I will show evidence that my theory's generality by by implementing it in a running computer program and making it work on several examples. The theory, as described in the section above, is a more complete description of how agents both humans and artificial agents might solve problems. The implementation will focus on a subset of these claims. For example, the agent will choose a problem solving strategy, but analogy will be the only one available.
I have already implemented some of the theory in an agent called Galatea. The fact that I already have a program working with two examples (the fortress/tumor and the Maxwell case) provides initial evidence for all three of my high-level hypotheses (Davies & Goel, 2001; Davies, Nersessian & Goel, 2002). In incorporating more examples from the Craig data, the expanded program will further support them, if indeed it works.
It has no non-visual representations and cannot do analogical mapping. It has the machinery for analogical problem solving for visually represented source and target problems, and much of this will be used for implementing the non-visual as well. The expanded program I will make for my dissertation evaluation will be called Proteus. First I will describe Galatea, which works for the fortress/tumor problem and the Maxwell case, and then I will describe what more Proteus will do.
The knowledge representation, at an architecture level, consists of propositions: connections of two ideas or propositions with a relation. The substance of the theory is Covlan, in the higher level visual knowledge types.
Galatea currently takes as input 1. a solved source problem, 2. an unsolved target problem (both represented visually), 3. an analogical mappings between the s-images, and 4. criteria for an adequate problem solution. When instructed to solve the target using the source, it analogically transfers the solution procedure. As can be seen in Figure 10, it outputs a series of s-images for the target problem, and checks to see if the solution transferred indeed solves the problem constraints.
The agent Proteus will be an extension of Galatea. It will be expanded as needed to model the Craig data problems, and make the models of the fortress/tumor and Maxwell case problems theoretically consistent. It will also implement much of the non-visual problem solving and representation as outlined in the theory section.