Models of this general kind, with complicated mechanisms being built up from simpler components having a simpler function, have been developed previously by workers such as Marr (models of vision) [4] and Brooks (robotics) [3], and these indicate the meaningfulness and practicality of such an approach. What is proposed here differs crucially from the work of Marr and Brooks, however, in the self-organisation aspects of the mechanisms concerned. The mechanisms, by examining their own performance, are able to diagnose imperfections in the functioning and make changes that improve on them. We do not have to start off with skilful machines; it is sufficient to have primitive mechanisms that crudely approximate to what is needed, combined with diagnostic mechanisms adequate to repair the defects in the initial performance, until an acceptable level (acceptable in terms of the kind of environment in which the organism will find itself) of competence is eventually reached.
In this latter kind of picture, it is not enough that the system as a whole contain various mechanisms for constructing skills. It must at the same time deploy the mechanisms concerned in an appropriate manner, that is to say at the appropriate times, and to the right extent. Thus the processes of development involve more than the mechanisms of construction already considered. Many factors will affect the efficiency of the constructive processes, for example the way the person concerned searches for new situations in which to attempt to develop skills, and the ability to utilise the knowledge obtainable from others. In the last analysis, these factors must all be implicit in the nervous system design. To design such a system without any clues from nature would be extremely difficult, but the knowledge of how children develop in actuality can simplify the task of understanding these matters.
We hypothesise that for each construction mechanism, and process constructed, there is an associated activation of the nervous system, which triggers off particular secondary processes, and leads to the system having a specific orientation for the relevant period of time, generally persisting as long as the system receives reinforcing cues indicating that it is being effective, but modulated by other relevant factors. Over the course of time, higher order skills develop that use earlier constructions and take over command of their use (cf. [3]).
We have indicated that development occurs partly through the agency of mechanisms able to diagnose imperfections. In general, diagnosis can be defined as any process that leads, through the results of observation, to the improvement of a system or skill. On this basis we can distinguish between innate diagnostic systems that have evolved through natural selection, and diagnostic systems acquired during the life of an individual or culture as a result of the observation that a certain action in a given situation has beneficial consequences. The `credit assignment problem', that of deciding whether a particular process is actually the one responsible for a particular improvement, seems to be solved in practice by having a mechanism that treats such a process as tentative (making a `provisional link') and observes its effects over an extended period, developing a tendency to excite or inhibit that process as a result. Eventually such a process may be linked in to the system in such a way that the process is activated automatically, the system then being in the position of having to find ways of coping with any adverse consequences of so doing.
As an illustrative example of a specific generative mechanism, take the process of rising to a vertical position, which has a pivotal role of providing the initial situation in which locomotion by means of walking can take place. We hypothesise that there are specific circuits to mediate this process. When they are activated they may, in the earlier developmental stages, activate certain `instinctive movements' possessing the feature of causing the body to tend to rise, combined with holding on to objects for support. They then evaluate the outcomes in particular ways and make modifications in the direction of rising as far as possible, and being as nearly balanced as possible (which can be judged in terms of the amount of effort involved in holding on to a support). The proprioceptive system can be primed to learn these nearly balanced states and thus guide further attempts, which can be made with gradually reduced support from objects.
In the phrase `construction of skills', there lies implicit both bottom-up and top-down mechanisms. The word construction implies putting components together to make larger units, leading by a process of accumulation to the formation of a hyperstructure as described in [2]. On the other hand, a prescribed skill can be considered as a whole that, given mechanisms that cause the skill to emerge, places constraints on how the parts behave. There is an analogy between this and the way that in a laser the microscopic systems create the macroscopic field, while at the same time the macroscopic field acts back on the microscopic systems. In human systems there are a number of high level systems that behave in this kind of way, for example the life that an individual chooses to live (which is partly a product of his skills and partly the determiner of the skills that he acquires), the modes of being of a society, the particular language of a society and so on. To take the laser analogy further, the selection of a process to attempt corresponds in the laser to selection of a mode to excite, while errors correspond to attenuation mechanisms, so that effective diagnoses have the effect of reducing errors and hence allowing the relevant mode to grow.
At a less all-embracing level there are cases of limited specific fields of endeavour, as can be typified by the example of rock climbing. Such a category of projects poses especial difficulties, the overcoming of which can be equated with the possession of particular skills. Acquisition of such skills expands the range of problems that can be tackled in the given domain. The mutual relationships between skills and domains of endeavour provides a natural mechanism for the marking out of particular domains for development by an individual or by a culture, which selection has marked effects on subsequent advances, especially facilitating particular possibilities.