Cognitive ethology 1 COGNITIVE ETHOLOGY: GIVING REAL LIFE TO ATTENTION RESEARCH

Studies of attention, often conducted in artificial laboratory experiments, may have limited validity when performance in the natural world is considered. For instance, for over two decades investigations of "reflexive" and "volitional" attention have tended to be grounded in methodologies that do not capture the demands of attention in everyday life. Recent studies suggest these laboratory investigations have lost touch with real life contexts and accordingly they may generate fundamental misunderstandings regarding the principles of human attention and behaviour. We identify the basic assumptions of laboratory research that has led to this state of affairs, and suggest a new set of assumptions that lead to a new research approach, which we call "cognitive ethology". The implication is that if one is to understand human attention in everyday life then research needs to be grounded in the natural world and not in experimental paradigms. Cognitive ethology 3 Patrick Rabbitt has never been one to accept the status quo. None of his students escaped this strong part of his personality, and for many of us, it helped to define who we became. One of Pat's most enduring lessons centered around the idea that the models in psychology are largely "snapshots" of idealized static cognitive sytems. Pat often punctuated this point by noting that experimental psychologists routinely throw away the first set of trials in an experiment as "practice trials". As Pat was fond of saying, these initial trials routinely produce the largest performance changes that will be found in a study. And yet, these initial trials are discarded and left unanalyzed because researchers desire stable systems that can be controlled, manipulated and modeled. This issue raised by Pat has plagued me over the years because seeded deeply within it was the notion that experimental psychologists were not really getting it right, nor were they going to get it right doing research the way they were doing it. That is, research in the pursuit of stability and control was not going to tell us what we really want to know – how people function in real life where things are highly variable and often outside the domain of experimental control. The present chapter represents an initial attempt by my lab to take on Pat's challenge to get things right and to learn how people function in real life. Our particular area of interest is human attention. A LABORATORY PARADIGM FOR STUDYING ATTENTION We begin by closely considering one of the most well known laboratory paradigms for studying attention: The Posner cueing paradigm (Posner, 1978; 1981). In this paradigm, a central fixation dot that is flanked by two boxes is presented at the center of a computer screen. The task is simply to press the spacebar on a computer keyboard as quickly as possible when a visual target object appears inside one of the boxes. This target object is preceded by an attentional cue, which is either a brief peripheral flash surrounding one of the two boxes or a central arrow pointing toward one of the boxes (see Figs. 1a and 1b). The standard and highly robust finding is that a target is detected fastest when it appears in the box that was cued by the peripheral flash or central arrow. On the assumption that the brain processes attended items more quickly than unattended items, it is concluded that target detection time was speeded for a target at the cued location because attention had been committed to the box that was cued. --Insert Figure 1 about here --Countless studies of this sort have led to the conclusion that there are two categories of attention, exogenous (reflexive) attention and endogenous (volitional) attention, both of which can be manipulated and measured by the Posner paradigm. When one of the two boxes is flashed briefly, as depicted in Figure 1a, attention is considered to be oriented reflexively to the box that brightened. This attention shift is thought to be reflexive because people are faster to detect a target in the cued box even when flashing the box does not predict where the target will occur (e.g., the target appears in the cued box 50% Cognitive ethology 4 of the time and in the uncued box 50% of the time). Attention can also be oriented volitionally. When a central arrow points toward one of the two boxes, as depicted in Figure 1b, attention is thought to be oriented volitionally to the box pointed at by the arrow. This attention shift is considered volitional because people are thought to be faster to detect a target in the cued box only when the arrow predicts where the target will occur (e.g., the target appears in the cued box 80% of the time and in the uncued box 20% of the time). THE POSNER PARADIGM AND REALITY It is perhaps instructive at this point to compare the displays used in the Posner paradigm, which are shown in Figure 1, with the type of scenes encountered outside the laboratory, examples of which are shown in Figure 2. Even a cursory comparison begs the following type of question: To what extent does the simple, impoverished and highly artificial experimental task of detecting a light within a