamic repository of expertise to be tapped into by learners and exponents of all levels of expertise.

Two Types of Promises

It can be seen from the above, that the first two promises of expert systems are directed at improving actual worker effectiveness and efficien­cy whereas the second two promises are directed at improving learning of a non expert and husbanding of existing expensively created expertise.

In the short run it could well be that the evolu­tion towards expert systems in managing con­struction will give the greatest payoff towards the latter two promises rather than the use of the expert systems in execution of managerial functions in construction. In the long run, the short run payoffs will remain viable whereas the replacement of the worker/manager by a computer program remains questionable.

CAP OF AN EXPERT SYSTEM

Expert systems constitute one branch of artificial intelligence which is the overall thrust of using computers to simulate human thinking on a particular topic. The other two major thrusts are natural language processing and robotics [8].

Expert systems should be developed from the knowledge base of experts in a topic and comprise (a) known facts about the particular situation which can be seen as input data which defines the domain in which the expert system will be applied and (b) heuristics. These heuristics are the thinking processes of existing experts and comprise rules of thumb, decision making parameters and even the values of the parameters in different situations under the same topic that comprise the thinking processes of the domain experts [4].

The process of use of the expert system is to fix the facts of the situation, feed them into the system and process them through the heuristic rules to produce the outcome or decision that would have been reached by the human expert or experts.

The system should be able to interact with and question the user on the nature of the situation to which the expert system is being applied so that the real world domain "fuzziness" is clarified to the specificity that can be handled logically to process information and make decisions by the heuristic rules [2] [8].

An expert system can be seen as a sequence of activities, information gathering, processing, clarification and consideration along with making decisions based on the information given. [2] Its format is usually in an hierarchical or flow chart format by which the computer carries out the thinking of experts for the non expert user to produce the output achievable by an expert. Especially in the complexity of a management science topic it is probable that the sequence or flow chart program has to use a computer, iteratively or interactively working with the user of the expert system. A more theoretical definition of expert systems is given as an intelligent computer program that uses knowledge and inference proce­dures to solve problems that are difficult enough to require significant human expertise for their solution. Knowledge necessary to perform at such a level, plus the inference procedures used, can be thought of as a model of the expertise of the best practitioners of the field. The knowledge of an expert system consists of facts and heuristics. The "facts" constitute a body of information that is widely shared, publicly available, and generally agreed upon by experts in a field. The "heuristics" are mostly private, little-discussed rules of good judgment (rules of plausible reasoning, rules of good guessing) that characterize expert-level decision making in the field. The performance level of an expert system is primarily a function of the size and the quality of a knowledge base it possess." [9].

Behind each expert system is the choice of the topic or domain and its scope as being narrow and specific or broad and general. Is the system intended to be generic as a general guide from which to create more specific processes under different domain factors or is it to be a totally programmed process to produce a choice or decision under specifio conditions? This multi parameter definition of scope or clarity of objective is very important to successful evolution and use of any expert system. Also establishing what the system will not do or consider as part of its scope is very worthy of expression to neophyte users.

The major parts of an expert system are (a) the domain data, (b) the inference engine and (c) the output.

The (a) domain data comprises the descriptors of the situation upon which the expert system will operate. These could be seen as the situation symptoms or factors to be dealt with by the system. The situations and their boundaries which call for experts solutions are usually fuzzy. Also, the knowledge bases of experts do not have exact or finite boundaries. Therefore, it follows that the expert system should have the capacity to inquire of the user about the situation/domain to clarify the actual situation and its boundary conditions to be as close as possible to the conceptual domain programmed into the expert system. An exception to this is in very narrow well defined situations/domains.

The (b) inference engine is the many heuristic rules drawn from one expert or set of experts on how they think about solving the given problem or situation. It is helpful to consider heuristics in two groups (a) accepted facts and processes common to all domain experts and (b) specific thought processes unique to particular domain experts. For all these to be used by the computer, their combination must be expressed in terms of logic couplings of "if" leading to "then" or "if not" leading to "then" [10] and closely linked by "arid" or "or" in groups of parameters rules and even and values of parameters [4]. Each parameter may be expressed as having a value 'greater than" or "less than" resulting in a different resultant state of the situation. All of these should be arranged in flow chart pattern in the most logical sequence of artificially thinking about the topic. Such a flow chart will be logical expression of the ingredients of the thinking of the experts even if each expert may make what is seen as subjective connections or less than optimum sequence of considerations in his personal thinking process. Each sub program in the flow chart should lead to a "do" something instruction at various points in the flow chart/inference engine where changes are made from one sub routine of the expert system to another.

