My own answer to the question was included after a list of suggested definitions. Here it is:
In science, arguing about a definition and attempting to produce a single binary division (e.g. between cognitive and non-cognitive systems/processes/mechanisms) is often a waste of time. What we need are surveys of the variety of possibilities, and comparative analysis of their implications, requirements, tradeoffs, etc. (The philosopher Gilbert Ryle referred to analysing the 'logical geography' of a system of concepts.) Instead of futile and unending debates about which is the best or right definition to use we can do more fruitful research into similarities and differences between many different subcases, whether in natural or artificial systems. For example, many definitions of cognition consider only human capabilities and would exclude the ability of insects to use landmarks. I am not saying there is a continuum of cases: on the contrary both in evolution and in the set of possible artificial systems there are many interesting discontinuities -- some big some small. We need to understand all of them. Focusing on one division that happens to interest particular researchers can divert research away from the more general kind research which in the long run will provide deeper insights into all the discontinuities.
Further discussion of Ryle's notion of 'Logical geography' and the deeper notion of 'Logical topograpy' can be found here.
School of Computer Science
The University of Birmingham