Dr. Robert A. McIvor is past director of the Canadian Defence Information Service and past manager of its data processing branch.
A NATION planning to build the world's first job-market might well consider running a computer-aided simulation of it first.1 This would permit that nation's engineers to study, assess, test, adjust, and ultimately perfect the performance of the job market's algorithms before adoption. This first, engineering stage of the simulation would, if successful, produce a tested prototype. The government of the pioneer nation could then, if it chose, run a second simulation, using the prototype itself, to familiarize its people with the job market's various services, possibly even before asking for their final endorsement. This second-stage simulation would amount to a "dry run." By participating in it, the citizens of the pioneer nation would be able to sample the flavor of life under a job market, as well as familiarize themselves with its novel procedures under non-threatening circumstances, before being asked to buy and sell real jobs in it.2
Even so, there are some circumstances in which a dry run is likely to be judged unnecessary. If movement into the job market sector of an economy is planned to be a voluntary one that might last several years (see chapter 14, "Transition to a Job Market,") a dry run would add little to the already gradual process of familiarization during transition. But where the government is planning an "overnight" adoption, putting the whole economy under the job market umbrella in a single night-as the Swedes went from driving on the left when they went to bed one night to driving on the right the following morning 3 -such a simulation would permit a people faced with that kind of sudden change in their habits to become familiar with the new options their job market will give them in a thoroughly playful setting before being required to make such choices in all seriousness in the real world. It is under these unusual psychological circumstances that a second-stage simulation conducted with a tested prototype might be judged very useful.
In the present essay we shall concentrate on the way either a one-stage or two-stage simulation could be run in a pioneer nation. We'll assume that our pioneer country is a small, culturally homogeneous, European-type social-democracy with a strong labor-party tradition in its politics. But obviously any nation is free to seize the initiative to become the job-market pioneer.
Most engineers would regard developing a tested prototype of anything as novel as a job market as absolutely necessary before turning it over to its users. According to our Western engineering tradition, things that have never before been built-and therefore put their first users at some risk-should, so far as possible, be built as harmless prototypes first. The prototype should be run, tested, stressed with the severest anticipated loads, 4 preferably without humans aboard, and then revised and rebuilt as many times as necessary, and in that way "perfected" before being permitted to carry real loads in the real world. It is in this tradition of assuring the safety of early users that this essay is being written.
Stage I. Engineering
Engineers tell us that they have successfully engineered something when they have made themselves as certain as they feel they need to be that it will be able to operate under the worst possible conditions anyone can reasonably expect. If it is a bridge, this means loading it bumper-to-bumper with heavy tanks and simulating the worst floods and winds that ever rise or blow; if it is a ship, it means testing it under the heaviest loads and impacts that the worst possible storm could impose on its decks and hull. Normal conditions hardly matter when testing a prototype. So no statistical estimates of the normal operating conditions of a job market in any particular country are required to put one through its Stage I engineering tests.
What will be required to conduct an engineering test of the first job market is (1) the same network of control computers that are planned to serve the actual job market in that country; 5 (2) a suite of software embodying in at least preliminary fashion all the algorithms that the computers in the network will require in order to perform the job market's various functions in timely fashion; and (3) several teams of scientists, engineers, and technicians, among whom are those who will actually create the job market's software but also others capable of devising various artificial loads to test it, and of correcting the flaws and defects discovered in its structure through such tests.
- The Control Hardware: This should be, as far as possible, the same suite of computational facilities that are ultimately to be used by the real job market. Whether these are a network of computers connected through the Internet, or a more centralized system that the engineers will access through workstations, will depend entirely on the existing infrastructure in the pioneer country.
