> linus, an outline of the potential new atomspace is here.
>

To be clear, we (the group of us in SingularityNET working on Hyperon,
alongside our other projects) do not yet have a detailed design for a
Hyperon Atomspace, though we do have a lot of ideas that are in an
intensive form of refinement.

Our hopeful timeline is

-- by end of August, to have a coherent, complete-ish description of
the type-system we want for "Atomese2" (the new programming language
that will be used to create and represent structures in the Hyperon
Atomspace). [this is down to Alexey Potapov and  myself]

-- by around the same timeframe, to have a specific proposal for the
distributed-processing and backing-store aspects of Hyperon Atomspace,
leveraging existing key-vector stores and such ... (DAS, Distributed
AtomSpace)

-- by around Sep 15, to have some decisions on what tools we want to
use for implementing Atomese2 and DAS

(e.g. what programming language to implement Atomese2, what key-value
stores as the basis for DAS...)

Note that none of the above are specifically about the in-RAM
"Atomspace" in terms of the containers behind the Atomspace API ...

It would be possible to have a new Atomese2 language and DAS, working
together with the current Atomspace.  On the other hand, if we're
implementing a new Atomese2 and DAS, it may end up being most
effective to implement a new local in-RAM Atomspace as well... this
will be considered in mid-September ...

Note that the SingularityNET team that is pushing this Hyperon design
initiative is also heavily using the existing OpenCog system, eg for
the Grace humanoid robot software system, for biomedical AI R&D, and
in the ROCCA project aimed at AI agents collectively playing
Minecraft.  So the motive for redesigning/rebuilding is not a feeling
that the current Atomspace/OpenCog is bad, but rather that through
this work (and prior work) we are seeing that some things that are
fairly awkward in the current system could be made much
cleaner/easier/more-efficient in a variant system with some different
characteristics...  (key examples are real-time learning interactions
btw external deep learning software and OpenCog; and large-scale
logical-inference meta-learning...)

The Hyperon wiki page contains links to a quite a lot of our
early-stage thinking on all this as it's been evolvlng over the last
year...





--

-- 
You received this message because you are subscribed to the Google Groups 
"opencog" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/opencog/CACYTDBduKz%2BcPHTpmaJvMRJS3Y4y56u6RGrHRzaDVq62Ysy3cA%40mail.gmail.com.

Reply via email to