Everyday Activities

1 Introduction

Much of behavior consists of everyday activities—mundane tasks that are nevertheless essential such as dressing, grooming, commuting, and preparing meals. These activities are performed on a regular—often daily—basis, in familiar environments, and in stereotyped ways, and often with limited overt attention. Yet, the apparent banality of such activities masks both their importance from a behavioral perspective and their complexity from a cognitive perspective.

With respect to the former, for example, successful performance of everyday activities is an essential part of independent living. Yet, and as testament to the latter, such performance typically degrades in old age. Thus, a major challenge brought about by demographic change is elderly care (Nicholas & Smith, 2006). There currently is a lack of qualified personnel and facilities to take care of the ever increasing part of the population that can no longer live completely independently. A full understanding of the cognitive complexities and demands of everyday activities thus has substantial practical benefits, as one way to address the challenge of elderly care is to enable people to manage necessary everyday activities (e.g., setting a table, cooking, cleaning up) without intervention from human care takers for longer into old age, and ideally for their whole life. To achieve this, systems are required which assist humans in their everyday activities based on a deep understanding of the mechanisms underlying everyday activity performance.

Cognitive Science can play a key role in enabling everyday activity assistance, and thus, tackling a major challenge arising from demographic change. Cognitive science is in a unique position to generate insights about the (neural) mechanisms underlying the abilities involved in everyday activities, how they can be measured, how these abilities develop, and how they may break down due to aging and cognitive impairment. Furthermore, computational cognitive science allows artificial systems to be equipped with the necessary knowledge and abilities to assist people in their everyday activities (e.g., by building robots or personal assistants that master everyday activities). At the same time, cognitive science can profit greatly from an in-depth investigation of everyday activities. First, such investigations shed light on how the cognitive system as a whole functions in naturalistic tasks that require the coordination of multiple cognitive faculties (i.e., beyond pure memory, attention, or language tasks). Second, research on everyday activities can serve as a “condensation point” for focused collaborations between research communities from various disciplines and end users.

Everyday activities may be viewed as “complex tasks” in the sense of Newell (1973). Accordingly, they constitute a domain in which the integration of various cognitive abilities can be investigated. Apart from providing insight into the mechanisms of integration, such an integrative approach facilitates honing theories and computational realizations of the individual abilities involved. Moreover, we believe that the investigation of everyday activities is necessarily an interdisciplinary undertaking that, in particular, also needs to and can help to bridge the gap between research on natural and artificial cognitive systems. In the light of the recent critique of cognitive science as an interdisciplinary endeavor (Núñez et al., 2019), investigation of everyday activities holds promise as a facilitator for uniting the contributing disciplines of cognitive science.

2 The contributions

Against the above background, this topic aims to: (a) promote everyday activities as a research topic in its own right within cognitive science, (b) bring together everyday activity-related research from a wide range of disciplines, and (c) lay the groundwork for thinking about how different abilities combine and are integrated to yield an overall cognitive architecture that supports everyday activities. To this end, this topic consists of eight papers that span different aspects of everyday activities, ranging from neuroscience through philosophical considerations and implications to lessons from robotics.

Starting at the lowest level, one fundamental question concerns the neural representations that support the execution of sequential action. Such representations must encode task context, first in order to distinguish between similar or overlapping stimulus response mappings, and second, because overlapping sequences cannot be generated through simple response chaining (Lashley, 1951). This is the question addressed by Shahnazian, Senoussi, Krebs, Verguts, and Holroyd (in this issue), who present a reanalysis of brain imaging data obtained from participants completing a lab-based tea/coffee preparation task. The task requires that participants make six binary decisions in sequence to produce a target beverage, and allows the authors' to differentiate between temporal (i.e., order) and contextual (task) information. Their reanalysis highlights the importance of the inferior-temporal gyrus and lateral regions of the prefrontal cortex in maintaining both types of information, with regions on the left appearing to be more strongly involved in the encoding of temporal, as opposed to contextual, information.

