Development Plan

A lot already has been implemented, including:

• SDR's and all necessary utility functions for encoding compound terms in SDR's.
• Events and implications.
• Event FIFO's with revision ability for goal conflict and sensory value conflict resolution.
• All the truth functions
• Inference rules.
• Voting schema
• Double-ended priority queue for attention.
• Attention-related functions in ALANN spirit.
• Basic SDR encoders for terms and numbers
• Stamp for checking evidental base overlap to avoid cyclic reasoning
• Concept containers, all stored in statically allocated memory
• OpenNARS style belief tables for pre- and post-condition implications, ranked by truth expectation.
• ALANN control cycle as described in https://github.com/patham9/ANSNA/wiki/Working-cycle
• tagging events with operations (SDR_Union) to learn and execute procedure knowledge.
• decision making rule (same as in OpenNARS): generate revised goal if possible, if not (overlap or it was the first in the table), for operation goals, check if above decision threshold, if yes execute the operation goal's associated C function.
• Conceptual interpolation

left to do is:

• exploring commonality and difference-based decomposition rules.
• investigating variables
• making sure the entire picture is complete: https://github.com/patham9/ANSNA/wiki/System-Overview
• showing what ANSNA can do that other systems cannot! :)

Expected capability:

• Dealing with rich sensorimotor experience, much richer than OpenNARS could handle.
• Effective procedure learning, applicable for real autonomous systems.
• Ability to learn useful conceptual representations.

A short-term goal is to make ANSNA replicate OpenNARS's ability to learn Pong, to deal with more complex environments like TestChamber: https://www.youtube.com/watch?v=hAO1zRj2z9A and to learn temporal correlations and make predictions based on them in a similar fashion as OpenNARS demonstrates here: https://www.youtube.com/watch?v=TlxoZ64cVvQ