cued or uncued box have to do with the many complex, rich real-life experiences that people share? In other words, what does the cuing paradigm have in common with everyday situations such as children playing under the supervision of an adult, a father teaching a son how to tie and tie, or a little girl watching another two friends kiss? On the face of it, not very much. --Insert Figure 2 about here --Even a cursory look at the naturalistic scenes shown in Figure 2 yields many hypotheses regarding attention that are unlikely to ever be generated from laboratory studies of attention. For instance, inspection of the scenes suggests that people’s direction of gaze might be a critical cue, perhaps even a reflexive one, for orienting attention in the real world. In 1998, Chris Friesen and I tested this idea by modifying the Posner paradigm in two important ways. First, arrows pointing to the left and right were replaced by a schematic face that looked left or right. Second, the predictive value of the central cue was eliminated; that is, eye direction did not predict where a target item would appear (see Fig. 1c; also Friesen and Kingstone, 1998). Note that because the eyes were centrally located and spatially nonpredictive, the traditional line of thinking predicted that they would not lead to shifts of attention. The result was that, contrary to this prediction, eye gaze did trigger shifts of attention, with all types of responses (target detection, target localization and target identification) being enhanced almost immediately for targets that appeared at the gazed-at location. This rapid onset of an attention effect, and the fact that it occurred in response to a spatially nonpredictive stimulus, demanded the conclusion that the attentional shift was reflexive (Cheal & Lyon, 1991; Müller & Rabbitt, 1989; Jonides, 1981). This conclusion, that centrally nonpredictive eye-gaze can direct trigger reflexive shifts of attention, led us to reconsider the fundamental notion that arrows only direct attention when they are spatially predictive, i.e., the shifts are volitional in nature. Since a classic study by Jonides (1981, Experiment 2), which failed to find positive evidence that nonpredictive central arrows will trigger a shift in attention, researchers have assumed that arrows do not produce a shift in attention unless they predict where an item will appear. We considered the possibility, for the first time, that central arrows might be much like eyes and Cognitive ethology 5 that they might orient attention even when they are not predictive. We tested this possibility by replacing spatially nonpredictive eyes with spatially nonpredictive arrows (Fig. 1c; Ristic, Friesen, & Kingstone, 2002). The results were unequivocal. People attend to where arrows point even when they know that the arrows do not predict where a target will appear. In other words, like eyes, arrows produce a reflexive shift in attention to the cued location. This result has been confirmed by several investigators (Tipples, 2002; Friesen, Ristic & Kingstone, 2004; Hommel et al. 2001) establishing that attention can be directed reflexively to locations indicated by nonpredictive arrows. The failure of Jonides (1981) to obtain this significant outcome may be attributed to several factors, including his arrow cue being hard to discriminate (it was flashed for only 25 ms) and his study lacking sufficient experimental power (fewer than 10 participants were tested and RTs at the cued location were based on less than 20 trials). We believe that the recent experiments on eye gaze and nonpredictive central arrows have important implications for laboratory studies of attention. Specifically, the findings indicate that the Posner cuing paradigm is fundamentally flawed. Central cues such as eyes and arrows, which were assumed to tap volitional attention, actually do not reflect volitional orienting. Thus, what was taken as a fundamental truth turned out to be completely in error. How could this mistake happen? How could this error be overlooked for over 20 years of research? We believe that the answer rests with the fact that laboratory research in the field of attention is based on two crucial assumptions – both of which are problematic. ASSUMPTIONS UNDERLYING LABORATORY RESEARCH Studies of attention in the laboratory are grounded on two basic assumptions. One is that human attention is subserved by processes that are stable across different situations. For example, the processes that are studied in the lab are assumed to be the same as the processes that are expressed in the real world. Second, one can maximize analytical power of a process by minimizing all variability in a situation save for the factor being manipulated. Note that the first assumption enables one to lay claim to processes in the real world without ever leaving the lab. And the second assumption demands that the laboratory situation be as controlled as possible. Ironically, the combination of these two assumptions has the effect of driving researchers further away from real life situations into highly contrived and artificial laboratory environments, all the while seducing the investigators into the belief that they are