Working at a higher, behavioral, level, Yanaoka & Saito (in this issue) consider how the abilities required to perform everyday activities develop throughout childhood. As their starting point, they take the proposal of Norman & Shallice (1986) of separable systems for the control of routine versus non-routine behaviors. Their interest lies in how routines (or action schemas) are acquired and how children learn to modulate or override these routines in relatively simple non-routine situations. They review a series of developmental studies and conclude that changes in the routine system are driven by both statistical factors (relating to transition probabilities between successive actions) and developing executive functions (as required, for example, in intention maintenance during hierarchical action and prepotent response inhibition). Moreover, they cite evidence for increasing flexibility with age in simple routines, and argue for a graded representation account where “clean” representations are required for flexible behavior, and where acquired routines help children detect (and ignore) goal-irrelevant information.

The development of flexibility in action control is also demonstrated by a remarkable ability of adult humans, namely, to accurately perform everyday activities based on vague and underspecified verbal utterances. A nice case in point is, for example, in recipes asking you to add eggs to whatever you are cooking, which, however, does normally not mean adding the whole egg. Besides being able to determine the necessary actions and action sequences based on the utterance, properly perceiving the meaning of an utterance often requires correctly interpreting vague verbal terms. Meder, Mayrhofer, and Ruggeri (in this issue) present empirical and modeling work that sheds light on how the human ability to interpret vague frequency and probability terms develops during childhood. One key finding is that younger children tend to interpret such terms more dichotomously, but that starting from the age of nine children exhibit more nuanced interpretations that are already very similar to those of adults. Arguably, flexibility in interpretations is supported by the same developmental mechanisms highlighted by Yanaoka & Saito (in this issue). Furthermore, Meder and colleagues propose a modeling approach that allows predicting the outcome of more complex inferences based on the developing interpretation of frequency and probability terms.

Failures in performing everyday activities—action slips—are characteristic of the action domain. Anecdotally, they have been observed in children as young as four (by which time routines such as washing, dressing, and toileting are established). Mylopoulos addresses the question of how such failures in performing everyday activities can occur despite the actor's intention to do otherwise. She stresses the importance of motor schemas as well as of selective and sustained focal attention (and the failure thereof) for understanding action slips, and explains how commonly observed types of action slips can arise from the interplay of these components. In doing so, her work dovetails with the developmental results reviewed by Yanaoka & Saito (in this issue), but her contribution also sheds new light on everyday activities, intentional action, and the nature of intention, more generally.

Parallels between action and language have frequently been made in the literature (e.g., Glenberg & Gallese, 2012). Thus, both appear to involve sequence, hierarchy, and argument structure. In this regard, the results of Shahnazian et al. (in this issue) concerning left inferior frontal gyrus, a region traditionally associated with language production, are tantalizing. Mis & Giovannetti (in this issue) highlight commonalities between everyday activities and language production, also being a sequential activity, which have so far received relatively little attention. Based on existing modeling work, the authors argue for functional similarities between, and overlap of processing in, these two domains. Against this background, Mis & Giovannetti (in this issue) review existing approaches for treating aphasia and relate them to the functional decline of performance of everyday activities observed in aging and cognitive impairment. The review suggests that several existing aphasia interventions are promising candidates for transfer to the treatment of impairments of everyday functioning.

The control of action is also an active area of research within robotics. Robotics work forces a degree of engagement with the environment that can be elided or completely neglected in less applied areas of cognitive modeling. Robotics work is also forced to provide a practical solution to the question of where motor control ends and action control begins. de Kleijn, Sen, and Kachergis (in this issue) consider a problem at this interface, namely, how an agent might learn to perform a simple touching task, where cued targets may appear at any of a number of preset locations. Some human participants minimize reaction time on this task by positioning their hand between targets, so it is equidistant from each—the so-called “centering” strategy. de Kleijn et al. (in this issue) train a robot arm controller on the task, and find that this behavior emerges when the controller is first allowed a period of free movement before imposing a cost on movement. If the cost is imposed early or late, the controller fails to exhibit the human-like centering behavior. The authors like this to critical periods seen in other aspects of development, but also link their findings to staged learning more generally, where learning of later abilities is dependent upon learning of earlier abilities (as in the theory of zones of proximal development: Vygotsky, 1978). The work relates to the earliest stages of the development of action control, but is arguably a precursor to the developmental findings summarized by Yanaoka & Saito (in this issue). Indeed, the behaviors reported by Yanaoka & Saito (in this issue) depend upon (or build upon) the earlier acquisition of simpler reaching behaviors.

