All 9 entries tagged Extended Cognition
No other Warwick Blogs use the tag Extended Cognition on entries | View entries tagged Extended Cognition at Technorati | View all 1 images tagged Extended Cognition
November 15, 2005
Section 10.6 of Andy Clark's book Being There is entitled with the question "Where does the mind stop and the world begin?". For philosophy this is a very significant question. For cognitive science and AI, much less so (its just a design issue). Why not just adopt the latter position? Would that be such a scandal?
Clark's answer to the question is both pragmatic and realistic, whilst promoting a proportionate, specific and sufficiently detailled investigation of real minds and environments. This is quite a contrast to the vague generalizations of some phenomenological models.
For someone with an AI/cog-sci background (that I in part share), the identification of a boundary (even a porous one) should only be significant when it could contribute to our understanding of the capabilities, limitations and developmental process of real cognitive processes. Our boundary marking conditions would have to be ones that really make a difference to the cognitive process itself. For example, one interesting boundary marking condition would be:
how replaceable or otherwise is a specific (internal or external) cognitive artefact? Could the individual agent simply swap the artefact with another similar or even totally different artefact? And to what extent would this change the character of the agent?
A related, equally important, but different question is:
how dependent is the development of an agent upon a specific artefact, such that it's abscence makes a significant difference to that agent?
This gets close to our understanding of what an agent actually is: it has a relatively consistent and pervasive character existing over time and to some extent surviving changes to the environment in which it exists. Whilst at the same time, its development and continuation is dependent upon the existence of key artefacts within that environment. It is as Clark says, closely coupled. Furthermore, the agent tends to influence the environment in which it exists so as to promote the continuation of these characteristics, so that an agent tends to be associated with an environment (reverse evolution), whilst the environment tends to promote certain characteristics in the agent and classes of agents (evolution).
This, to readers of recent dynamical systems theory (and the likes of Deleuze and Guattari), is quite an obvious model: 1) there are arangements of mechanisms that interact with and consume other mechanisms through processes of ordering, selection, managed preservation and controlled degradation; 2) these mechanisms have selective principles (the character traits) that are repetetively applied over time; 3) some of these repetitive mechanisms reproduce the conditions of their own production and reproduction; 4) and fewer still reproduce the conditions that make their own reproduction more likely, more desired by the environment in which they exist. Or in shore: they are desiring machines.
I would say that this is stating the obvious. Certainly there is a degree of convergence towards such a model in evolutionary biology. And I'm sure there will also be such a convergence in AI development. So why is it likely that philosophers will still consider it to be controversial? Why does it seem OK in biology, but radical when applied by, for example, the psychotherapist Felix Guattari, to the problem of fixing broken minds and bodies?
Thinking is selecting, is doing.
November 06, 2005
Page 215 sees both an indication of just how radical its implications might be, followed by a clause that is perhaps an attempt to avoid an engagement with a whole set of possibly esoteric philosophical issues.
Firstly, there are significant ethical implications of an argument that sees a persons mind being extended into the environment. Hurt that environment and you hurt the mind. This goes a little beyond the arguments of human rights lawyers, who could safely say that long term deprivation may affect the development and sustanance of the mind. Clark's argument clearly indicates that damaging the extended cognitive apparatus has an immediate and damaging effect on their mind.
If, for example, a human rights court assessing the actions of the regime at Guantanamo Bay were to accept that abuse of the Quran were a direct physical attack upon the minds of the inmates, then the range of crimes would be greater and more extreme.
The first point opens up a radical debate. The second is just as dramatic in closing one down. Clark makes a threefold differentiation:
He also talks about "self", and seems to be referring to the totalized singularity of the collection of factors that make a person individuated. That, I think, is closely tied to what he wold call "consciousness", but the issue is not properly explored.
His argument clearly shows that minds are more than brains, being extended out from the brain into the environment (or perhaps coming in from the environment and parasitising the brain). But he is careful to say that there is something called consciousness that is not extended into the environment. Individual consciousness, my attentional experience, is packaged back inside the individual. It is philosophically safer to say that there is still something, some inelliminable feature of being human, that is not dissipated out into the apparatus of extended cognition.
The clause avoids some very difficult philsoophical ground, but only at the risk of begging a very big question. This clause has the following effects:
- a separate set of apparatus must be implicated in consciousness;
- this apparatus cannot be reducible to, dependent upon, and part of the environment, as being such would again make it porous and subject to an extended cognition argument.
