The future of the project

[patham9]

ONA received only minor updates since the AGI-23 release of v0.9.2 in June 2023.

But now big changes are underway and v0.9.3 is getting closer.

Time-less implication links between concepts derived from functional equivalence, e.g. when learned

<(<(loc1 * X) --> shape> &/ ^left) =/> G>

and then for instance much later learning

<(<(loc * Y) --> shape> &/ ^left) =/> G>

the system can despite time gap effectively learn

<<($1 * X) --> shape> ==> <($1 * Y) --> shape>>

and

<<($1 * Y) --> shape> ==> <($1 * X) --> shape>>

which are stored in the new (timeless) implication links table of concepts <($1 * X) --> shape> and <($1 * Y) --> shape>.
Essentially the system can learn that X and Y can replace each other (to different degree) without having to be given an explicit <X --> Y>, <Y --> X> or <X <-> Y> relation (and ==> being more reliably controlled now due to the table).

Unified <(location * value) --> attributeName>. :|: representation of stimuli as we learned from the AniNAL experiments, which will lead to many examples getting updated.

http://91.203.212.130/AniNAL/demo_complex_continuous_verbal.html

http://91.203.212.130/AniNAL/demo_complex_continuous_shades.html

https://github.com/patham9/AniNAL

Overhaul of declarative inference

As we are moving to making the system acquire meaning of terms through sensorimotor rather than relying on human-provided concepts, we might find issues with the old ways of how NARS handles inductive reasoning on event streams and the declarative inference pathway in general. Progress with AniNAL will show us the way. For now

#define SEMANTIC_INFERENCE_NAL_LEVEL 0

allows to deactivate the previous declarative inference pathway, relying only on the new mechanisms which allow NARS to derive meaning of terms by itself by ways of how they can substitute each other without changing behavior.

Unification with NACE

Likely beyond v0.9.3 will be unification of ONA principles with NACE principles of curiosity-based exploration and coherent prediction of the future from individual predictions + choice rule, and without reliance on a grid world.

Tricky will be the coherent forward prediction which ONA does not yet have but seems to be crucial for more wholesome situation-based planning.

https://github.com/patham9/NACE

Good however is that NACE already relies on the idea of <(location * value) --> attributeName>. :|: however can only have one attribute per location and the locations are tied to the grid locations.