The inference engine or program is usually structured in one of two ways, backward chaining or forward chaining [4] [10]. Backward chaining starts with an hypothesis and then tests it through the system to choose yes or no as the output. On the other hand, forward chaining is used in sit­uations where there are many alternative solutions to a situation and the objective is to find the best process/outcome from the input, albeit fuzzy. In the forward chaining process the output or result is created or built by the given facts being examined against the heuristic rules of the infer­ence engine. In effect, the input data drives the expert system through a series of iterations of examining the heuristic rules. It would appear that management issues in general and management of construction processes in particular, are suited to the forward chaining structure of the inference engine than to the backward chaining approach.

The (c) output is the final "do" something instruc­tion from the whole expert system. This may be a single simple instruction of what do or it may be the choice that would have been made by an expert in the topic. Alternatively, it may be a com­plicated multi part result, each made up from an array of alternatives to form the best solution to the situation situation for which expert advice has bee.,. sought.

DEVELOPING AN EXPERT SYSTEM

Fruitful Environment

One aspect of man's curiosity has been to simplify both the learning process and the subsequent process of use of knowledge gains. This is compar­atively well advanced in areas of study of physical sciences and is evolving rapidly in medical sci­ences. In physical sciences the evolution of say, structural engineering from the empirical to the contemporary mathematical level, has taken a long duration of time but is now at a level which, when combined with computers, is causing a dramatic diminution of the required numbers of human beings as a structural engineers. In medical sciences the longevity of study of the human body allied to the importance of the desire for continuity of life by humans and the comparative uniformity of the working of the human body has facilitated advances in that science and its use all over the world. The volume of funding for medical research and the reduction of the results of that research to writing combine to create a fruitful environment for the production of medical expertise in a systematic format.

In both these examples the past systematic encapsulation of the requisite knowledge in written format has proceeded, facilitated and enabled a forward movement towards some tentative computerized replication of how an expert in such fields would think or react to a given set of inputs or ingredients to produce at least a tentative conclu­sion or finding equivalent to that as a normal expert if not a super expert. Even with such a written data base available, these expert systems in comparatively physical areas of science, cur­rently only approximate the quality from that of a normal exponent of that science.

Clearly, the more written valid data that exists, the more fruitful the environment for successful development of an expert system.

Strategy

Probably the actual process of developing an expert system will be more ad hoc and interactive or of simultaneous evolution of the parts than presented below because of the interactive nature of the end product. However, the following parts would all be required and if each is developed sequentially, those which are already developed should be modified to ensure the compatibility of the resulting whole expert system.

Probably the development of an array of narrow individual expert systems which can later be built together to form a very useful broad system may have the greatest potential to building a viable and useful strategy of management using expert systems.

As mentioned at the beginning of the paper, there are different levels of human expertise on a topic and a major feature of strategy could well be defining the level of human expertise to be sim­ulated by the expert system. This is related to deciding what existing costs or resources will be replaced by the expert system and how it will relate to other management activities and resources still in place as before.

Setting the Topic

The problem to be solved, opportunity sought or situation to be explored has be to established clearly in all its relevant parameters. The universe of the problem for which the system will be used has to be carefully defined. Setting (a) the boundary of the scope of concern and (b) the degree of detail or generality are of importance and is usually derived from the clarity of statement of the objective of the expert system.

The topic can vary between (a) narrow i.e. specific and precise from the beginning of it use but which risks being trivial or (b) broad i.e. with great potential use but be so interactive or large as to jeopardize successful creation. Alternatively, combined within a broad system can be an array of more specific individual processes subsequent to (i) testing of that broad approach or (ii) a series of uses of the broad approach in specific situations by a semi expert.

