Cognitive Development

Suppose stable patterns in behavior themselves change over time, that is, suppose that human behaviors change over the larger time-scale of human development. Then the question is ofcourse: what are the general characteristics of such a developmental change, and how does such a developmental mechanism work? Again, I chose to assume that the changes in behavioral regularities over developmental time are self-organized patterns in a complex system: in this case the developing cognitive system. So we already encounter the subject of levels and timescales: at a fast timescale behavioral variables change in regular ways, forming stable patterns. But at a slower, developmental timescale, the form of these stable patterns themselve changes over time during the course of development. Now there are two general views on how these two levels interact. The one is that they don’t interact, because the developmental change is purely and completely governed by the initial conditions of the developental system. That is, a particular development takes place no matter what, and what is developed is by effect purely explained by innate factors. This is the nativist point of view. The other view is that the two levels do interact. As each stable pattern is in fact an interaction with the environment, we could propose a mechanism that uses the consequences of your behavior as a guideline to the direction of the developmental change. So, in effect, you learn from doing. But the specific developmental change will ofcourse change the set of stable patterns and their specific forms, so that ultimately you will start to behave differently. If in the previous section I talked about interaction between agent and environment, here's another type of interaction: The developmental level and the behavioral level influencing one another.

To recap:

• Complex systems have several levels, with several accompanying timescales indicating the rate of behavioral change at these levels.
• Complex systems have several components, which are relatively independent in their behavior but at the same time influencing each other.
• Components and levels are ofcourse part of the greater family of subsystems: levels, such as the immediate behavioral- versus the developmental level, can be seen as subsystems divided on time-scale, whereas components, like agent versus environment can be seen as subsystems divided on spatial-scales.

One of the interesting existing type of models with which to investigate complex systems are neural networks. Neural networks are sets of behavioral variables changing over time, and the interesting thing about them is that one can specify certain connections between them (interactions) after which patterns stable patterns often emerge. Furthermore, one can construct neural networks that change their connections over a developmental timescale and what we see is that by doing that the stable patterns also change over this developmental timescale. And last but not least, we can device a mechanism, called the learning mechanism, by which the the actual behavior and its consequence in the environment influence the way in which the connections change. So in sum we have:

• A set of behavioral variables changing over time
• …that are relatively independent but also connected with each other
• …leading to regular patterns at the system level
• The possibility of developmental change by changing the connections
• …leading to a change of the stable patterns
• A learning mechanism that embodies the interaction between the behavioral and the developmental level

If neural network models would ultimately behave like humans, that is, if they would show X-like behavior as we do, and if they would show some characteristic pattern in the developmental change like we do, then we would have a point in saying that perhaps behavior is governed by a self-organizing process in a complex system, and development is governed by a self-organizing process in a complex system, and that development is not purely governed by innate factors alone.

In my thesis I focussed on this characteristic pattern that humans show on the developmental time-scale. According to a lot of empirical research, cognitive development in humans comes in stages. Now Piagetians have deviced a lot of tasks to measure this and the thing is that in all of these tasks children show discontinuous jumps on the performance-scores of these tasks. That is, you do not gradually improve on these tasks, but you stay on a lower level of performance for some time and then you suddenly jump to a higher level of performance. This they called stagewise development and in my thesis I wanted to model this pattern at the developmental timescale using a neural network model.