vendredi 17 avril 2015

Practice & Context.

Prof. Bourbaki, Inventing frames & diffusing overflows, 2015.
A logic of learning

One type of learning is 'learning by doing' in the sense familiar to economists as a downward shift of an average cost curve as a function of cumulative uninterrupted production (Yelle, 1979). There is also learning in the sense of (radical) innovation; Schumpeterian 'creative destruction' and 'novel combinations'. The challenge is to explain the latter type of innovation and learning. That is subject to uncertainty rather than risk. And the point of uncertainty is that it yields the problem of induction or, more precisely, the problem of 'abduction', as Peirce (1957) called it. Quite apart from justifying a novelty once it is achieved, in the 'context of justification', how do you go about the development of novelty, in the 'context of discovery'? How do you go from ('abduce from') and existing working modus operandi to a new one that in the future will turn out to be better but which now is not known? Of all the new things you could think of doing, how do you choose the correct one, or even one that is viable, and how do you know whether you know all the options (Holland et al., 1989)? The problem of uncertainty is precisely that you do not. This connects with and important issue in Austrian economics: how is it that an entrepreneur can be right s/he venture into new areas? In Herbert Simon's terms: if uncertainty precludes the 'substantive' rationality of choosing the best from available options, we need a 'procedural' rationality or heuristic, in the form of some modus operandi that is likely to succeed.

Of course, one possibility for a process of adjustment is random trial and error. This would make economic evolution as blind as biological evolution. And, indeed, the lack of rational evaluation by especially small firms injects an element of randomness that contributes to variety. But although rationality is bounded and entrepreneurship entails an element of gambling, people do think, make inferences and limit risk, and they are not always wrong. So how, as a firm, could one go about 'abduction', and maximize chances of survival? Usually, the answer in theories of abduction is to proceed via 'adjacent possibilities': to project an existing practice into a context that is sufficiently similar to have a chance of success, and allows for some prediction of likely results, while it is sufficiently different to yield novel experience and indications for further change with a chance of success. That principle is extended below into a few basic principles of a 'logic of development': principles that maximize the chance of survival in an economic selection environment. 

The first requirement for survival is ongoing production during adaptation: without it we starve even on the road to success. The second requirement is adaptation to novel opportunities and threats. How to do both? How to reconcile continuity and change? How to combine exploitation with exploration? How to go from utilization of existing resources to the development of new ones? The core principle here is that one should not surrender an existing way of doing things before both the motive and the opportunity for a replacement are evident. Before the need arises such a move would be wasteful, and before the opportunity arises it would be impossible. Thus a certain amount of conservatism is rational, but it can easily become excessive and block innovation. There is a trade-off between the need to adapt and the costs involved in terms of uncertainty of whether novelty will be successful, and uncertainty about the organizational repercussions (March, 1991). To make the step to novel practice, one must be prepared to 'unlearn' (Hedberg, 1981): no longer taking established procedures for granted. Thus a necessary (but not always sufficient) condition for innovation generally is that there is perceived need, mostly from external pressure, a threat to continued existence or a shortfall of performance below aspiration levels, as has been the dominant in the literature on organizational learning (see the survey by Cohen and Sproull, 1996).

The following heuristic principles of development and learning are now derived. The first principle of abduction is that one needs generalisation of a successful practice to novel but 'adjacent' contexts, where it is likely to succeed (so that it satisfies the requirements of ongoing production), while it is also likely to run into its limitations, so that we may discover the boundaries of its validity (so that it contributes to the requirement of exploration). Next, as the practice runs into its limitations, it should be adapted to the local context to solve them. This is the principle of differentiation. Again, attempts to adapt contribute to ongoing production as a condition for survival as well as adaptation. Typically, such adaptations are inspired by comparisons with similar, 'adjacent' practices, which in the given context are in some respects more successful, and elements from these more successful practices are imported into the local context at hand. This exchange of elements from different parallel practices, in a given context, is the principle of reciprocation. In language, reciprocation is the operation of metaphor or analogy. As the practice becomes more and more differentiated across contexts, efficiency losses appear owing to lack of standardization, economies of scale foregone, complexity of ad hoc add-ons. Novel elements inserted from outside often do not fit well in the structure of current practice, and for the full utilization of their potential require a more fundamental restructuring of practices in a novel practice. This is the principle of novel combinations or accommodation. This is where the previous preparatory steps lead to novelty. But then, at first, the novelty is indeterminate. Knowledge is partly or wholly tacit: to the extent that novelty works, it cannot yet be fully explained. Much experimentation is needed to find its best form(s), for the novelty to 'come into its own', and to become standardized in a 'dominant design' (Abernathy and Utterback, 1978). This is the principle of consolidation. After that has been achieved, one can move into the next cycle of development, starting with generalization. What feeds exploration while maintaining exploitation is an alternation of variety of practice and variety of context. The resulting cycle of learning is illustrated in Figure 1. The claim is that this procedure is the best answer to the problem of how to maintain continuity (exploitation) while preparing for change (exploration).

The hypothesis now is that these principles of abduction, being conducive to survival, constitute a fundamental 'logic' or heuristic of development which is applicable, with appropriate elaborations and enrichments, at all levels of learning/development/adaptation, and at the level of individual people, organizations, industries and national economies. Thus it should, among other things, serve to specify the relation between equilibrating (Walrasian, Austrian) entrepreneurship and dis-equilibrating Schumpeterian entrepreneurship. Furthermore, this 'logic' should also help to indicate how in processes of development the different levels of people, firms, industries and national or regional economic systems tie into each other. Of course, the hypothesis has to be argued more in detail and then tested extensively. The hypothesis has been inspired by the work of the developmental psychologist Jean Piaget (1970, 1974; Flavell, 1967) on individual cognition. 

Note that, while in a procedural sense the heuristic is optimal, it need not yield a unique or optimal outcome. It allows for path-dependence and suboptimal outcomes, and the path taken depends on context and coincidence. Different economies can develop different structures. 'Logic' is put between quotation marks because it is a heuristic rather than a logic in the sense of indicating a sequence of stages that is logically or epistemologically necessary. It is a heuristic in the sense that it is generally the best answer to the problem of abduction; the best way of exploring while maintaining exploitation. However, stages will overlap: there is generalization during consolidation, differentiation during generalization, exploration of novel combinations during reciprocation. Stages may occasionally even be skipped, and innovation can occur less systematically, more randomly and spontaneously (Cook and Yanow, 1993), when an obvious opportunity presents itself without much exploration. But as a general rule one needs to accumulate failures to build up the need for change, as well as hints at what directions to look in: indications of what changes could be made with some chance of success.

Bart Nooteboom, Innovation, Learning and Industrial Organization, Camb. J. Econ. 23, 1999, pp.131-133.

Aucun commentaire:

Enregistrer un commentaire