what is tacit knowledge?

 ch7. What is tacit knowledge?

In this chapter, H.M. Collins takes on the problem of whether tacit knowledge can be absorbed, so to speak, by neural nets. 

Interestingly, this book (The Practice Turn) was published in 2000, so 21 years ago...good to keep in mind. 

Collins tests the proposition that computers can't replicate practices by looking at three different theoretical models. 

1. motor skills metaphor
2. rules regress model
3. form of life approach

He returns to each section. The first time, I believe, is just to give us a background on the approach, whereas the second time is to test is with the idea of neural nets. 

Collins concludes by arguing pace Dreyfus that neural nets can absorb practice, but they can't absorb what Collins calls "social embedding."

I understand social embedding to be the traditions out of which practices emerge, but that can't be right. at the end of the day, social embedding makes me think of neural nets and training data. You can train a computer to see whatever you want, I'm sure, but what the computer sees depends on whatever training data you give it, and this training data is heavily depended on human biases, biases that themselves come from society, traditions, culture, etc.

Collins doesn't define social embedding per se, but he does say this about the forms of life approach (which is third in the list and therefore probably the most important):

10. The inspiration for my own initial research was the ‘form of life’ approach and only because a referee drew my attention to Polanyi’s tacit knowledge idea did I use that formulation in my 1974 paper. In some ways I have been regretting it ever since because, as I will go on to explain, the form of life approach is the most interesting and fundamental. (127n11) 

This is from the social embedding section:

What I have done so far is to consider three ways of approaching the idea of tacit knowledge. In the cases of the first two routes I have taken examples that have allowed me to be prepared to concede that what we call tacit knowledge may have a lot to do with the calculative processing limitations of the human brain and that there is nothing philosophically fundamental about the tacitness. That is to say, I have not found it difficult to imagine that the tacit knowledge involved

But, even if it is the case that all tacit knowledge as understood via the motor-skills metaphor or the rules-regress model could, in principle, be replaced with something like the programs of artificial neural nets, the formation of these rules, which we see as we watch the formation of new scientific consensuses, cannot be so replaced. This is because the history of scientific consensus formation is the history of groups of scientists, and scientists

imperfections out of science.

WHAT IS TACIT KNOWLEDGE? 123

124 H.M.COLLINS

might be transferred to a subsymbolic computer program and thence, perhaps, to a symbolic computer program. I have argued, however, that even if this is the case, the idea of tacit knowledge is vital for the analysis of science as we know it, and has permanently altered the way we analyze science in so far as anything ever could.

I have then gone on to look at a third way of approaching tacit knowledge and suggested that this creates a new kind of problem—the creation of new consensual knowledge. Even if we allow that all consensual tacit knowledge is potentially transferable to artificial neural nets, the neural nets that would be needed to have taken part in the creation of science as we know it would have had to fit into society in the same way as the human scientists fitted, and we do not know how this could have been managed. 

This separation and amplification resonates with the prior footnote. 

The improvisation thing is also in the social embedding seciton:

The same two aspects can be found in dancing. Mr Data may have learned some steps from Beverley Crusher but putting these into practice on the dance floor is another matter altogether. In the Star Trek episode in question, Commander Crusher, impressed by the instantaneous nature of Data’s mastery of the steps she has demonstrated, remarks: ‘Now all you have to do is improvise.’ It is here that Star Trek departs from its customary scientific conservatism, for ‘improvisation’ is exactly what we would not expect Mr Data to be able to manage. Improvisation requires exact attention to, and a very deep understanding

physics of bike-balancing is consensual and frozen. )

The other aspect of bike-riding is riding in traffic. This is much more

of, human society because it is not a matter of following a behavioral repertoire but of making departures from it to a degree that must be consistent with quintessentially unspoken convention. There is little easier to caricature than the gauche dancer desperately trying to impress with inappropriate improvisations, but there is little harder to explain. 

When we move to the aspects of human action that are socially embedded it is

far harder to imagine them being inscribed in a computer model, whether the

computer be symbolic or subsymbolic. These kinds of actions require that

behavior be varied, not randomly, but according to the social circumstances.

Such actions are what Kusch and I call ‘polimorphic actions,’ as opposed to the

rarer kind of ‘mimeomorphic actions’ which humans prefer to instantiate with 

the same behaviors every time 

So this is basically getting us to atypical rhetorical action (Christopher Basgier).

Dreyfus is all over the footnotes, btw. 

Collins is def not articulating something like ANT. 

2. It also follows that this paper is an implicit critique of analyses of knowledge that reduce society to something like interactions in a network. This paper defends a notion of society that is very much more than the sum of the interactions of its parts. Certain of the capacities of the members of a society are seen as available to them only because they are members of something larger and irreducible (Collins 1998). In this the paper draws on the philosophy of Wittgenstein (1953; Winch 1958). For related arguments see Schatzki 1996, 1997. (126n4) 

"idle wheels"

It follows that I am not terribly interested in the analysis of the term ‘tacit knowledge’ as it is used in theories of practice. Many contributors to this volume consider that terms such as ‘tacit knowledge,’ or ‘tacit rules’ are idle wheels in theories, or worse. Some think that to use such terms indicates, misleadingly, that there are hidden structures which underlie practices, whereas all we need to refer to are the practices themselves. I am not trying to make a direct contribution to that debate, at least, not in the terms in which it is usually cast;3 my contribution is meant to be that of a practicing sociologist, for whom my own (partial) mastering the practices of the communities I study is a very concrete achievement akin to what the members of those communities themselves achieve as they become members. A language describing what I and my respondents do as we move from a state of incompetence to a state of competence seems anything but an idle wheel. Thus, I note that mastery of a practice cannot be gained from books or other inanimate sources, but can sometimes, though not always, be gained by prolonged social interaction with members of the culture that embeds the practice. 

I wonder if proprioception could help me out with Bloor, with the idea that you can think something thinkingly and blind at the same time, with the difference between mere conformity and genuine involvement. 

Bike-riding is complicated so let us pick a simpler example for some deeper analysis. Let us consider dancing and its relationship to what psychologists call ‘proprioception’—awareness of the position of our own limbs and muscles. Mr Data, a friendly computer in human form portrayed in the TV series StarTrek: The Next Generation, has more proprioception than most mortals. In one episode, Mr Data is being taught to dance.