- The Software: All the software to be tested must exist in at least trial form before engineering tests can begin6 . Any realistic test may require that two or more of the job market's many algorithms be called into action simultaneously. So the software must include at the outset usable versions of (a) algorithms for cutting work into jobs, for optimizing job-sizes, for adjusting the average size of jobs, and for assessing and then responding to the rates at which jobs enter and leave the market; (b) algorithms for monitoring the flow of shoppers into and out of the market and for responding to changes in these flows; (c) algorithms for adjusting credit-rates, for keeping all credit-rates within a specified range, for spreading credit-rates out along that range, and for altering their allowable range when told to do so (presumably by the national legislature); and (d) algorithms for responding to changes in the maximum value of the average annual-work-year (as this too will be a legislatively-imposed numerical constraint on any job market's behavior). Finally, (e) the algorithm for keeping the average credit-rate at unity must also be present in the original suite of software to be tested
It would also be desirable, but not absolutely necessary, to test trial versions of algorithms that (f) cause the computer to "learn" what properties of work are likely to affect the credit-rates of the jobs made from it, and that (g) tell the computer how to use this knowledge to assign credit-rates to totally new work; for ultimately the software of any job market must be able to assign trial credit-rates to novel work with increasingly accurate predictions.7 This, then, is the short list of algorithms to be tested during the simulation, and, when possible, improved.
Software that will not be necessary for engineering the prototype job market should also be mentioned. None of the accounting, commercial, and financial software used in a job market economy will need to be tested. While such price-setting and money-handling algorithms will be essential to the performance of any real job market economy, they are, first, peripheral to the computer's job-creating and selling roles, and, second, do not require engineering. Software of this commercial and financial kind is already in widespread use in the Western world. It is the ability of the job market to make and sell jobs that can be doubted, and so must be assured by prior engineering.
- The Engineers: The engineering group can be partitioned into at least four teams. In some ways these teams will operate independently:
- The Developers. These are the computer scientists, programmers, and software engineers who wrote the software in the first place and who are presumably prepared technically to fix it once its defects are discovered.8
- The Employer Team. This is the group of statisticians, economists, and "human resources" experts who will be preparing the input stream representing the needs and desires of employers. They will generate an artificial data-stream informing the job market of the types and quantities of work employers need, the dates by which they need it, and the qualifications they require of the workers who perform it. The shape of this data-stream will be deliberately changed from time to time, partly to represent long-term trends or seasonal changes, but often as part of a more severe test of the algorithms. By modulating this data-stream in concert with modulations provided by the Employee Team (see below), a series of progressively more stressful tests of the prototype can be conducted.
- The Employee Team. This is the group of statisticians, demographers, sociologists, and social psychologists who will be representing both the job-seekers and the job-holders in the simulation. It will be the task of the Employee Team to generate and modulate two further data-streams impacting the algorithms. The first and most important of these will be the stream of simulated responses being made by job-seekers to the jobs available to them. This data-stream will reflect the mix of backgrounds, educational levels, ages, states of health, tastes in work, work experiences, and other characteristics of the job-seekers whose presence in the market is being simulated. The second data-stream will provide similar information about the job-holders in the country, the people who already have jobs and expect to keep them. Both these data-streams will be deliberately modulated by the Employee Team, sometimes in concert with, and sometimes independently from, the changes being made in the data-stream coming from the simulated employers.Varying the shape of the job-seeker data-stream will provide the severest tests of the prototype. Suppose a group of 1,000 inexperienced and under-educated youth suddenly materializes, and all need work. This is the kind-but certainly not the only kind-of stressful event that will be simulated by the Employee Team. Tests of this and many other kinds will be continued until the job market's algorithms are robust enough to handle all of them.
- The Referees. The fourth team of workers will be the Referees. These will include, not only the computer scientists who oversee the simulation, but also the political scientists who will be representing the changing interests of the nation and its people in the simulation. In real life these political forces will impact on the job market's operations through legislative actions or through popular initiatives or referenda, in any case altering the numerical parameters under which the job market is obliged to operate. It will be necessary to test the alacrity and accuracy of the prototype's response to such changes in its directives, The Referees will also assess the results of all the other engineering tests, and call for new ones when they judge them to be appropriate.
