Human performance: from then to now
DOI:10.1093/acprof:oso/9780199228768.003.0020
Abstract and Keywords
This chapter examines changes and developments in the study of human performance in psychology. It discusses major findings concerning signal detection, the relation between choice reaction time (RT) and amount of stimulus information, the notion of limited capacity, and linear stage theory. This chapter suggests that psychologists should not become too fascinated by the new intellectual climate of cognitive neuroscience so as to continue neglecting human factors.
Keywords: human performance, signal detection, choice reaction time, stimulus, limited capacity, linear stage theory, cognitive neuroscience
Then
I was a young Psychology graduate and my employer, the Institute for Perception at Soesterberg, allowed me to attend the Brussels 1957 International Congress of Psychology. I was thrilled by my first scientific meeting ever but rapidly became miserable because I failed to comprehend almost anything that I heard. I was not yet aware that, aside from the difficulties imposed by the typically poor presentation skills of most academic speakers, a thorough background knowledge of a topic is required to understand and appreciate an oral presentation. Moreover, I suffered from a language problem. However, in retrospect, two symposia on the final day of the Congress shaped my interests. One was on Applications of Information Theory and the other was entitled ‘Too busy or too bored’. Both were mainly inspired by researchers from the Applied Psychology Research Unit in Cambridge.
The central message of the first was that processes in between input and output are controlled by the information contained in the input rather than by the input itself. The prime concern shifted from S-R associations to ‘information processing’ between stimulus and response. The information metaphor was primitive—a fixed and limited-capacity telecommunication channel, which replaced the equally primitive Behaviourist metaphor of an associative telephone exchange. The related research was eloquently summarized in Broadbent's (1958) influential Perception and Communication, which argued that previously neglected topics of research could be unified by the guiding principle of the telecommunication metaphor preceded by a selective filter and a preselective short-term memory system to safeguard the system against temporary overload.
It was striking how quickly this new set of concepts and interests replaced the earlier Behaviourist ones—no more than 15 years after the publication of (p.254) Hull's (1943) Principles of Behavior. After the mid-1950s Behaviourist theorizing and results were almost totally neglected. Incidental attempts to stress communalities (e.g. Berlyne 1957) were effectively ignored. It may well be a characteristic of a paradigm shift that previous work is viewed as obsolete. Central themes of Behaviourism—learning and motor behaviour—were undervalued, and animal laboratories rapidly disappeared from psychology departments. Though renewal is necessary, this carries the cost of loss of appreciation of earlier accomplishments. As an illustration, effects of long-term work had been studied extensively in the context of effects of massed versus spaced practice on performance in tracking and in serial reaction tasks. The usual finding was a negligible practice gain during a massed session followed by a marked improvement at the beginning of the next massed session—contrasting with continuous gains during spaced practice with brief rest intervals between trials. The obvious conclusion was that the absence of practice effects in the massed condition was not due to poor learning (i.e. deficient habit formation), but to a suboptimal drive state, referred to by the term ‘reactive inhibition’. Such research is highly relevant to the interpretation of studies on sustained attention and vigilance. Yet, references to reactive inhibition are rare.
Another consequence of the paradigm shift was that relations between Psychology and Physiology were largely disconnected—perhaps with the exception of the study of evoked potentials and galvanic skin responses, The guiding metaphors of information theory—as well as of its successors—were not inspired by Physiology but by Technology and Computer Science. I remember a remark by a leading American psychologist during the early seventies that Physiology had done nothing but carry Psychology on wrong tracks—with clear reference to Behaviourism. Nobody took issue.
The second symposium in Brussels included a discussion of two applied research themes: perceptual-motor load and vigilance, springing from applied problems during World War II. They launched much research on reaction processes, selective attention, visual search, performing under adverse conditions, and short-term memory, topics that had been neglected by Behaviourists. These were main themes at the first Attention and Performance conference (Sanders 1967), at which Human Performance was defined as the combined basic and applied analysis of human skills.
It is relevant that most applied work is commissioned by non-academic bodies and so is described in reports rather than in the open literature At Soesterberg I was involved in applied research, concerned with semi-automatic systems of the 1960s, such as perceptual-motor load, air-traffic control, guiding aircraft in airports, ship manoeuvres in harbours or in shallow waters, (p.255) industrial quality control, the relative merits of paced versus self-paced work, display design, human signal detection in radar and sonar, and a variety of issues related to traffic safety. Some questions had simple solutions, such as. a licence plate number that can be easily encoded and retained. Others, such as a new ship-traffic control system for the Rotterdam harbour system or optimal visual search strategies during driving, were much more complex, requiring extensive research and collaboration. However, the boundaries between process-oriented and task-oriented research (Schmidt 1987) were weak, transient, and viewed as mutually dependent and inspiring.
Signal detection
The popularity of Information Theory was short-lived because it had no provision for feedback and learning, and could not account for strategic interaction between an individual and their environment. Moreover, some main predictions were falsified. For instance, the linear relation between choice reaction time (RT) and amount of stimulus information, suggesting an information-based constant processing rate, was hard to reconcile with the marked effects of practice and S-R compatibility. In addition, the effect of information load was confounded with the effect of the probability of stimulus repetition between successive trials. When repetition probability was held constant, the relation of RT to information load simply vanished (Kornblum 1969). The demise of the information metaphor was the first step from ‘then’ to ‘now’. It was followed by numerous new technology-based models and a considerable theoretical diversification (see Chapter 15).
Some of the new models aspired to the status of unifying principles, whereas others aimed at covering a smaller domain. Control theory, concerned as it was with manual tracking, originally belonged to the latter category; yet subsequent optimal control theory had the wider claim of modelling supervisory as well as manual control (Sheridan 1987), thus also pretending to be a general Human Performance model. I remember heated discussions with engineers, interested in performance, who did not see the necessity of testing the optimal control model. It was only a matter of parameter estimation; the model itself represented an engineering truth. Signal Detection Theory—with its measures of sensitivity and response bias—was the major unifying principle of Broadbent's (1971) second main book, Decision and Stress. Decision-related notions were fostered by merging signal detection theory and Bayesian revision of opinion. In 1964, I was introduced to all of this while undertaking post-doctoral work at the Michigan Human Performance Center, where Paul Fitts (ergonomics, reaction processes), Ward Edwards (decision-making), and (p.256) Arthur Melton (memory) were the main investigators. Edwards (1966) proposed a Bayesian model of two-choice reaction time, whereas decision-type models for recognition memory were under way. Many RT models were proposed, based upon various decision rules (Luce 1986) that were inspired more by mathematical convenience than by psychological reality. Consequently most were short lived, with the notable exception of Ratcliff's (1978, 1988) stochastic diffusion model which, in line with the mainstream, proposed a continuous signal-detection process feeding into an equally continuous random-walk decision process. This model appears to cover a wide domain of data on RT in traditional choice reactions, in target classification, and in same–different responses.
Limited capacity
The notion of limited capacity survived the fall of the Information metaphor and remained popular for two further decades. It simply shifted from a telecommunication device to a small-capacity digital computer. Practice effects were accounted for by ‘more efficient programming’ and strategic factors by ‘allocation policy’. There were differences between views—fixed computer capacity (Norman and Bobrow 1975), energetic ‘effort’ capacity with a variable limit (Kahneman 1973), and scarce fixed economical resources (Navon and Gopher 1979)—but all shared the common assumption that performance required some portion of the same general and undifferentiated processing capacity. The conceptual framework was particularly useful for researching perceptual-motor load in multiple task performance. Indeed, capacity theories reflected the applied interest in task load—or lack of task load—arising from rapidly developing industrial automatization. Operators were freed from boring repetitive work, but faced with equally boring supervisory control. A combination of different tasks might provide optimal operator load (e.g. Moray 1979).
I was never charmed by the capacity concept. It seemed to me void of psychological content, because, although in different amounts, all mental processes were supposed to consume the same general capacity. As computers became more powerful, the limited-capacity metaphor faded (Navon 1984; Neumann 1987). A new conceptualization of parallel-distributed neural nets had obvious references to Neurophysiology, and stressed the enormous brain capacity and the prevalence of in-parallel distributed processes. Again, behavioural evidence accumulated that time-sharing of different tasks is widely possible. There remain instances, though, in which a combination of seemingly simple decision processes creates a serious bottleneck. In a thoughtful analysis, Neumann (1987) argued that limits are met as soon as even simple task (p.257) demands conflict with respect to appropriate action. Response conflict evokes widespread inhibition so that appropriate responses can be selected and errors prevented. But this is far removed from the limited-capacity view.
Linear stages and choice reactions
Linear stage theory had roots in Donders (1868)—just as capacity theory had in Külpe (1905)—and was revived by Sternberg (1969) in his influential paper to the Donders Centenary at Eindhoven. A reaction process would consist of a number of successively operating processing stages, the outcome of a stage serving as input for the next one. Sternberg's method of inferring these stages from patterns of additive and interactive effects of experimental variables was called the Additive Factors Method (AFM). The method has little to say about the sequence of stages and about processes occurring within a stage, each stage needing its own processing model.