Capturing the Expertise

This is probably the most vital part of developing an expert system and it is probably the most difficult. However, it could have the greatest fail/safe potential from the evolution of expert systems.

The system developer has to recognize which people are expert on that topic and how can they be induced to be willing to give their expertise to someone who may use it to make them partially redundant. There may be many experts in a particu­lar topic and each may have unique factors and heuristics and thought processes as well as factors and processes canton to the group of experts. Both unique and common factors, heuristics and processes should be gathered from all source experts.

Given the selection of a set of experts can the expertise increment of each over that of the group be captured along with the fundamental factors of the topic common to each knowledgeable expert? Is direct questioning a valid way or is indirect, open ended discussion better [11] or is a sophisticated Delphi system approach sore useful or are all approaches required to extract the required body of expertise from which can be created the heuristic rules of the inference engine? Use of carefully structured questionnaires appear to be of consider­able use for finding basic factors and heuristics common to many experts but open ended discussions many be more fruitful for finding unique or deeper heuristics of individual experts.

The issues of duration and cost in capturing the expertise by the chosen approach cannot be ignored. Also, the differences in vocabulary of each expert may cause unnecessary variances in the data base of expertise gathered unless careful analysis and distillation of captured expertise's carried out.

The level of knowledge of the system builder on the topic/domain is another variable to be considered. Some such knowledge is useful in the recognition of the relative value of different inputs from the various experts but the system builders level of knowledge may bias the structure and content of the inference engine for the better or worse. This is a major factor in creating questionnaires and carrying out open ended interviews with experts to gather expertise.

Creating the Inference Engine

The inference engine has to be created by combing out the factors and heuristic rules provided by the experts and arranging them in the most logical flow in the inference engine as well as relating each to the required input information which describes the situation or problem in which the expert system has to produce its results. This is the building of the system and it might be best to develop the whole system as a flow chart and test it prior to programming it for execution on a computer.

The process of an expert system requires inputs of knowledge, information, thought patterns, factors and heuristic rules which may have variable values in different situations and appropriateness of consideration of those ingredients. Also, ability to appropriately consider the past and present in evaluating what to do about something that will affect interactions of people and sit­uations in the future may have to be built into the system. To create an expert system all of these have to be gathered, blended and balanced to produce high quality results under all situations faced in the set topic within which the process or decision may occur. The whole system is required to provide results at least better than that of the normal, average human exponent in that topic and if possible emulate or better results by top experts. Achieving this level of quality might well be a major cost/benefit hurdle to be considered prior to starting to develop any expert system.

Testing the Expert System

Testing the system can be (a) by its examination by a number of experts or (b) by trial by use of the system or (c) a combination of both (a) and (b).The testing of the system should be directed at iteratively improving the expert system to the desired level of equality with available human experts in the topic.

Examination can be by the experts from wham exper­tise was received or a second set of experts who were not involved in providing input to the creation of the system. Alternatively, a mix of these two categories of experts could be used.

The system could also be tested on examples of the particular problem or situation for which it was designed while parallel execution by human experts is carried out and the results of the two approach­es compared for similarities and differences and subsequently to use these to improve the expert system.

Training of Users

Once the expert system has been created it remains highly probable that the non experts who will use it will require training. This will require providing full information on what the expert system can do and cannot do, how to interact with the system in reducing situation fuzziness, how to present the final output to its user or the partic­ipant on the next phase of work on the project, along with the mare prosaic aspects of how to use the system.

SOME COSTS AND RISKS IN CREATING EXPERT SYSTEM

Fundamentally, in the creation of expert systems, capital investment is required. It is hoped that when the capital investment to create an expert system is spread over the many tines the process is used in the future that such cost as an increment of each use, when added to the cost of the non-expert using the system, will be less than the current cost of the evolution of human experts in that domain and one of them making his best judgment on the problem being faced. In the above cost equation it is assumed that the quality of the expert decision by the system and the human expert will be the same.

The more sophisticated, more completely encompass­ing is the desired scope of the expert system to be produced, the higher the probability that a greater capital investment will have to be made. Furthermore, the larger the system to be produced, the larger the probability that it will take longer to produce and the larger the probability that the system produced will be incomplete in its contents. Simultaneously, in the world of expertise in a particular subject the evolution of the subjects data base or demands of the marketplace will likely change over time. This is especially true in the fields of non deterministic sciences such as management within which falls the topic of management of construction. Thus the capital investment in creating the expert systems produced should be evaluated carefully as to the probability of its ultimate economic success in use in the future.

Major questions to be considered in creating an expert system can encompass the following but there may be others to ask as well. Who will create the expert system by consuming the capital investment? Can someone who is a non-expert in a topic, but may be an expert in computers or artificial intelli­gence, create such a system to be better than an expert in that topic? Does it require the expert or experts in that particular topic to create the system? If more than one expert exists does it require all experts to be consulted or can this be done by the pooling of expertise of some of the existing experts in the topic? Who will carry out such pooling of the diverse approaches of the experts? Will the process of pooling of diverse expertise destroy the expertise of each individual expert? Does the person pooling the expertise require knowledge of the subject to interpret what each expert says or does?

Other broader questions to be considered are - what is the life span of such an expert system in a field of human expertise which is not finite like a physical science but is dynamic and probalistic as in a science dealing with human thinking, i.e. management of construction? What if the dissimi­larities of different segments of the construction marketplace cause them to put different values on the same parameters of the expert systems being created for subsequent use in all these different segments of the marketplace? In addition, there is the risk that in the development of expert systems so much attention is paid to developing the compu­terized system that less than appropriate time and mental energy is given to establishing the factors, thought processes, heuristics, variables and values of the experts and their interactions in that topic.

Another segment of concern is that the educated population of the world is increasing and the work force of most developed nations is evolving from producer type work to service type work. It is the volume of this latter service type of work that will be diminished by use of the expert systems by programming of the thinking services provided by humans thus displacing work from humans to comput­ers. The increasing availability of educated humans could impinge negatively on the initial cost/benefit calculations to decide to (a) develop expert systems and their mode of use, dared to (b) continuing the normal evolution/education of human experts of either as is now practiced or assisted by the recorded expertise of experts.

Above all others is the major concern in creating expert systems is the probability that the system resulting from such a capital investment of research and development will be at least feasible and may equal or even be better than human experts or ordinary exponents at doing what the system is required to do.

Furthermore, with such a complex system being created through such a maze of risks and being used subsequently by a no.-expert to carry out work currently only trusted to an expert raises the always present spectre of Murphy's haw that states that if something can go wrong it will go wrong. Then it is not the inadequate human user who will be blamed for the failure but the "expert system" itself will carry the criticism. Then the expert system may be cast aside as useless. Then the search begins for the next panacea rather than continuing to steadily work to evolve a better expert system from the embrionic one.

Counter to these somewhat specific doubts about the potential of expert systems are the overwhelming forces of human curiosity and desire for improve­ment and that such systems can be built by a few people who have the faith and dedication to achieve their goal and reap their benefits from many people using what they have created despite the rational odds against them.

In the development of expert systems it would seem sensible to establish the factors, thought process­es, variable and values of the experts in that specific topic as a precursor to the creation of the computerized or written vehicle which is the expert system. This approach may be argued against by people who want to concentrate on the evolution of computer software, etc., for generic or specific uses. However, if the objective is to produce an expert system for a specific topic in the manage­ment of construction, then that is where the evo­lution should begin by studying what such experts do. After that has bee. done, the resulting written expert system ca. be computerized with greater probability of producing a working con­struction management expert system, than would be by concentrating on the computer programming

CONSTRUCTION PROJECTS AND THEIR REQUIRED EXPERTISE

In construction projects generally the expertise required can be biased towards (a) deciding what the end product shall be or (b) the actual process of constructing the end product.

The expertise required to crystallize the end product deals with its compatability to subsequent use and choice of and juxtaposition between the physical entities in the end product and the

relationship of the end product to its future surroundings. This expertise is mostly exercised in the design phase of the project to produce the unique end product to meet the users needs and owners objectives broadly across the parameters of quality and cost and to time in the context of economic and social longevity.