Obviously the second of these points is hugely controversial, and heading towards the kind of mind/body separation that Clark set out to dispel. But I would say that it is essential to Clark's attempt to keep some kind of separation between subject and object (individual and world). Without some kind of absolutely non-porous subjectivity, his thesis gets increasingly radical. The supposition of an individual consciousness, for example, provides some limitation to the damage that extended cognition could do to our established legal and ethical assumptions.
How to escape from this? Phenomenology leading out from Kant has recourse to time. In fact once can see the predominately spatial way in which much of the extended cognition debate is framed. We have a bounded, territorialized layout of minds and environments. One could argue that consciousness is the experience of this layout in time. The link between "self" and "consciousness" seems to rely upon this, with time, history, evolution and its experience from a specific perspective being the individuating feature:
…the flow of reason and thoughts, and the temporal evolution of ideas and attitudes, are determined and explained by the intimate, complex, continued interplay of brain, body and world. p.217
We could investigate how the environment/mind relationship unfolds differently to an individual consciousness. A kind of "pure time", a "duration", could be the inelliminable fact of consciousness. But what are the origins and effects in the world of this pure time? It can't be an organizing super subject, because that would be subject and hence porous to the world it organizes. As you can see from the passage given above, Clark does actually acknowledge that the experience of being a singualrity in time is the result of a complex dynamical coupling.
There's another possibility. Rather than being an organizer, a Kantian transcendental rationale, perhaps it is exactly the opposite, a hesitation at the core of conscsiousness, the force that holds the folds of time open and inelliminable. A chaotic attractor at the core of being in the world?
My argument is that the "chaotic attractor" of consciousness, and its temporal incarnations, is in fact crucial to perception and cognition. It is the drive behind inquisitiveness and the dynamical engagement of minds and environment. Far from being outside of cognitive science, it will prove to be the key.
But perhaps to obtain this key we have to accept that our ethical and legal assumptions need to be re-thought?
If you are interested in this entry, then please contact me by email.
October 22, 2005
As I descibe in a previous entry , Clark uses the lifecyle of a mangrove island as a metaphor for how we sometimes use public language to speculatively play with ideas and see how they grow. I summarise metaphor as follows:
a mangrove seeds itself in shallow water, grows roots, traps other roots and particles, forms a network of roots with other mangroves that seed nearby (helped by the first mangrove), and eventually forms a more solid island within the sea.
Clark's argument is that by stating words publicly using an external cognitive apparatus (a web log would be a good example), even when they do not represent well formed ideas, thoughts and sense can be encouraged to form. We can easily play with the publicy stated words, manipulating them and seeing what emerges as they connect with other words. As a social process, involving other minds, this technique can lead in unexpected and successful directions.
It seems to me that "mangrove" style social thinking could be a core activity of the localized cliques that are probably the context for most blog entries. The clique tends to select phrases to be repeated locally, and provides opportunities for social thinking in a controlled space. But the small world network of Warwick Blogs offers two other possibilities:
- ideas that become well-formed can be transmitted to other localized networks through 'hubs' (people or pages that are authoritative in presenting selections of worthwhile content);
- ideas can also just free-float from one clique to another, for example via the "show all" page, where they may take root, or at least make the transmission through a hub more likely (hearing a message through several channels makes it more likely to be trusted and valued).
I have commented previously on how cognitive science has been dominated, and possibly limited, by a conception of the nature of cognitive tasks that reduced them to being classical problem solving activities. This 'problem solving' ranges from trivial pattern matching to complex strategy formation. In every case, the problem is well stated, along with the conditions that demonstrate the achievement of a solution. Cognitive science has of course existed to a great extent to support development in artificial intelligence, which has itself been skewed towards industrial and military applications (see De Landa's War in the Age of Intelligent Machines for an account of this relationship).
Coming as I do from a very different philosophical background, one that is more interested in artistic and literary creativity, it would be easy for me to dismiss cognitive science for its obsession with such trivia (even considering that the actual problem solving isn't that simple). However, Clark's extended cognition thesis fits neatly with accounts of artistic and literary creativity given by the likes of Deleuze and Guattari. Indeed I plan to explore these links in much more depth. I can see occassions in which it finds a line of escape from the 'cognitive task as problem solving' trap.