At a task level, Caccavale & Finzi (in this issue) consider how insights from the cognitive psychology of sequential action selection can inform the design of service robots. They argue that the use within service robots of a schema-based system for action control, where known routines may be modulated by executive control, has multiple advantages. Such an approach provides behavioral flexibility that is essential in a joint action (as would be required when a robot is assisting in elderly care), where different elements of a shared intention might be performed by different agents. It also supports learning from human guidance. In both cases, it is the intention that serves to link individual actions together within a sequence, and that allows recall of those sequential actions when appropriate.

Vernon et al. also consider the execution of everyday activities from a cognitive robotics perspective. Their focus is on the gross cognitive architecture required to support such activities. They specifically consider how the challenges of everyday activities are addressed by their cognitive robot abstract machine (CRAM) architecture. CRAM comprises a set of interacting modules (plan, perception, and action executives; a metacognition process; a knowledge base that embodies reasoning; and an episodic memory). Some of the concepts within CRAM, for example, that of a generalized action plan, which is “a computational encapsulation of an underdetermined action description that can be deployed in many everyday contexts,” are closely related to those developed in other literatures, but Vernon et al.'s work goes further by demonstrating both the sufficiency of CRAM as a model of a large part of human everyday activities, and the limitations of CRAM when it comes to what the authors' term complex everyday activities (i.e., those where the constraints on the order of steps are weak, meaning that there are multiple degrees of freedom at each step). To address these weaknesses, Vernon et al. appeal to the situation model framework—a framework that distinguishes between habitual (i.e., routine) and deliberative (i.e., non-routine) behaviors in a way not dissimilar to that proposed in the cognitive psychology literature.

3 Conclusion

The picture that emerges from the various contributions is that everyday activities draw upon both precompiled or cached routines (“action schemas”) and systems for general problem solving, including those involved in perception, motor and action planning, spatial reasoning, and the generation and maintenance of intentions. The underlying mechanisms develop throughout childhood, with statistical learning and the maturation of executive functions playing key roles. Returning to the themes of the introduction of this overview, one can see that the various contributions span the key abilities, from the generation and maintenance of intentions to monitoring and error correction. For example, the generation and maintenance of intentions is central to the contributions of Mylopoulos, Mis & Giovannetti, Caccavale & Finzi, Shahnazian et al., Vernon et al., and Yanaoka & Saito, while the role of perception is touched upon by Meder et al., Caccavale & Finzi, and Vernon et al., and monitoring and error correction is a critical element in the discussion of Mylopoulos, but also present in the work of Caccavale & Finzi, de Kleijn et al., and of Mis & Giovannetti. Cross-cutting these key abilities is the role of learning and development.

The overlaps in topics support our initial contention that everyday activities can serve as a “condensation point” for focused collaborations within cognitive science. The robotic work highlights the potential of work on everyday activities to inform, and improve, service robots in the care sector. At the same time, the contributions highlight that there are many outstanding questions, concerning, for example, the neural substrates that support flexible everyday activities, the relation between the cognitive systems that support everyday activities and language production, the role of goals and subgoals within everyday activities, the relation between CRAM and the architecture of Caccavale & Finzi, and the interrelation of cognitive subsystems that allow flexible, and even opportunistic, everyday behavior. Everyday activities thus remain a fruitful ground for research.

4 Papers in this Topic