getting closer and closer toward a true understanding of how attention operates in real life. While the assumptions of process-stability and situational-control are commonly held and readily applied in studies of attention, adopting them comes with a high degree of risk. The assumption of stability, for example, eliminates any need or obligation by the scientist to confirm that the factors being manipulated and measured in the lab actually express themselves in the real world. The field does of course check routinely that the effects being measured are stable within the lab environment, by demanding that results in the lab be replicable. Unfortunately, a Cognitive ethology 6 result that is stable within a controlled laboratory environment does not necessarily entail that it is stable outside the lab. Indeed, there are many examples within the field of human attention indicating that even the most minor changes within a laboratory situation will compromise the replicability of an effect (e.g., Soto-Faraco, Morein-Zamir & Kingstone, in press; Kingstone et al., 1993). Indeed, the fragility of laboratory findings whereby small changes in an experimental situation results in radically different experimental outcomes, should perhaps not be particularly surprising. There is now a growing body of literature indicating that "process stability" is tied intimately to the specifics of a situation, with brain re-configurations occurring continuously in response to even subtle environmental changes. Neisser (1976) referred to the dynamic reconfigurations as “schemata”, Mosell (1996) referred to them as “task-set reconfigurations”, and Di Lollo et al. (2001) have referred to them as “configurable input filters”. Thus, while the assumption of process stability is remarkably convenient because it allows one to think that the work in the lab has some relevance to real world situations, the fact is that there is a very real risk, indeed a high likelihood, that the process being studied in the lab will not exist outside the lab. Associated with the above is the fantastic risk of wasting many years and dollars conducting research that is replicable but which has little, if anything, to do with cognitive processes beyond the lab in real life environments. This follows because by simply assuming a process is stable across situations – specifically from the lab to real life –means that one never has to check whether the results in the lab are at all relevant to real life performance. This seems to be the reason why for more than 20 years researchers failed to notice that the Posner cuing paradigm, while producing results that were highly replicable, did not measure what people believed they were measuring. It was only after considering the results derived in the Posner paradigm against real life situations (e.g., how eyegaze might influence attention) that this error was detected (Kingstone et al, 2003). The problem gets even worse, however, because it is often the case that attempts to test the first assumption of stability against real life situations are immediately met with apparently insurmountable obstacles. These obstacles arise from the second assumption of experimental control. The first obstacle is that processes in the lab often become defined by the experimental controls that were used to examine them. In addition, if this obstacle were somehow overcome and the controls that define a phenomeon were reproduced in the real world, a researcher is immediately posed with the challenge of making the case that the data collected in a real life situation is in fact a manifestation of the process that was measured previously in the lab. This is a daunting, and perhaps ultimately impossible, obstacle to surmount. This is because data that are attributed to a particular process in the lab can always be re-attributed to other factors that were left free to vary in a real life situation. We believe that these two obstacles actually reflect a more fundamental problem that arises from the assumption of control when it is applied to nonlinear systems like human cognition and attention. While experimental control can be Cognitive ethology 7 effective at revealing basic characteristics of simple linear systems, general systems theory has established that experimental control is unlikely to be effective at revealing important characteristics of complex, non-linear systems such as the human attention system. This is because certain characteristics of complex systems are only revealed, or emerge, when several variables vary together in highly specific ways (see Ward, 2002; Weinberg, 1975). This is precisely what is not allowed to occur in highly controlled laboratory situations. Considered together, it appears that there are both practical and principled reasons to question whether selecting tasks based on the assumptions of stability and control is likely to inform us about cognitive processes as they are expressed in real life situations. Moreover, it is sobering to contemplate the fact that these two assumptions can lead researchers to unwittingly and unthinkingly commit their lifetime to studying processes that may be expressed only within artificial lab situations.