Skip to main content
Social Sci LibreTexts

5.12: What is Embodied Cognitive Science?

  • Page ID
  • To review, the central claim of classical cognitive science is that cognition is computation, where computation is taken to be the manipulation of internal representations. From this perspective, classical cognitive science construes cognition as an iterative sense-think-act cycle. The “think” part of this cycle is emphasized, because it is responsible for modeling and planning. The “thinking” also stands as a required mentalistic buffer between sensing and acting, producing what is known as the classical sandwich (Hurley, 2001). The classical sandwich represents a modern form of Cartesian dualism, in the sense that the mental (thinking) is distinct from the physical (the world that is sensed, and the body that can act upon it) (Devlin, 1996).

    Embodied cognitive science, like connectionist cognitive science, arises from the view that the core logicist assumptions of classical cognitive science are not adequate to explain human cognition (Dreyfus, 1992; Port & van Gelder, 1995b; Winograd & Flores, 1987b).

    The lofty goals of artificial intelligence, cognitive science, and mathematical linguistics that were prevalent in the 1950s and 1960s (and even as late as the 1970s) have now given way to the realization that the ‘soft’ world of people and societies is almost certainly not amenable to a precise, predictive, mathematical analysis to anything like the same degree as is the ‘hard’ world of the physical universe. (Devlin, 1996, p. 344)

    As such a reaction, the key elements of embodied cognitive science can be portrayed as an inversion of elements of the classical approach.

    While classical cognitive science abandons Cartesian dualism in one sense, by seeking materialist explanations of cognition, it remains true to it in another sense, through its methodological solipsism (Fodor, 1980). Methodological solipsism attempts to characterize and differentiate mental states without appealing to properties of the body or of the world (Wilson, 2004), consistent with the Cartesian notion of the disembodied mind.

    In contrast, embodied cognitive science explicitly rejects methodological solipsism and the disembodied mind. Instead, embodied cognitive science takes to heart the message of Simon’s (1969) parable of the ant by recognizing that crucial contributors to behavioral complexity include an organism’s environment and bodily form. Rather than creating formal theories of disembodied minds, embodied cognitive scientists build embodied and situated agents.

    Classical cognitive science adopts the classical sandwich (Hurley, 2001), construing cognition as an iterative sense-think-act cycle. There are no direct links between sensing and acting from this perspective (Brooks, 1991); a planning process involving the manipulation of internal models stands as a necessary intermediary between perceiving and acting.

    In contrast, embodied cognitive science strives to replace sense-think-act processing with sense-act cycles that bypass representational processing. Cognition is seen as the control of direct action upon the world rather than the reasoning about possible action. While classical cognitive science draws heavily from the symbol-manipulating examples provided by computer science, embodied cognitive science steps further back in time, taking its inspiration from the accounts of feedback and adaptation provided by cybernetics (Ashby, 1956, 1960; Wiener, 1948).

    Shapiro (2011) invoked the theme of conceptualization to characterize embodied cognitive science because it saw cognition as being directed action on the world. Conceptualization is the view that the form of an agent’s body determines the concepts that it requires to interact with the world. Conceptualization is also a view that draws from embodied and ecological accounts of perception (Gibson, 1966, 1979; Merleau-Ponty, 1962; Neisser, 1976); such theories construed perception as being the result of action and as directing possible actions (affordances) on the world. As such, the perceptual world cannot exist independently of a perceiving agent; umwelten (Uexküll, 2001) are defined in terms of the agent as well.

    The relevance of the world to embodied cognitive science leads to another of its characteristics: Shapiro’s (2011) notion of replacement. Replacement is the view that an agent’s direct actions on the world can replace internal models, because the world can serve as its own best representation. The replacement theme is central to behaviour-based robotics (Breazeal, 2002; Brooks, 1991, 1999, 2002; EdsingerGonzales & Weber, 2004; Grey Walter, 1963; Sharkey, 1997), and leads some radical embodied cognitive scientists to argue that the notion of internal representations should be completely abandoned (Chemero, 2009). Replacement also permits theories to include the co-operative interaction between and mutual support of world and agent by exploring notions of cognitive scaffolding and leverage (Clark, 1997; Hutchins, 1995; Scribner & Tobach, 1997).

    The themes of conceptualization and replacement emerge from a view of cognition that is radically embodied, in the sense that it cannot construe cognition without considering the rich relationships between mind, body, and world. This also leads to embodied cognitive science being characterized by Shapiro’s (2011) third theme, constitution. This theme, as it appears in embodied cognitive science, is the extended mind hypothesis (Clark, 1997, 1999, 2003, 2008; Clark & Chalmers, 1998; Menary, 2008, 2010; Noë, 2009; Rupert, 2009; Wilson, 2004, 2005). According to the extended mind hypothesis, the world and body are literally constituents of cognitive processing; they are not merely causal contributors to it, as is the case in the classical sandwich.

    Clearly embodied cognitive science has a much different view of cognition than is the case for classical cognitive science. This in turn leads to differences in the way that cognition is studied.

    Classical cognitive science studies cognition at multiple levels: computational, algorithmic, architectural, and implementational. It typically does so by using a top-down strategy, beginning with the computational and moving “down” towards the architectural and implementational (Marr, 1982). This top-down strategy is intrinsic to the methodology of reverse engineering or functional analysis (Cummins, 1975, 1983). In reverse engineering, the behavior of an intact system is observed and manipulated in an attempt to decompose it into an organized system of primitive components.

    We have seen that embodied cognitive science exploits the same multiple levels of investigation that characterize classical cognitive science. However, embodied cognitive science tends to replace reverse engineering with an inverse, bottom-up methodology, as in forward engineering or synthetic psychology (Braitenberg, 1984; Dawson, 2004; Dawson, Dupuis, & Wilson, 2010; Pfeifer & Scheier, 1999). In forward engineering, a set of interesting primitives is assembled into a working system. This system is then placed in an interesting environment in order to see what it can and cannot do. In other words, forward engineering starts with implementational and architectural investigations. Forward engineering is motivated by the realization that an agent’s environment is a crucial contributor to behavioral complexity, and it is an attempt to leverage this possibility. As a result, some have argued that this approach can lead to simpler theories than is the case when reverse engineering is adopted (Braitenberg, 1984).

    Shapiro (2011) has noted that it is too early to characterize embodied cognitive science as a unified school of thought. The many different variations of the embodied approach, and the important differences between them, are beyond the scope of the current chapter. A more accurate account of the current state of embodied cognitive science requires exploring an extensive and growing literature, current and historical (Agre, 1997; Arkin, 1998; Bateson, 1972; Breazeal, 2002; Chemero, 2009; Clancey, 1997; Clark, 1997, 2003, 2008; Dawson, Dupuis, & Wilson, 2010; Dourish, 2001; Gallagher, 2005; Gibbs, 2006; Gibson, 1979; Goldman, 2006; Hutchins, 1995; Johnson, 2007; Menary, 2010; Merleau-Ponty, 1962; Neisser, 1976; Noë, 2004, 2009; Pfeifer & Scheier, 1999; Port & van Gelder, 1995b; Robbins & Aydede, 2009; Rupert, 2009; Shapiro, 2011; Varela, Thompson, & Rosch, 1991; Wilson, 2004; Winograd & Flores, 1987b).

    • Was this article helpful?