In chapter 10 of Being There, Clark describes a recognizable form of 'extended cognition', which he calls the "mangrove effect". The metaphor is this: a mangrove seeds itself in a shallow water, grows roots, traps other roots and particles, forms a network of roots with other mangroves that seed nearby (helped by the first mangrove), and eventually forms a more solid island within the sea. Clark argues that in some cases linguistic elements (both publicly spoken and internally contained) can work in this way. A word is uttered, not to fill a definite space or necessarily solve a well–defined problem, but rather to probe the cognitive and social environment, to see what connections form around it, or even to change the cognitive and social environment. This sounds much closer to the activities of musical and painterly composition described by Deleuze.
That is what Deleuze and Guattari would call "rhizomatic".
Most importantly, we should consider how this model of cognitive task frees us from a theoretical dependency upon a well integrated goal oriented super–subject. During the discussion following Clark's paper, the practical question of how to differentiate cognitive apparatus properly belonging to an agent from those belonging to the environment. I think he responded with an answer that relied upon the existence of such a super–subject (and its plans and goals). My alternative argument (which I think comes from Deleuze) would be that it isn't so much a well–integrated well organised super–subject that provides the drive for the cognitive task, but rather a dissonance producing chaotic attractor, speculatively dispersing fragments of sense into the world in order to simple make things happen. As Deleuze and Guattari say "the machine only works when it breaks down" (spot the double meaning).
So in fact, I argue, the cognitive tasks that drive the "mangrove effect" are closer to artistic creativity – composition, especially poetry. But this is not alien to Clark's thesis. In fact on page 208 he brilliantly identifies poetic composition as a form of thinking that exploits the "mangrove effect".
This leads onto Clark's more recent thoughts on other way in which thought is dependent upon an extended apparatus in a non–trivial way (that is to say, more than just as a means of cognitive off–loading, as in the case of simple note taking and note reading). Gestures and other rhythmic, haptic techniques were discussed. Someone asked if an exercise machine could ever form part of the extended cognitive apparatus (a musician would certainly say yes). Clark did talk briefly about non–linear couplings between mental and extended apparatus. A consideration of rhythmic apparatus could be drawn from there, leading into time and complexity.
At this point I remembered Cezanne's description of how his hands and the paintbrush and the canvas would merge together in the act of painting – what Deleuze called the "diagram" (see my entry on Cezanne Unlocking Sensation ). The act of painting for Cezanne, this merge between mental and external apparatus, rhymthically moving together, is a "mangrove effect". Cezanne:
Our art must shock nature into permanence, together with all the components and manifestations of change. Art must make nature eternal in our imagination.
October 20, 2005
After spending a few weeks away from my research project, I usually restart my thinking by quickly reading a popular-science/history/art-theory book of some kind. My means of choosing such a book is usually quite aleatoric: the criteria being "something that sounds good, is easy and fast to read, and which may provide some unforeseen empirical data for my conceptual activities".
Mark Buchanan's book, selected from the popular-science section of Waterstone's in an instant, without much consideration, met these criteria well. The aim of the book is to convey and contextualize what is a fairly simple idea. But what makes that simplicity into something more significant is that it is an idea that seems to have been overlooked, and which once brought into consideration, gives explanatory sense to lots of seemingly unrelated phenomena. For me as a philosopher, that is interesting: what thoughts were impossible before the arrival of this concept? what errors were made?
The concept is, superficially, that of the "small world network": that is to say, how networks seem to naturally form from highly chaotic and random situations, into simplified but still random organisations, connected together by specially priveliged nodes that do much of the work of maintaining order and flow in the system. As a result of this priveliged position, and the dynamics of its connectivity, all kinds of sophisticated behaviours (economic, social, cognitive, ecological) can be seen to emerge: results of what has been called the network effect (although there is much more to this than the business model). It plausibly demonstrates how it is quite feasible for one random person to be only six degrees of separation from another random person. And then it extends this model to many further domains (physical, ecological, computational etc).
I'm thinking: Kant, sensus communis – what if he new about small world networks? Or conversely: Nietzsche – what did he know? And of course it is there in Deleuze and Guattari (concepts such as transversality), but rarely with the very definite examples that we now know of. A fascinating question to consider would be: when exactly did the concept of "network" arise? – and at what point did people start realising that the conditions for the establishment and operation of a network may cause certain behaviours, patterns and organisations to emerge (the network effect)?
Connect it to Andy Clark's extended cognition theses (with its tightly coupled systems). And throw in our experiences with the small world system known as Warwick Blogs. Interesting. Very.
Even more so because it turned out that, as I discovered on page 16 as the author described the kind of surprising coincidence that a small-world network can cause, I am in fact only two degrees of separation from Mark Buchanan! He is a friend of a friend As I read…
I moved a few years ago from the United States to London to take up an editorial position with Nature
…I dropped the book when I realised that he may well share an office with my friend Karl, who is also an editor with Nature, also working in the physical sciences. Karl is a Moosehead, with whom I occassionaly drink, eat chillis, and bellow loudly. On Friday night (whilst in the Bilash in Botley) I explained this to Travis (also a Moosehead, if not the head Moosehead). Travis was actually planning, that night, to write an email to Mark Buchanan. It may even be that I have met the author at some Moosing event at some time (although I may have been drunk and therefore incapable of remembering the names of any new acquintance). Spooky. Or just the result of a small world network.
Buchanan's book does well to rapidly explain the work of Watts and Strogatz, Granovetter, and other pioneers. It is entertaining and full of fascinating examples throughout – especially when dealing with the ingenious experiments of the sociologist Stanley Milgram. The connecting-up of cases from such a wide range of domains begs many questions (important philosophical questions that I think Deleuze and Guattari address effectively in What is Philosophy?). But it is very much worthwhile because of that.
I'll give the book 4 stars (not 5, as it could do with a bit more detail on the mathematics and mechanics of the networks).
If you have feedback on this entry, please contact me
August 25, 2005
Can Clark's open-minded and pragmatic cognitive science be applied beyond the scope of problem-solving behaviour to which it is aimed?
By Chapter 9 of Being There, Clark has etablished good arguments for a more open-minded and pragmatic cognitive science, combining dynamical, computational, componential, symbol-processing and representational models so as to match the richness that we should expect of real minds. There is also the start of a set of heuristics to guide us in deciding which approach (or blend of approaches) to use in a given situation. The discussions of 'continuous reciprocol causation' and 'representation-hungry problems' provide some clues as to what indicators to seek in choosing methodologies for specific situtations.
Chapter 9, Minds and Markets, opens this up further, arguing for the importance of seeing minds as embedded in and formed from a set of other 'external' open systems. These real-world cognitive tool kits provide the material required for the various tricks and techniques that allow a brain to extend its intelligence and power significantly:
The idea, in short, is that advanced cognition depends crucially on our abilities to dissipate reasoning: to diffuse achieved knowledge and practical wisdom through complex social structures, and to reduce the loads on individual brains by locating those brains in complex webs of linguistic, social, political, and institutional constraints.
This may in fact sound rather obvious. Research in psychology, linguistics, sociology is concerned with just this. However, we could argue that there is a tendency to 'protectionism' that guards certain cognitive activities from this analysis. The discourse on 'creativity', for example, even now is surrounded by notions of genius. Studies of creativity as an embedded, machinic process (such as Deleuze's study of Bacon) are rare. And yet such studies have proven to be revealing and powerful.
My argument is that the 'higher' we can set the aims of the 'higher reasoning' addressed by Clarke, the better. Apply his methods without limit.
But at the same time, be critical and wary of his bias towards reasoning as pattern-matching, goal achieving, problem solving. Note the emphasis on constraints, and the lack of consideration of 'wild' less-directed revolutionary creative activities, in which arrangements of 'continuous recipricol causation' emerge rapidly, redefing the problem-space faster than the selection of problem solving apparatus.
August 21, 2005
Despite some interruptions, I'm now up to chapter 8 of Andy Clark's Being There. The chapter on The Neuroscientific Image presented a plausible model of cognition based upon brain research, but combining functional-hierarchical, distributed, embedded and dynamical approaches. But I am still not satisfied that the question of methodology is addressed sufficiently.
The important point is that dynamical and computational explanations are not exclusive, but rather, describe actually distinct forms of organisation and mechanism, and hence our task is to identify when one or the other is more appropriate (and indeed the combinations of the two).
