Monday, February 18, 2013

Where is the mind, and why does its location matter?

According to the late George Miller, cognitive science was born on September 11, 1956 at a symposium organized by MIT’s Special Interest Group on Information Theory. Miller left this interdisciplinary conference excited by his sense that experimental psychologists, linguists, and computer scientists faced similar problems; he believed that their search for solutions to these problems exhibited a deep theoretical unity. 

Where did this unity come from? The pioneers of cognitive science confronted issues rooted in the then-new science of information processing. A post-war invention -- the electronic computer – provided some fresh suggestions for solving these problems. Indeed, these pioneers assumed that cognition is computation; to them computation meant rule-governed symbol manipulation, exactly the type of processing brought to life by digital computers. Researchers who adopt this view are classical cognitive scientists. Classical cognitive scientists found unity because in the 1950s there was only one meaning to the term ‘information processing’. 

By the mid-1980s, the unity in cognitive science that inspired Miller was fragmenting. First, the promise of classical cognitive science had stalled. Classical cognitive science produced many promising, small-scale computer simulations of reasoning and language, but general-purpose machine intelligence still seemed a distant dream. Second, and more importantly, new ideas about information processing arose. Connectionist cognitive scientists championed artificial neural networks. Embodied cognitive scientists endorsed physically embodied agents -- behavior-based robots. In today’s journals, one finds extensive debates about the relative merits of classical, connectionist, and embodied cognitive science. 

Embodied cognitive science is the most recent reaction to the classical approach, and in many respects its ideas are the most revolutionary and dangerous. Embodied cognitive scientists describe classical models in terms of what is known as the classical sandwich, defining cognition in terms of a cycle of sense-think-act processing. That is, classical models do not permit direct connections between sensing and acting. Instead, the environment provides raw information to thinking processes that build an internal model of the world and use this model to plan appropriate action. Action only takes place in a classical model after this extensive modeling and planning and thinking have taken place. In a classical model, the middle of the sandwich -- the thinking -- really is the meat; planning is taken to be the ultimate goal of cognitive processing. 

Embodied cognitive science challenges the importance of planning, and defies the need for the classical sandwich. Rodney Brooks famously asked why we need to model the world when we can (through sensing, or ‘situation’) use the world as its own model. He built behavior-based robots whose sense-act reflexes could quickly react to the sensed world without modeling and reasoning about it. In the embodied revolution the purpose of cognition is to act, not to plan. 

Behavior-based robots do not merely sense their world; being physically embodied they can act upon it and change it. Extending this idea to human cognition, embodied cognitive science has renewed scholarly interest in how humans use the external world to support or scaffold cognition. For instance, an enormous amount of our memory is deliberately recorded outside our brains, in books, in electronic devices, in the World Wide Web. Our memory may not require us to think, by retrieving internally stored information. Instead it may require us to act on the world, storing (and later finding) information externally. 

If cognition is sense-acting, if it is scaffolded by the external world, then it is reasonable to wonder whether the world itself is literally part of cognition or of the mind. Philosophers Andy Clark and David Chalmers introduced the extended mind hypothesis, which was the proposal that the external world was literally part of the mind. According to the extended mind hypothesis the boundary of the mind is not the skull. Not surprisingly, this hypothesis is controversial; its supporters and detractors have published many articles and books to debate the whether the mind is extended. 

The debate about the extended mind hypothesis is important to academic cognitive science. If cognitive science is the scientific study of cognition and the mind, then it is critical to know exactly where the entity of interest resides! However, might the issue of whether the mind is extended or not be only of interest to the scholarly residents of the ivory tower? 

Importantly, it is not. 

For instance, I often see news articles and letters to newspaper editors critical of current educational practices. A frequent theme is the need to return to “common sense education”.  For instance, there are frequent appeals for educators to instill better thinking in students; a “common sense education” should produce students who can use their excellent memories to learn important concepts. What prevents this from occurring? A frequent claim is that technology is the culprit -- because of calculators, students can’t perform mental arithmetic; because of word processors with spellcheckers, students can’t spell; because of smart phones and social media, students are distracted, and so on. 

Such commonplace criticisms of education seem strongly related to classical cognitive science, and emphasize the importance of internal modeling, planning and reasoning. But we have seen that cognitive science is now seriously considering alternative notions of mentality. If embodied cognitive science’s extended mind hypothesis is true, then perhaps we should seriously pursue an educational system that promotes students’ use of technology, and which develops their skills in manipulating the world -- and its technology -- to scaffold their cognition. If cognition is acting, and not thinking or planning, then the educational implications of the extended mind hypothesis are staggering.


  1. I enjoyed this concise introduction to the Hypothesis of Extended Cognition. I was wondering though, when speaking about information being stored in external objects, say, a notebook, and this allowing for scaffolding, doesn't this picture concede the classical sandwich in some sense? It seems like the HEC, at least how Clark presents it, admits that there are symbols being manipulated, they are just external ones. So, my question is, how radical do you see the HEC being? Or is there some continuity with the classical approach? And, why can't we say the "meat" is both in the world and in the head?

  2. Personally I'm comfortable with the "meat" being both in the world and in the head. One hint at this is my blog post on memorizing pi; this trick uses a lot of scaffolding, but it also leans heavily on internal representations.

  3. Could you maybe elaborate on what you think internal representations are? I ask because I'm just thinking that if internal representation are distributed like as is generally assumed in connectionism, and they are not symbolic like in classical theories, then it's not clear to me how they are supposed to be sufficiently similar enough to external "meat" to figure into cognition? This is the reason I brought up the classical sandwich picture, because if internal representations are symbolic then this gels rather well with have external content be included in cognition, so long as it is included in the right sort of functional way. But if internal representations are distributed then doesn't this raise a concern for the extended story we tell about how the external "meat" or content figures into cognition? Hopefully that makes sense.

  4. I believe that the (potentially) distributed nature of representations is a separate issue to the extended mind. If one looks at the literature (e.g. van Gelder) one finds that 'distributed' is a very difficult term to define, and that it doesn't serve as a 'mark of the classical'. I don't see why the mind couldn't be both extended and distributed, but that is just speculation on my part.