This quadripartite group of workers-no matter how assembled by the pioneer government, or by whatever other funding agency commissions their work-will run the Stage I simulation until the engineering tests have been concluded. At this point a tested prototype of the job market, one ready to be used, will presumably be delivered to the commissioners of the simulation, and they will then either authorize a Stage II extension of the simulation, as described below, or they will not. If they do not, it will presumably be because they have decided on immediate acceptance of the prototype as their new job market, but perhaps installing it only gradually as is described in Chapter 14. In any case, all that will then be left to do is to get the banks involved and open its doors.
If, on the other hand, an all-at-once overnight adoption is planned, then a Stage II simulation of the sort described below is likely to be judged appropriate.
Stage II. Familiarization
The primary purpose of this second stage of the simulation is to allow a people who have, in principle, already accepted the idea of a job market to familiarize themselves with its practices. A secondary function will be to allow the engineers to fine-tune the prototype's algorithms for the setting in which they will actually be operating, using data generated by the country's actual employers and job-seekers.
What would be the elements of such a Stage II simulation? There are five: (1) the prototype itself; (2) the control hardware of Stage I supplemented by (3) the access hardware with which to make a "dry run" on a national scale; (4) the human players who will be participating in the simulation; and (5) the team of engineers and scientists who will be managing it. Let's deal with each of these five elements.
- The Prototype: This is essentially the suite of software, now tested, that has emerged from the Stage I tests.
- The Control Hardware: This is the same large computer, or network of large computers, used in Stage I to engineer the prototype.
- The Access Hardware: While the Stage I hardware consisted of the job market's control network-the computers that ran the software-as well as the work-stations that gave the various teams of engineers access to it, everything else used in the Stage I tests was simulated. In Stage II, however, the public will be invited to come on board; so the widest possible access to the prototype must now be provided. Whether this is accomplished through a network of personal computers supplemented by public terminals, or by a more centralized, national system that is accessed primarily through public terminals supplemented by a few personal computers, will depend entirely on what is already available in the pioneer country.
What the pioneer country will not require, even for a Stage II simulation, is whatever physical job market floors (like the one described in Chapter 3) it has planned to provide its job-shoppers once their real job market is in operation.
Let us assume (a) that our pioneer country is one in which its telecommunication network already reaches nearly every corner in it; (b) that many citizens already own, or have ready access to, personal com-puters; (c) that most of these computers are already equipped with modems (devices for linking computers to telephone lines); and (d) that inter-national computer-networks like the Internet and the World Wide Web are already in widespread use.
In that case, the Stage II simulation need only supplement the existing hardware with additional modems (made available to computer owners who don't have them yet), additional computer terminals (installed in public places like post offices, schools, libraries, shopping malls, and city halls), and additional connections to the Internet (in those places that telephone lines do not yet reach).
The access hardware will be ready to run the Stage II simulation when every economically active resident of the country-every job-seeker, job-holder, or employer-is a potential player in it.
- The Players: The fourth element is the human players. These will be of two principal types: Job-Seekers and Employers. Players are potential Job-Seekers in the simulation if they actually would be potential buyers of jobs from a job market if one existed. That is, if they are real youngsters poised on the brink of work, real employees dissatisfied with their present jobs and looking for new ones, real retired persons thinking of coming out of retirement now that a job market is going to be available, or real unemployed persons wishing to find work again whether they are currently supported by unemployment benefits or not.
Players are potential Employers in the simulation if they either own a firm or manage the employment activities of a private- or public-sector organization. It is these real players in the nation's economy who will now be supplying the simulated work arriving in this simulated marketplace. They will report to the market-only when and as such real work requirements do emerge for them, of course-that they require a certain kind and quantity of additional work to be done (additional, that is, to what their present employees are already doing), and they'll report the qualifications required to do that work, when they require it to start, and where it is to be performed. In short, all the real employers of the nation, public and private, will be invited by the government to participate in the simulation by supplying its managers with true and complete facts about their real work-requirements.
