Adventures in Cognitive Science 2: What's that Thing called Embodiment?
The second paper I am going to present is pretty much a follow-up to the first one. It was written by Tom Ziemke, a full professor of cognitive science at the University of Skövde and it is called What's that Thing called Embodiment. Like Wilson, Ziemke is also a proponent of embodied cognition and highlights the general agreement that humans are embodied cognisers (which is true today as it was back in 2003 when this article was published), but much less agreement on how this body looks like, or if an artefact could also be considered embodied. Some researchers claim that “intelligence” can not exist just in the form of an algorithm, but requires a physical body, but this view is actually much less widespread than you might think. Similar to Wilson he identifies and evaluates six different notions of embodiment (although interestingly enough he does not really pick up on the differentiation between on-line and off-line cognition that was one of the central topics of Wilson's paper):
- structural coupling between agent and environment
- historical embodiment as the result of a history of structural coupling
- physical embodiment
- organismoid embodiment i.e. organism-like bodily form (e.g., humanoid robots)
- organismic embodiment of autopoietic, living systems
- social embodiment
Embodiment as structural Coupling
This is the broadest notion of embodiment: a system is embodied, if it is somehow structurally coupled to its environment. Structurally coupled means that there exist channels between the embodied system and the environment with which the embodied system can influence the environment and the environment can influence the embodied system. This notion does not necessarily require a body and so also software systems can be embodied and “intelligent”. A drawback of this notion is, that it is so broad that even a stone that lies on a field and gets blown by the wind would be an embodied system, since it follows this definition of embodiment. But still cognitive scientists do not attribute stones with cognition, even tough according to this definition a stone on a field would be a cognitive system.
Cognitive systems are not only structurally coupled in the present, but their embodiment is also a reflection of their past. The embodiment that follows from being in an environment is necessarily effected by the history of this structural coupling. Some cognitive scientists go so far to say that the adaption of a system to its environment is the definition of embodiment.
This notion is more restrictive than the previous one, but it still applies to living systems and does not exclude non-physical bodies. It is useful to identify the developmental roots of embodied cognition.
This notion of embodiment does exclude software agents and claims that embodied systems require a physical body (if it's held in a strong sense). Ironically this definition of embodiment still views a stone on a field as a cognitive system, since it does not discriminate between living and non-living bodies, it only has to be physical. In its weaker sense physical embodiment does include software agents, but only if they either adapted to their environment through self-organisation or are somehow connected to the physical world with sensors.
Similar to physical embodiment, but more restrictive, since it limits cognitive systems to have organism-like bodies, so physical bodies (no pure software agents) which have the same or similar features as living bodies, especially in terms of receiving inputs and performing outputs (so no stone on a field). A good example for this type of embodiment are humanoid robots: even though they are artificial systems their bodies receive inputs which are similar to those received by living systems, and they perform actions similar to living systems and are therefore organismoid embodied.
The organismic notion of embodiment is a very restrictive one, since it says that cognition is limited to organisms, so to living bodies. This notion of cognition has its roots in the work of Jakob von Uexküll and was picked up and extended in the 1980ies by Humberto Maturana and Francisco Varela (as you probably noted so far, I tried to avoid giving too many links to authors, other than the one of the paper I'm presenting, in this case I have to make an exception, since the work of Maturana and Varela is probably the foundation of my master program in cognitive science). They claim that the main difference between living organisms and man-made machines is, that organisms are autonomous and autopoietic. Autonomous means that they are can do whatever they want, as opposed to heteronomous machines which can only “follow the rules” of their constructor. Autopoietic means that a living organism maintains itself and is also able repair itself and able to reproduce, something machines can not do, they are allopoietic. When those theories were formulated, technology was far from what it is today and today it is possible to think of machines that could overcome those limitations.
This notion is different from the other five, since it does not look at what kind of body is required for embodiment, but focuses on the role of social interactions in embodiment, how social interaction influences human cognition. Social “inputs” do not only produce cognitive states, but also bodily states. The question if social embodiment also requires a physical body or could be also realised as a software agent is open.
Ziemke discussed six different notions of embodiment, looking at how the embodied body would look like. The organismic notion of embodiment is the most strict one, since it eliminates all kinds of non-living bodies, the other notions are more liberal and also view software agents and even stones on a field as embodied body (although I have to admit that cognitive scientists do not study the “behaviour” of stones).