That is good, but rather than giving a systematic guide as to how we can apply these various approaches, we instead see a cognitive 'science' built upon a patchwork of interdependent conjectures concerning the many distinct aspects of embedded and evolutionary cognition. The conjectures add up to an 'engineering model' of the intelligent organism, with the aim of offering a plausible story. They concern, amongst many other aspects:
- Components available for the construction of the system, and the energetic and physiological limitations imposed on those components (both in the brain and the rest of the body);
- The ontogenetic generation and regeneration of those components within the life of the individual, in isolation and through complex co-evolutionary relationships;
- The diversity of the modes of organisation of those components, and their interactions (including either computational and dynamical operations in combination);
- The generation and regeneration of those organisations;
- Components in the environment available for simplifying and extending cognitive processes;
- The feedback and feedforwards loops between these internal processes and the external environment within which the organism is embedded;
- The limitations and requirements (temporal, spatial) imposed upon the individual by the environment;
- Co-evolutionary links with other complex entities in the environment of the organism;
- The phylogenetic development of all of the above in the evolution of the species and its environment.
But how can these individual conjectures be 'falsified'? Much of the work of cognitive science is to assess the fit of one of these conjectures with the many others. Of course that may lead to complex but consistent theories that seem plausible but which turn out to be entirely wrong. What other means do we have for assessing the plausibility of the conjectures. Clarke relies on two:
- Selective neural damage evidence – the mainstay of neuroscience, showing how damage to a part of the brain has specific effects on behaviour;
- Economic plausibility – assessing whether a conjecture describes mechanisms that are just too extravagant and costly to be likely (for the individual, species or environment).
Method 1 is relaible where a theory depends upon the existence of a localized centre of mental functionality or control that is suspeptible to damage. However, even in these cases there may be important distributed and dynamical elements that are difficult to assess (and hence get ignored by the theory).
Clarke gives some relatively trivial examples of method 2. However, I am not convinced that such 'economies' of resource (including time) are often simple or stable enough for easy analysis. And furthermore, if we consider the organism not to be an individual, but rather to be an assemblage of highly mobile components (especially at the 'higher' levels, at which an individual may become possessed by powerful, nomadic memes driving its behaviour), economies become much less predictable and reducible.
More fundamentally, i'm not at all convinced that there is some 'rule of efficiency' driving and limiting all phenomona. That I think is the really interesting question.
If you are interested in this entry, please contact me
August 08, 2005
By Chapter 5 of Being There, Clark has reviewed the relevant work in robotics, cognitive science and developmental psychology. The slightly understated conclusion seems to be that a vital ingredient is missing: reality, messy, complex, non-linear reality.
This is a key paragraph:
This approach ignores one of the factors that most strongly differentiate real evolutionary adaption from other forms of learning: the ability to coevolve problems and solutions. p.93
The question is, as I read on, will he propose some kind of mechanism for introducing this factor into simulations, and thus quantifying its likely effects and patterns? Or perhaps he will explore the non-linearity of co-evolution further, with the conclusion that it renders a science of embedded cognition to be of limited explanative power?
My bet would be on coming up with toolkit for identifying the situations in which co-evolution occurs within lmits, as distinct from cases when its non-linearity renders problems obsolete faster than the emergence of solutions: a set of systems co-involuting through a shared Body without Organs, with degrees of stability and relative velocities.
If you are interested in this entry, please contact me
August 02, 2005
I am now reading Andy Clark's book Being There: Putting Brain, Body, and World Together Again. This is a successful attempt to bring together strands in situated robotics (which I studied with Dave Cliff at Sussex), cognitive science, evolutionary and adaptive systems (from von Uexkull, also Deleuze and Guattari's inspiration), developmental psychology, and phenomenology.
The book contains many good observations and conjectures derived from all of these fields. But the guiding principle is that nature tends towards the simplest and most efficient solution. This is, in many cases, a devolved 'subsumption architecture' of the kind employed by the roboticist Rodney Brookes. In some cases however, a more complex solution emerges. An example of this may be simulations that act to provide "virtual feedback" within action loops:
…proprioceptive signals must travel back from bodily peripheries to the brain, and this takes time – too much time, in fact, for the signals to be used to generate very smooth reaching movements. To solve the problem, the brain may use a trick (widely used in industrial control systems) called motor emulation. p.23
If you are interested in this entry, please contact me