Experiments and prototyping for interactive learning agents.
Version 0.0.2 -- Extremely early prototyping stage.
The Opencog Learn project showed that the basic conscept of structure learning works and is scalable and can deal with observing and learning real-world structures. Great!
But it was built as a batch system, training on large blobs of data, moving from one training stage to the next. This turns out to be unmanagable: its complicated, its fragile. It's hard to track status and to run experiments. Hard to redo batch steps when bugs get fixed and system architecture changes.
Most importantly, the above is a roadblock to scaling: the core idea is that the learning can be done recursively. But if the recursive steps teeter precariously on earlier batch runs ... ugh. It became too hard to develop the recursive step.
Rather than refactoring the code in that git repo, it seems better to start with a clean slate, here. The general ideas get recycled. The framework changes dramatically.
One current goal is to rewrite important parts of the pipeline in pure Atomese. This seems like the "right thing to do", so that, for example, AS-MOSES can be used to mutate and optimize important subsystems: this would be impossible if the processing was done in a conventional programming language. But these details are unclear, and so all of this is a big experiment. Don't know how it will turn out.
The hope is that a pure Atomese structure will make the recursive jump easy. Without the recursive jump, well, its just Symbolic AI from earlier decades.
A general overview can be found in the AGI 2022 paper: Purely Symbolic Induction of Structure.
Background reading: the various PDF and LyX files located at:
- AtomSpace Sheaves & Graphs
- OpenCog Learn Project and especially the "diary" subdirectory there.
Additional general details and concepts are described in scattershot locations, in diary entries, or nowhere at all.
The demos here are built on top of the perception-action framework developed in the Sensory project. The code here adds some non-trivial processing steps on top of the basic sensori-motor API provided there. Be sure to go through the examples there.
For examples in this project, see the examples directory here.