I accidently found this somewhere on my computer, so why not put it in the open:
Why Dynamical Systems Theory is relevant to engineering
Commentary on “Being There”, Andy Clark (1997)
Jelle van Dijk
Utrecht University of Applied Sciences
18 march 2008
There are various ways in which cognitive science is different from its precursor, cybernetics. I will try to pinpoint one of these ways. The starting point is a remark made by Andy Clark in his groundbraking work ‘Being There’ (Clark, 1997). Clark discusses how the best explanations of cognition might not be the, then standard, computational-representational accounts, rooted in the philosophical position of cognitivism. One of the emerging trends in the nineties was Dynamical Systems Theory. At present, it has turned out to remain rather a niche in the cognitive science community, despite original interest awakened by works such as (Thelen and Smith, 1994) and (Kelso, 1994). Perhaps one of the reasons that DST is not so popular today is because it was dismissed rather explicitly by Clark, who remained much more sympathetic to an informational-processing account. He allows for DST to be an interesting complement, but refuses to go ‘all the way’ . In discussing the drawbacks of DST he states:
“The pure Dynamical Systems theorist is seeking mathematical or geometrical models that give a powerful purchase on observable phenomena. This is good science, and it is explanatory science (not mere description). .. But [its] power comes at a cost: these ‘pure’ models do not speak directly to the interests of the engineer. The engineer wants to know how to build systems that would exhibit mind-like properties, and, in particular, how the overall dynamics so nicely displayed by the pure accounts actually arise as a result of the microdynamics of various components and subsystems” (Clark, 1997, p. 120).
But these are very specific kinds of engineers Clark is referring to. They are, to be precise, the people working in artificial intelligence (Simon, 1996). Other engineers, including many of which have, in fact, been brought up within the academic cultures of cognitive science, computer science and/or psychology, are not interested in artificial intelligence at all, at least not primarily. They might be in need of artificial intelligence (and for them, a lot of systems quickly go as ‘intelligent’ as seen from the practical perspective), but only if it suits their needs. If not, then they would just as quickly turn to other means. So what ends are these means for, then, if their goal is not primarily to create synthetic ‘minds’? What are the ends of these non-AI engineers? Here, we have to turn back to the old roots from which AI has branched, the fundamental engineering philosophy, cybernetics. In cybernetics control is the key-word. I would hold that most engineers are interested not in building systems that mimick, or model, some kind of behavior, rather, the engineering discipline is primarily a practice of engineering control (over environments, including people). Thus, the typical engineer is not primarily interested in building a thermostat for a living room that ‘models’ its environment, like as if the goal was to build a system that was ‘just like the living room’. Rather, the thermostat should be able to control the environment, in certain desired ways, e.g. holding the temperature within a fixed bandwidth (the necessary normative aspect of engineering). If it can do so better by modeling the world, this would be a possible solution. If it can do so using a purely analog, mechanical device, this will provide another solution. Engineers will sort out what means to use, depending on various factors present.
The question is whether DST affords control. I believe it does, because if the dynamical systems scientist has extracted a relevant control parameter within the system, the system has thereby been ‘opened up’ for control from the outside. The engineer is thus presented with a set of dials (the control parameters), for which he can seek desired settings so as to control the system in such a way that certain goals are reached, or rather most probably, certain homeostatic states are being maintained.
Now regarding the understanding of human intelligence (the science of cognition), this indeed comes at a price. The price is that we let go of the naive goal of actually building an artificial intelligent system. This kind of engineering has proven to be an utopia. It has not worked. And it will not work, precisely for the same reasons that scientists have turned to studying complex nonlinear interactions instead of modularized simple computational components. At least, this is my strong hunch. Instead, the principle engineering field that cognitive science should connect to, then, is human-system interaction. This is an area where Engineers build the systems. These systems then interact with human beings. And if the system is able to provide appropriate levels of ‘control’, this means that the engineer has understood something about the system that it is trying to control. In fact, this has been a practice for years, and cybernetics and systems-theories have naturally influenced and blended into practical engineering areas such as educational design (Bateson) and computer-supported-cooperative-work (Winograd & Flores; CSCW), and so on.
In all, Clark is, I am inclined to believe, wrong in thinking that DST can do nothing for engineering in general. It does do little for any type of AI that is aims for a full-blown engineering of artificial autonomous intelligent systems. But cognitive engineering in general can greatly benefit from DST, both as a conceptual position and from it’s empirical analyses. Here I see a cognitive engineering that aims for improving human-machine interaction, asking the question of how technological environments can provide constraining structure (control), such that certain desired states will be reached and maintained. (NB: fill in the kinds of ‘desired states’ you feel most comfortable with, e.g.: better education, less frustration in daily life, succesful farming in third-world countries, world-peace: note that ‘control’ in the way I use it has nothing to do with oppression or specification or determination, although I grant the negative connotation of the word in our culture) If anything, the fact that DST can do nothing for artificial intelligence might signal the death of artificial intelligence, not the irrelevance of DST to engineering.
References:
Clark (1997) Being There
Kelso (1994) Dynamic Patterns
Thelen & Smith (1994) Dynamic Systems Approach to Cognitive Development
Bateson
Simon (1969/1996) The sciences of the artificial
Winograd & Flores
Varela Thompson & Rosch
Popularity: 17% [?]
