1) his basic thesis There is quick & dirty biological cognition: action orientated, local problem-solving, 'on line' stuff. This is what we do a lot of the time, without thinking , automatically, and its what animals do and robots...
And then there is advanced cognition: the clever 'exotic', abstract, open-ended, 'off-line' stuff (where Clark wants to hold on to the notion of representations) which only we do, cos ours are 'scaffolded' brains.
There isn't a difference in kind tween us and animals, there is continuity, an 'extension of the basic framework', not a gap.We're the same but better .We're cleverer cos we deal with a wider world.
Traditionally, it has seemed like we've got a special inbuilt capacity for rarified, cool, detatched, logical,(language-like?) Ratiocination, but it's not so.
What we have is access to Diffused Reason, this is cos we are part of a larger social/cultural machine, cogs in it. We are scaffolded by external social, cultural, institutional structures.
Our brains 'offload onto' and interract with external props (institutions like families, schools, businesses, governments.....cultural artefacts like language (more in the next chapter- its the main thing that makes the difference) , slide rules, filofaxes, laptops,TV...)
'Our brains make the world smart so we can be dumb in peace'.
What Minds- these clever rational inference-making machines are is biological brains plus external scaffolding.
So, There is:
2) The classical theory of substantive rationality. How come it seems to work so well for some things?
Traditional economic theory assumes we have perfect knowledge about the way things are and the options available, weighted ordering of preferences, and unhurried, cool, perfect responses.
Yet the latest theories show its not like that- its rule of thumb,not precise rule following, 'good enough' or satisficing, not optimising.
We know classical economic theory doesn't work well at the very high/general/long-term level or very low/individual/immediate level (eg it can't deal with economy as a whole over long perods, or individual consumers behaviour at a moment...)
Yet the classical model does work (at the mid level) for predicting the behaviour of firms, political parties, 'double auctions'. That's cos it deals here with 'scaffolded choice'. As Satz and Ferejohn, Denzau and North show, it works best where choice is limited.
Classical theory works where it does, cos it succeeeds in modelling the diffused reasoning and behaviour of the larger machine in which the individual is embedded. It posits, reflects and confirms a structurally determined theory of interests - what 'maximising returns' means for the system in question.
It's the constraining structure in which the individual is embedded that does the work, and enables the theory to work. The individual is unconstrained, odd, unpredictable. (Gode and Sunder show it doesnt make a difference whether in an 'auction simulation' the trader is of zero intelligence, a normal human , a rat or a coin-flipping device- what they do is pretty much the same.) The larger machine is constrained, patterned, predictable.
The larger machine is (natch) itself one that has proved evolutionarily successful. That's why it's there.
3) Then he goes back to the 'stigmergy' of chapter 4, to show how the individual fits into the whole, and the sort of individual that needs, makes and uses an external scaffolding.
We can best understand individuals' responses as operating within the constraints of a broader social/cultural/institutional context.
Once one accepts that there is:
Then he does some fast sleight of handing to say- just as individual humans are cogs in the broader structure, so individual neurons are cogs in the brain, and what individual human brains do is support 'a grander kind of reason', ie what scaffolded humans typically do.
4 )inside the machine
In this part Clark goes more into the relation tween individuals and the collective whole, stressing that what counts is the dynamics of structures, the interrelations that go on inside collectives. Systems have their own momentum, different from the sum of their parts.
He looks at 2 bits of recent research by Hutchins, a Cog Scientist. The first is interesting,the second silly. The first looks at how simple neural networks sift data to match with a programmed -in concept (about whether something is a dog or not) ,activate when the criteria are met in a 'good enough' fashion, then inform other networks of their findings, which results in the collective of networks plumping for the easiest option, the lowest common denominator, they can all agree on, and then tending to ignore evidence that contradicts this. This is fascinating. Interestingly, when each network is left to its own devices at first, then there is communication across the collective, it 'gets it right', whereas if they all communicate from the start, they often dont, or its only 'good enough'. (Moral- think for yourself first, don't go with the crowd.)
