Since neither designs nor requirements are static, and since changes in each can produce changes in the other, we can think of various sorts of causally related trajectories in design space and in niche space: Evolutionary trajectories, Individual development trajectories, Social/Cultural trajectories, and in the engineering world, what I've misleadingly labelled "Repair trajectories" since an engineer can take a working system and do some repairs, possibly stopping it functioning for a while in the process, unlike all the other trajectories which involve changes in continuously functioning systems.
One way seems to be to consider changing pressures from an environment that starts simple and then gets more complex in various ways. A consequence is that the designs of some of the organisms get more complex. Those new more complex designs together with the features that they are responses to provide constraints on the control mechanisms that are needed, or more generally the required information-processing architecture.
E.g. we can think of microbes in chemical soups, using only chemical gradients as a basis for decision making about where to move, and organisms that need to cope with larger scale persistent structure in the environment, or organisms that cannot simply approach and ingest their nutrients, but have to break food open, using teeth, beak, hands, or other tools.
The more complex the structure of the environment, as measured for example, by the variety of types of relationship, and changes in relationship that the organism needs to perceive, explain, think about (e.g. find explanations, make plans to produce or avoid things), act on, or control, the harder it is for the knowledge involved to be expressed in patterns of relationships between inputs and outputs. (Compare these comments on Margaret Boden's discussion paper on creativity in machines.)
Some of the sources of complexity are discussed in this paper on the need to take account not only of the existence of and changing relationships between surfaces, shapes, and objects, but also the many kinds of 'stuff' of which objects can be made.
I have tried to argue elsewhere that this has led to the development of abilities to understand structures and processes in the environment in such a way as to enable their consequences to be 'worked out' instead of having to be based on statistical predictions. The slides give many examples but here are two cases
There are very many more cases in which we can distinguish statistical or probabilistic learning, learning of correlations, from the ability to work out from more general principles what must happen or what cannot happen, in a specific structured situation.
I suspect that humans are not the only species that can do this, and that some of the capabilities evolved independently in some birds, in primates, in elephants, and possibly other species.
School of Computer Science
The University of Birmingham