The expertise in the construction process is mostly a sub part of management science to provide the project owner with a service to effectively and efficiently procure the end product as part of the future built environment given the current state of the marketplace for its required resources. In management of construction there is a whole array of somewhat specialized management tasks and people who have varying degrees of expertise in each such task. These specialists can work in narrow fields of expertise such as estimating, scheduling, safety, choice of crane or hoist, etc or their work may utilize more general expertise such as project management, site management or human relations, etc. involving the interactions and coordination between specialists, etc. on specific projects. All these topics of expertise are not a part of a physical science although a minority of their operations may involve aspects of physical science. Managerial science, as distinct from physical science, involves human beings both as constituents to be considered in the subprocesses and as the exponents of the subprocesses of that procurement service which the construction industry provides to society.

The above narrow fields of expertise can be seen as more repeatable and thus more easily programmable [12] because they constitute the somewhat fixed management system through which passes the vari­ables i.e. the different end products from each project.

The broader topics of more general management and strategic management are less suited to repetition and more difficult to program due to their 'domain dependant' nature which is a major facet of general management in that the problems and situations it has to solve are not similar to each other and come from the unique combination of factors in a specif­ic situation [1]. Here the recognition and identi­fying of a problem and establishing its nature and format tend to be the major difficulties in solving it rather than in the execution of the solution.

Thus, to begin to develop expert systems in managing construction would seem to be more successful by starting with the topics or processes which are repeated in the same manner sufficiently often as to be worthy of capital investment in creating a programmed approach i.e. an expert system to reduce the cost or duration of each execution of that process.

It must also be realized that management of con­struction is a dynamic evolving process and that the speed of change in management topics can vary at different points of time. This raises another major issue that it may only be worth the capital investment in a. expert system that is either (a) directed at a very slowly evolving topic whose caution can be fixed for a sufficient period of time to recoup the investment or (b) couched in somewhat broad terms for general guidance as to how to execute that process in all situations faced by exponent of that topic. In the latter, the expert system user will use his own judgment when to veer away from the expert system on a particular part of that topic on a particular project.

EARLY EXAMPLES OF EXERT SYSTEM DEVELOPMENT IN CONSTRUCTION

Various university construction programs are entering the expert systems area.

Stanford is developing systems to better the site layout of temporary facilities [13] and carry out quality control as well as generic "shells" for expert systems. Carnegie Nellon [14] and Illinois [15] developing systems for structural design and project management and scheduling. Loughborough is developing systems for selection of cranes for projects and for scheduling of construction (161. There are also efforts to computerize building regulations and codes which may or may not really be expert systems. Georgia Tech [17] and MIT are examining projects risk analysis. Most of these approaches use the single person expert approach which does reduce costs but also increases risks as to ultimate quality and virtually all of them are not yet operational.

Two less global topics have been tackled by this author. One such system distilled to a summarized format the written factors used by twenty corporate project owners to evaluate the performance of construction contractors on their work. To this was added a weighed and quantified scoring system from private project owners heuristics to create an expert system to include contractor performance evaluation as an ingredient of evaluating bids and selecting contractors for future work [18]. The other system was carried out for a city government by questioning all appropriate staff members for the ingredients involved in the creation of a special assessment bond for a construction project the control processes on such a construction project [191. All of these inputs were pooled to flow chart the whole process with decision and choice nodes where required. The whole process was then expressed in sub flow charts of sub routines which operated during a construction process either repeatedly or once per project. These sub routines could be used as a general expert system for use in managing any such project to be carried out by City Hall. It would appear that this approach to programming the repeating activities could be very useful to guiding lower level staff in a con­struction contractors organization on how to carry out their work as it would be done by a senior staff member i.e. an expert system.

BENEFICIAL/FAIL SAFE FEATURES OF DEVELOPING EXPERT SYSTEMS IN MANAGEMENT OF CONSTRUCTION

As stated above, some people are beginning to try to develop and apply expert systems to the non-physical and non-medical sciences of managerial processes such as management of construction. There is a dramatically greater level of complexity and increase in variety of situational probabilities of activities and variables in the managerial and human sciences compared to the physical sciences. In management sciences different processes carried out by different people may be the best way for each such different persons to reach the same desirable or satisfying conclusion, process or decision.