The second bit of reasearch by Hutchins and Hazlehurst looks at how collectives can select for 'better' cultural structures, better symbols over time, and thus pass on more successful external structures. There is something deeply worrying about the circular thinking in and sheer bloody obviousness of the result of this experiment... However, it is used to show that it is not inner, genetic stuff that counts in making a system that gets things right and predicts accurately (ie not a matter of hard-wiring) but evolution in the symbols used, the external stuff, the 'software' which is I think right. Also it is used to show that cultural traditions are tried and tested, the result of adaptation,which is why they are so useful and 'prop'-like, which I also think right.
It goes like this: A collective of artificial connectionist networks was set up, each using 'cultural artefacts' /symbols representing constant correllations in features in the environment (phases of the moon and states of the tide)that were passed on to other networks over time .Some of the symbols used were 'better'/more accurate and thus more useful in predicting regularities than others, so the networks using them were more 'successful'. In some manner I can't work out over time more and more of the networks copied the 'good/useful' symbolic structures, so that over time there was a gradual accumulation of 'better' external artefacts. (the networks cant have 'chosen', but must have been somehow programmed to have a 'selection bias' that caused them(surprise surprise) to pick out and copy the most 'successful'systems (How could they tell who was being 'successful'?)).
Finally 5)
Clark sums up the chapter and asks how all this successfulness and diffused reasoning can go on if there is no centralised control, no inner agent or self (let alone soul) that runs the show, no homunculus reading off meaning and rational connectedness from its inner states or representations, but only quick and dirty multiple inner processes that basically don't talk to each other much.
To Rodney Brook's 3 'constraints' (natural coherence due to physical laws) designed coherence (the 'built- in' goals the system has) and 'cheap global modulation(??), he adds 'stigmergic self modulation' ie the creation and exploitation of external scaffolding for itself. (nb who is the 'self'?: the individual human, or humankind at a time and place , or across its history...?)
To our basic unaugmented biological brain that deals with iterated, local, action- oriented ,pattern- completing -type responses, we add the external scaffolding which enables Human reason to outstrip basic brain power. We participate in the workings of a large machine which is the embodiment of all achieved knowledge, a Leviathan of diffused reason. It is this machine we are part of that is the primary vehicle of our distinctive cognitive successs.
It's all rather Aristotelian- we are social and rational animals, marked off by having language and culture and doing things by conventions.. Its also rather Wittgenstinian- there is 'agreement in judgements as well as agreement in definitions'. The norms and what we do cannot come apart.
Worries:
1)My main worry with Clark , tho I agree with the idea that Minds are operations within a cultural tradition, is the old one of the problem of teleology. How you define 'good, better, best', how you define 'success' what you pick out as 'selected, adapted' etc all depends on what type of thing you think the entity you are looking at is, what its functions/purposes/aims/goals accordingly are, or seem to be. Hutchins and others know what their networks are cos they set them up to do what they wanted. But we are not necessarily like them, or like robots or like animals. Clark et al assume we must be basically all the same.
2)Clark seems to follow Dennett in adopting the intentional stance and seeing rationality as a heuristic device, enabling the observer to make sense of behaviours that would otherwise appear be just one physical action following another. But there is also Davidson's way of seeing rationality as constitutive of what it is to be human, a matter of living up to an ideal of doing and thinking things for reasons, which is not capturable by scientific methodology (which only deals with causes).
3) Clark, for all his good stuff (getting away from the 1st person, putting the individual in the wider context, stressing culture, public language and society, noting the dynamics of structures 'take on a mind of their own', going on about the embodiment of mind, the way minds essentially interract with others and the environment, and the externalist notion of the 'spreading out of mind over the world') that many non-physicalist philosophers (eg McDowell et al) would share, doesn't leave any elbow room for choice, except as a matter of scanning amongst limited options and being caused to go for one of them or picking one at random. There is only the automatic, not the autonomous (except maybe Derrida- style). It doesn't feel right.
4) A final quibble is the 'levels of description' one: the way he seems to see the relation between: