Moving MOSES Representation into Atomspace
I wonder if it would make sense, at some point, to replace the VertexTrees inside MOSES with comparable structures living inside the Atomspace?
If we implemented a general-purpose Atomspace rule engine, as has been discussed, then this would likely allow Reduct to be implemented as an application of this general rule engine.... This would go along well with Nil's long-sought idea of refactoring the Reduct rules to b
Or maybe it should go the other way around:
- The AtomSpace is highly tuned for PLN: it has hard-coded truth and attention values.
- It is difficult to extend the current atomspace to support general types (ints, strings, booleans, functions) due to the above.
- It is not clear what benefits there are from using the large, complex atoms to represent these types.
So... maybe it would be better to add atoms to the MOSES vertexes, and have MOSES work on that!?