Then the subject might decide that if this is the case more of attribute A and less of B would be preferable, as represented say by Point 3. A computer, properly programmed with constraint equations, can return a judgment of feasibility or infeasibility of an initial aspiration, as for example in Figure 2.6, by letting the person know that a closest allowable point is 2, assuming a rough equivalent weight on the two attributes. Then, by successive interactions with a computer, all neighboring points (combinations of the attributes) might be considered to find a point that is satisfactory in all respects. Point 1 in Figure 2.6 might be such a point in a two-attribute space, where A is, for example, the ride quality or comfort of a trip and B is the speed. What now appears much more tractable is to let individuals initially specify “aspiration” points in “attribute space,” that is, say what they would like. Though utility theorists have applied their methods to a number of complex and real decision problems, the empiricists have run into serious problems getting people to render judgments in terms of objects or events which correspond to nonfamiliar situations. Thus to talk of the “optimum human–machine system” is usually either a gross misuse of words or a revelation of naïveté. This is especially true when people are part of the system, for we know less about human behavior and what is good and bad with respect to people than we do about physical systems. For example, what are the equally good combinations of speed, safety, and comfort for an air trip? Clearly it depends on both the person and the situation. In fact, usually the interaction of all the key variables in the system is not well understood and, furthermore, there is little consensus on what is the trade-off between the system variables in terms of relative goodness or badness. Unfortunately even the determination of the optimum trade-off seldom makes sense. When there are reasonable quantitative specifications of system constraints and objective function, even if the equations are not well behaved or if probabilities must be assigned to many conditions, a computer can be set to searching for the optimum. We all know of politicians who blissfully assert demands for “the greatest performance at least cost, with 100% safety” and seem not to understand that such statements make no sense. If one only had to optimize the objective function, one simply would choose the highest speed, greatest accuracy, least energy, least error, and least cost, with no constraints. Formally this amounts to minimizing or maximizing this objective function (depending upon whether it is defined in terms of cost or reward) together with solving the set of system equations (sometimes called the system constraints, meaning nature's constraints on the trade-off of desirable performance variables apart from what any individual intends). If this objective function and the equations representing the system to be optimized are well behaved (i.e., continuously differentiable and unimodal), then the optimum trade-off can be determined with relative ease. When this “ultimate trade-off” function can be defined in numbers, theoretically the optimum or very best trade-off can be found. That is, a single variable which defines ultimate goodness or badness is written in terms of time, accuracy, energy used, errors, and costs. Ideally the goal can be specified as a scalar “objective function” of system input and output variables. It is important to consider how this goal should be, could be, and is, in practice, specified. In more rigorous terms, that means to make some output or dependent variable conform more closely to some goal or norm. System analysts practice their art (some of it may be science) for a purpose-to eventually make the system perform better. Sheridan, in Human Factors in Aviation, 1988 Goals
0 kommentar(er)