If there is a shortfall of either Employers or Job-Seekers-and there is bound to be, for not all the members of any group invited to perform some service actually volunteer to do so-that shortfall can be made up by the same teams of workers who managed the artificial data streams for the Stage I simulation. Facts about the entire economy will of course be available to the simulation's managers. In particular, they will know the characteristics of the populations of real employers and real job-seekers at any given time; so the nature of both shortfalls can be accurately ascertained. The two shortfalls can then be made up by constructing two supplementary data-streams of the sort on which the Stage I simulation relied.
The totality of such work requirements (real and simulated) will, in the aggregate, be the work that the computer will then cut up into the required number of simulated jobs. And the totality of job-seekers (real and simulated) will be the group of people looking over the simulated jobs and pretending to buy the ones that attract them most from among the nearly always larger set of jobs for which they find they qualify.
- The Managers: The fifth element in a Stage II simulation will be its managers. These are the scientists and engineers who will be either fine-tuning the computer's performances during Stage II-readying the prototype for actual use-or supplying information to the players that will familiarize them with their options and obligations.
The managers will also have the preliminary task of designing the two or three separate sites on the Internet-on the World Wide Web or its local equivalent, call them "websites"-that will keep them in touch with the players throughout the simulation. One website will inform the Job-Seekers of the jobs available...just as the real job market will. It will also tell them of the qualifications currently required for each job, of its contract length and width, and of the (simulated) circumstances under which the (simulated) work will be performed. At this website, too, the qualifications of potential buyers for specific jobs will be evaluated, and if a player actually qualifies (or simulates qualifying) for a certain job, and then decides to buy it, a (simulated) contract between that buyer and his or her new employer will be drawn up. All this will contribute to the familiarization of players of both types with job market procedures, as well as to the fine-tuning of the prototype.
Another website will be available for interaction between the simulation's managers and employers. Here the questions of real employers will be answered, instructions will be given them on how to participate in the simulation, and eventually on how to use the real job market, when it opens, to their best advantage. It is at this website, too, that information will be elicited from employers about the changing work-needs of their real or imaginary businesses, or of the organizations for whom they manage employment, during the period of the simulation.
The simulation's managers will also design certain data-collecting features at both websites, as well as the statistical programs by which the responses of both types of players will be collected and analyzed.How long should such a Stage II simulation run? For as long as it takes the designers of the simulation and the officials of the government, or, alternatively, of the political party or foundation that has been the source of its funding, to convince one another that an infrastructure capable of supporting a real job market is now in place, and that the algorithms of that job market are now ready to cope with the real world9 .
