The document starts
The overall goal proposed here is to construct physically instantiated systems that can perceive, understand, and interact with their environment - but also evolve in order to achieve human-like performance in activities requiring context-specific knowledge.I posted the following comment on 15 Feb 2006
But there are many kinds of competences that biological evolution has produced including the ability to think about the past, the distant future, remote or inaccessible happenings in the present, and to construct and perform operations not only on physical objects, such as food, nests, weapons, tools, obstacles, but also abstract objects such as plans, proofs, unanswered questions, and the thoughts, goals, preferences, and intentions of other individuals.
More subtly, unlike most other animals, humans are able not only think about our own actions and their consequences, but also those of others (e.g. children, predators, etc. -- I call this seeing 'vicarious affordances'), and we can understand and reason about what is common between grasping and moving something with our teeth and grasping it with a hand, or a pair of pliers, or a remote controlled motorised mechanical device Understanding that commonality, and using it to make predictions, requires representing grasping independently of the specific sensory motor contingencies involved. (I suspect most animals cannot do that.)
Focusing on the particular collection of problems identified in this Consultation Report might be a good way to study what humans have in common with insects, but it seems perverse to exclude study of cognitive competences that are distinctly human, or in some cases shared between humans and other altricial species, such as nest-building, tool making birds, hunting mammals and other primates.
There is a way of surveying the space of possible research goals in terms of a grid of topics exploring how
(a) different sorts of competences (represented by rows in the grid) can be applied toThen most research proposals in AI, Cognitive Science, robotics, etc. can be seen as addressing specific subsets of the 3-D requirements grid. There's nothing wrong with that as long as the proposals don't include spurious explicit or implicit claims that ONLY those subsets are worth investigating.
(b) different sorts of entities (represented by columns in the grid -- including both concrete, abstract and social entities)
(c) at different levels of sophistication and complexity, represented by depth in the grid cells (more precisely a partial ordering rather than linear depth).
A first draft overview of the requirements grid can be found here:
Discussion paper: 'A Conceptual Framework and Draft Tool for Generating Requirements and Scenarios'
A related document on learning orthogonal recombinable competences, in altricial species, is http://www.cs.bham.ac.uk/research/projects/cosy/papers/#dp0601
Comments and criticisms, and suggestions for improvement welcome.