Compared to the comparatively deterministic nature, of the more physical science examples of early expert systems, managerial science is more one of a combinational and situational context in which there may be no one specific best way of carrying out a whole process and in which there may be equally good but different processes to reach a valid conclusion.

There are a fail/safe features for capital invest­ment in the development of expert systems if their evolution is carried out wisely in the context of the above risks, costs and questions. These fail/safe characteristics can introduce benefits to the construction industry even though the resulting expert system may only approximately replicate the results created by the expert exponent or decision maker. These features derive their benefits from (a) the temporal nature of human life, i.e. that the human mind loses expertise over time and (b) the needs of the educational processes of future experts.

Currently construction expertise is locked in the brains of particular human beings whose mental expertise on a topic may vary over the years and who may or may not divulge his expertise to others depending on his emotional make-up. Without doubt, current experts will change work to create a new challenge for themselves or to increase their money rewards and eventually retire and die and thus current expertise will be lost. The expertise in a human brain erodes and dissipates over time whereas in an expert system the expertise is recorded in writing or a computer program. It is this record­ing of the current expertise and data which is of great value to the future potential to allow further study, analysis and improvement in process­es of executing management of construction.

There appear to be at least three major fail/safe features to be gained by trying to develop expert systems in construction, (a) capturing of exper­tise, (b) acceptability by current experts by using a multi phased evolution process and (c) education­al benefits from the evolution of expert systems.

Capturing of Expertise

The creation of the expert system requires the process and decisions making ingredients of the experts in a topic to be recorded, written, etc. firstly in a raw captured format from each expert and then in a common pool or analyzable state describing their factors and processes of exe­cut ion.

By capturing the expertise of current experts in writing or in the format of flow charts or computer programs enables, at the very least, potential new experts to start their learning by studying exist­ing expertise hence reduce the duration, risks, costs and improbabilities of become a future expert. This also releases time and energy for these future experts to push the wheel of knowledge further forward because the recording of the expertise has given them kinetic energy and traction to that. wheel of knowledge.

It appears that in its embrionic state of evolu­tion, the field of expert systems in management of construction should be trying to establish and record how experts think and act on the topic. Even where such work has been attempted e.g. selecting cranes for projects, it has had great difficulty in producing heuristics common to a number of experts without that expertise being firstly written down [1].

Put simply human expertise may be seen as ephemeral expertise but expertise in systems may be seen as eternally (!) recorded expertise.

Acceptability By Current ExpertsBy Usinq A Multi Phase Evolution Process

There is the ultimate potential that an expert system could be one in which a construction mana­gerial problem is typed into a computer as a 'black box' and it will print out the answer of what to do without the human responsible for the subsequent action knowing anything or little of what went on in the computer. This would be like receiving a little card out of a fairground machine telling you what your future life will be like - and believing it! It would appear that humans with a deep knowledge of a topic might balk at accepting following the above. Certainly, experts in the topic would most likely be at least querulous at following the computer's instructions without knowing and understanding the process and body of expertise used by the computer. Furthermore, the experts in that topic might not be willing to allow their subordinates for whom they are responsible, to carry out their work based on the computer's instructions without first fully understanding and accepting the contents of the computer software.

This raises a major issue in the use of expert systems in that before they can be used to do work they must be accepted as to their ability by the humans who will have to accept responsibility for the work results pursuant to the use of the output of the expert system. It also points up that an initial stage of evolution should be reached before moving on to the above situation of the 'black box' expert system if only to have the initial state of evolution in readable format to be evaluated and accepted by the current experts in the organization who will be responsible for the results of the use of the expert system. It has been implied that the need for almost perfection in the expert system may be to defend against future law suits based upon poor expertise utilized and provided by expert systems rather than to add to the quality of the expert system [1].