- It should be understood that what is being proposed here is not a "computer model" in the commonly-used sense of that phrase in natural science. This computer simulation is not like an astrophysicist's model of the evolution of quasars, or a glaciologist's model of the breakup of the West Antarctic ice-sheet. Such models are always meant to simulate some natural process, and, as Gibbs (1997a) remarks, they have inherent limits: "computer programs can only simulate the stuff you know." In contrast, the simulation we'll be discussing here is an engineer's model. It is like the model of a sailboat or a bridge that is meant to be tested in a physical laboratory...if necessary, to destruction. That our model is made of software, as are most scientific models, and not of wood or metal, as are most engineering models, should not be allowed to obscure this distinction. The real job market will also be made very largely of zeroes and ones.Back
- Nemacek (1996) also cautions against uncritical acceptance of the results of computer modeling. Because of its essential arbitrariness, until and unless the results of running the model are compared with the natural process that it models, and are then corrected, modeling is enlightened guessing. Lab experiments and sampling from nature are still "vital parts of the scientific armamentarium." The two stages of the simulation proposed here are more like Nemacek's extended "lab experiments" (Stage I) and his "sampling from nature" (Stage II) than like ordinary computer-modeling.Back
- This is what was intended; but in actual fact, an enormous number of Swedes stayed up until the switching hour, and were on the roads in the middle of the night, waiting to make the switch. One of us (McIvor) was in England at the time, and his English friends facetiously suggested that if the English were to make the same change, they would want to do it gradually, say, lorries first, for a six-month trial period, then buses, followed by cars. Sometimes he wonders if gradual introduction of a job market would be as feasible as this method.Back
- Plus a safety factor, which, for economic reasons, is usually as small as can be confidently predicted.Back
- It is likely that a network of control computers would be used, rather than a single computer, both to spread the work, and to ensure redundancy for continued operation in the event of failure of any one element in the system.Back
- The many ways in which the job market's challenge to software developers can be met are not discussed here. But to the experts with whom we've discussed this problem, the tasks of devising, programming, and improving the suite of algorithms that are described functionally in this book all seem quite feasible. As to meeting that challenge in efficient ways, Gibbs (1994) reported that certain then-new "clean room" techniques of software development were apparently making error-free development of large suites of software (such as the job market will require) quite possible. In 1994, according to Gibbs, ordinary software-development was still not a "mature engineering discipline." Debugging, for example, was still a hit-or-miss art, programmers finding, on average, only about a third of the errors their programs were found ultimately to contain. But the new "clean-room" technique described by Gibbs, which involved "growing" small parts of the ultimate program to "run perfectly the first time," was already changing that picture. It offers the highest promise to date of solving the "hidden error" problem.In a later article, Gibbs (1997b) described an even newer system of software development-one that incorporated the clean-room technique but went beyond it-used to develop a new automated tracking system for the U.S. Department of Defense. The new system appears to have great general promise. It consists essentially of six innovations: (a) the computer-displays and user-interfaces were developed first, not last as is usual; (b) the "riskiest" segments of the program were worked on first, not allowed to "slip to the end" as is also usual; (c) large chunks of off-the-shelf software were reused in the program; (d) programmers reviewed one another's code periodically, "catching more than 200 major design errors while they were still easy to fix;" (e) programmers were also required to use the clean-room technique of "perfect[ing] each segment [of a program] before moving on;" and (f), instead of "avoiding contact with the [intended] users of the software," as is apparently normal, programmers were encouraged to invite the users frequently "to come and see what they had done." This produced, in record time, an enormously complicated suite of relatively bug-free software that worked the first time.Back
- Abu-Mostafa (1995) reports on a "computer learning" technique that might be useful for generating and/or maintaining job market software. Using this technique, programs can be written that cause computers to emulate "humans learning from hints" and from other sorts of potentially instructive experiences.Back
- Lloyd (1996), in a review of Holland (1995), discusses the latter's work with an interdisciplinary group of associates-then still incomplete-in designing a computer model of stockmarket behavior. This work may be relevant to the job market simulation in two ways: (1) Holland and associates are clearly the kinds of scientists who could build, or advise on building, an engineering test of the job market, and should probably be consulted by the team that actually does the engineering; and (2) studies of what might happen to stockmarkets under job markets-developing a scientific computer-model, this time, in the sense developed in Note 1 above-could also be made. See Note 10-9 for more on the issue of stockmarket volatility under job markets.Back
- Simulations of the job market's software might spontaneously develop on the internet. Even before more formal agencies had become interested in funding the engineering of job markets, groups of programmers interested in the job market idea might find each other, and so "assemble" electronically, where they could organize themselves by e-mail into teams, develop trial suites of software, and test them on similarly volunteer groups of "beta-testers." These last could simulate the responses of employers and job-seekers, respectively, to the developing software; and the results obtained by such unfunded groups of workers could contribute materially to the adoption of job markets in the real world...especially by poorer countries. Persons interested in joining such groups of programmers and/or beta-testers may contact each other by sending e-mail messages via this website.Back