A main difference between stage and capacity theory is probably that the first suggests a modular and the second a holistic organization of performance, a contrast that also has classical roots (e.g. Goldstein 1938; Lashley 1929; see also Chapter 8). I vividly remember a 1980 stay with Daniel Gopher at the Technion, Haifa, Israel, and our intense discussions about the relative merits of either approach, including the possibility of multiple resources as potential bridge (Gopher and Sanders 1984). The AFM evoked considerable debate, which I reviewed in my Elements of Human Performance (Sanders 1998). The objections mainly concerned the credibility of some of Sternberg's original axioms, such as the unidimensional and serial nature of the stage sequence and the absence of feedback and feedforward processes. Some of these proved to be less compelling than they seemed at first sight. Thus, evidence from computer simulation suggested that the AFM still holds in most models of continuous flow between stages (McClelland 1979). As a result, the issue of discrete versus continuous processing evoked considerable research (e.g. Sanders 1990). The work of Miller (1988, 1993) and of Meyer and collaborators (1988) deserves mention, because of their methodological power in developing critical tests and conditions.
Parallel processing within stages poses no problem for the AFM, but stages operating in parallel are clearly problematic and would lead to inconsistent patterns of interactions and additivities of variables. A preponderance of parallel routes, of strategic manipulation, and of feedforward/feedback loops would entail chaotic outcomes of relations between experimental variables, and this would entirely discount the AFM. Most of my research during the 1970s and 1980s investigated the prospects of a stage structure in traditional choice reactions (Sanders 1980, 1990), suggesting a pattern of some six stages. (p.258) Stage robustness was illustrated by replicated findings that unusual conditions—such as sleep loss, long-term work, sedatives, and stimulants— left the stage structure intact and specifically affected some, but not other, stages. A one-dimensional processing flow was usually sufficient to account for the results; nevertheless, evidence for a second dimension occurred, under coherent conditions of highly arousing stimuli (Sanders 1983), thus setting a limit to the AFM. The AFM also encountered a limit in the case of separable multidimensional stimuli, confirming the widespread evidence for parallel processing of separate stimulus dimensions, and inspiring Miller (1982) to his view of asynchronous discrete processing. Given its powerful role in other areas of performance research, the lack of evidence for feedback/feedforward processes in choice reaction times may be most surprising, Yet loops within stages would not pose a problem for the AFM, whereas traditional choice processes may be too fast for larger loops to affect RT. It remains to be seen whether evidence for feedback/feedforward would arise in conditions of longer RTs, exceeding, say, two seconds.
In my view, the AFM has been instrumental in building cumulative evidence in favour of a modular processing flow in choice reactions. However, as I argued in Elements of Human Performance (Sanders 1998), the AFM never claimed to be a general unifying principle for all cognitive performance. It has a limited domain of application, and the obvious question is how far this stretches. Perhaps the question of domain extent is more useful than an endless quest for general unifying principles. Too small a domain, though, would of course limit the relevance of the AFM. It could well be that choice reaction tasks force subjects to one-dimensional processing, which, in turn, might have little bearing with regard to real-life tasks. I will return to this point in the next section.
There is evidence that the AFM domain exceeds elementary choice reaction tasks. In a large range of studies (Sanders 1998, pp. 196–213), I found that shifting the eyes from one stimulus to a another, presented at the right side of the first stimulus, is triggered when the identification stage of that first stimulus is completed. This conclusion derived from findings that effects on choice reaction time of variables, affecting stages up to and including the identification stage, did recur fully in the fixation time of the first stimulus, whereas the effects of variables affecting later stages did not.
Simple paradigms and real-life tasks
For nostalgic reasons I have spent much space discussing stages. Now, I will turn to the critical issue of generality. It used to be assumed that results on (p.259) simple experimental paradigms (choice reactions, short-term memory, tracking, etc.) were reliable reflections of real-life tasks, such as manual tracking to driving, and choice reactions as a prototype for interaction with certain environments. From the 1970s, Ergonomics became concerned with much more complex human-machine systems, raising doubts about the validity of simple paradigms (e.g. Flach 1990). Many felt that total task simulation would offer much better prospects, and increasingly sophisticated simulators of, say, flying an aircraft, driving a car, or manoeuvring a ship, found their way to the applied laboratories, whereas basic research was gradually abandoned. My Soesterberg Institute was no exception. Along the same lines, Allport (1980) expressed the gloomy view that common laboratory paradigms were artificial and irrelevant to real life and system design.