While the black box computerized approach to expert systems is perhaps interesting to those system developers in the theory of expert systems but not to those desiring only to carry out a management of construction process, there is the middle ground of the 'open' expert system alternative. This is one where all the logic of expert systems is in a written package but the user has to create the data base of parameters of the topic and establish values for the parameters dependent on the situa­tion in which the resulting expert system will be used. This open system approach could be the second phase of evolution of expert systems in ,management of construction. It would guide the construction executive through the thought process­es but leaves the construction executive with the task of gathering all the information and parameter values from the situation in which he is making and injecting it into the package before it can be used. Such open systems should be developed out of the pragmatic expertise gathered from human experts and while supporting the activities of staff less capable than experts and who may be working at their full capacity or may be evolving towards becoming the experts of the future. Such an open system could be a sieve to capture additional ingredients and features over a series of uses, that need to be in the expert system as it develops towards the "black box" state in the future.

Educational Benefits From the Evolution of Expert Systems

Initially, it would appear that the creation of expert systems applicable to management of con­struction should be directed towards creating expert systems to be used to replace current practitioners. However, complementary to that objective, there is the fallback or failsafe objective that such efforts in creating expert systems will have to establish the information, thought processes, parameters and values in various situations, etc., etc., which constitute the human expertise in that topic. Thus, the recorded expertise in a written and humanly readable form can be a very strong educational tool without in any way detracting from the first objective of trying to create expert systems than can replicate the output of an expert in action. The learner can be given situations and allowed to make his choices in conjunction with the expert system as part of his education to became a human expert. Alterna­tively, to enhance his expertise, the learner can explore all the alternatives posed by the expert system just as sportsman may train under all potential circumstances under which he may have to play.

SOME CONCLUSIONS ON DEVELOPING EXPERT SYSTEMS IN MANAGEMENT OF CONSTRUCTION

The existence of expert systems could have consid­erable benefits to various sub processes in manage­ment of construction. These benefits are mainly in (1) achieving the results of an expert exponent from the services of a middle level exponent or neophyte or (2) recording in writing, or computer programs, the expertise of the current experts rather than allowing such expertise to remain only in the ephemerally of a human brain. The first benefit appears to be rather far in the future whereas the second benefit provides the opportunity for the initial expert system (or its raw ingredients which were gathered from, recognized arts) to be studied objectively and improved without having to repeatedly go through its evolutionary process of ingredient, etc., capture from human experts. A third major benefit of recorded expert systems enables the ingredients and process­es to be learned by neophytes from its distillation or raw expertise of the human experts. This can be a considerable educational tool for advancing expertise of new entrants to the construction industry.

The process of developing expert systems is akin to venture capitalism in that there is no surety that full success will be achieved. Expert systems, especially in the areas of human managerial sci­ences, are very complex and thus there will be considerable risks, costs and improbabilities involved in their creation. The veracity of the research processes used to ascertain the features which constitute expertise thinking in the field of managing construction may be the most difficult aspect to carry out but must be the first phase of creation of expert systems. Also the larger the managerial topic to be transformed into an expert system, the greater the risks to success.

Given the evolution of research into artificial intelligence, the availability of computers and primarily the human curiosity and desire for improving the way man carries out his activities, the evolution towards expert systems will continue and has the promise of being the way in which many managerial processes will be guided and supported in the future even without reaching the far out ultimate of a computerized 'black box' non-human manager of construction.

It would appear valid that the most viable long term approach to developing expert systems for construction managerial processes would be to first capture the required expertise and second record that expertise in writing with its knowledge factors and relative values, etc. These should then be reviewed, discussed and refined etc., by current experts in that topic. Then the package can evolve to be expressed as a flow chart and again reviewed and approved by those responsible for using the output. Only then should it be programmed for outer handling prior to use by knowledgeable users. After satisfactory use by these people, the evolution can go to further approval and subsequent evolution towards "black box" use of the expert system.

CONCLUSION

The promises for the future from expert systems may or may not be realized very soon, but the movement towards fulfilling these promises of "computer black box process" making decisions, monitoring or guiding the execution of managerial tasks in construction will certainly cause an increase in the capturing of human expertise on paper from the ephemerality of human experts thought processes. That written record of human expertise is then in a format which can be more easily transferred to others - either to follow as in an expert system or simply read as in a book. Thus, the wheel of human knowledge can move forward with traction created by that recording rather than continuously slip as

each new generation has to learn slowly and pain­fully on the job.

REFERENCES