I addressed the dilemma in Attention and Performance X (Sanders 1984): simple paradigms are experimentally manageable and easy to analyse, but potentially artefactual and perhaps hardly valid ecologically; in contrast, full simulations, approaching the ‘richness of reality’, may be of direct applied significance but are too complex to enable detailed analysis or to choose between theoretical views. A caveat regarding the negative validity of a simulator is that a slight deviation from actual reality may render it counterproductive, as illustrated by some studies on the transfer of training from simulation to real driving (Moraal and Poll 1979). This does not imply that the relevance of simulation should be underrated (e.g. Sanders 1991). On the contrary, Performance Theory has always been interested in the description of complex everyday skills such as typing, driving, flying, human–computer interaction, and monitoring. In all of these cases simulations are obviously relevant.
As a way out, Gopher and I (1984) proposed a back-to-back strategy to test how far findings from simple paradigms can be applied to increasingly complex simulations. This can establish the domains of validity of models based on simple paradigms. It is recognized that a particular model or conceptual approach is never all-encompassing and must somewhere reach its limits. If findings in an elementary paradigm fail to hold in a more complex setting, the question arises as to how the situation has changed—perhaps in interesting ways.
The back-to-back idea concerned tests of the validity of results in artificial simple situations to increasingly closer approximations of the real world. This was actually a major research proposal but, to my regret, has not been followed up seriously. Instead, the last decades have shown a pronounced decline in common interests between Cognitive Psychology and Ergonomics, perhaps due to another major development, to which I will now turn.
(p.260) The biological revolution
Some readers may consider all that has been said so far as belonging to ‘then’; for these, ‘now’ is what is commonly called the Biological Revolution, the consequence of developments in brain imaging techniques—positron emission tomography (PET), functional magnetic resonance imaging (fMRI), and magnetic electroencephalography (MEG)—during the last 10–15 years. These techniques allow more precise and extensive measurements of brain correlates during performance of simple paradigms. PET and fMRI recordings suggest that the dream of establishing the anatomical and physiological basis of mental processes may be coming true. Two main features of the previous paradigm shift from Behaviourism to Cognitive Psychology were a new conceptual framework and a neglect of older work. Now behavioural, or technological, notions are being replaced by brain terminology. I first encountered this during Donald Broadbent's retirement Festschrift in 1993, where Mike Posner— an early and important leader in this field—reported PET studies in relation to his three components of attention: alertness, conscious attention, and orienting. Increased blood flow in the frontal lobe appeared to be related to alertness (e.g. maintaining sustained attention), and conscious attention (i.e. investing effort) had connected active brain structures centred around the anterior cingulate gyrus, whereas orienting had a prime site at the posterior parietal lobe (Posner 1993). Posner and his collaborators were careful to relate their brain imaging measures to performance on well-researched elementary paradigms, but clearly their terminology shifted from behavioural to neuro-scientific concepts (i.e. ‘anterior and posterior’ attention systems).
I would be the last to underrate the relevance of the new brain imaging techniques to theoretical descriptions of mental processes and human performance. Uncovering brain areas involved in the various types of mental activity obviously involves more than correlating brain structures to behaviour. It adds to theory and enables control of intervening variables that have, so far, been hard to trace. Aside from extreme statements like ‘I am my brain’—which raise considerable philosophical problems—I am convinced that the biological approach will enable rapid progress in relating brain function to behaviour. It would be disappointing, though, to see a rupture with the past, as happened during the 1950s. This may not occur as long as the new movement sticks to the experimental rigour that characterized behavioural experimentation during the last decades. The problem could arise that the fascination with brain images leads to neglect of rigorous behavioural experimentation, which is crucial to render the pictures meaningful. There are many experimental studies in which the cognitive tradition of the last fifty years has (p.261) been enriched rather than abandoned. I was pleased to read Posner's plea (see Chapter 15) for combinations of behavioural, electrical, and haemodynamic methods. This will not be easy, but is not impossible. Not surprisingly, I was also pleased with Dehaene's (1996) work on the AFM, showing converging evidence from brain and behavioural measures. Stages were laid out in separate brain areas, supporting a modular nature of information processing.
These new studies are hopeful signs of a gradual incorporation of the ‘then’ into the ‘now’. Time will tell whether this will happen or whether ‘then’ will be abandoned in favour of a new ‘now’. This latter move is not uncommon in science and may be useful so long as the antithesis, evoked by a paradigm shift, is eventually followed by a higher-order synthesis.
I end by expressing the hope that, in the course of events, psychologists will not be too fascinated by the new intellectual climate of Cognitive Neuroscience so as to continue neglecting human factors and leaving that field completely to engineers. The point is that I do not envisage combined fMRI–simulator research. Thus, the Soesterberg Institute has chosen to ignore Neuroscience and to concentrate solely on applicable behavioural research. That is where their money comes from. I am not convinced, therefore, by the term Neuroergonomics as a weak attempt to link human factors to the high tide of the biological revolution.
No comments:
Post a Comment