Neural-symbolic approach to exploratory learning in MInecraft

[Ben Goertzel]

Check this out from Alexey Potapov,

https://blog.singularitynet.io/towards-general-purpose-minecraft-ai-agents-41fe415d6dd

While it does not use OpenCog, the work was done as a preliminary step
toward doing similar but more advanced stuff in OpenCog Hyperon (this
uses some NNs together with a custom specific rule engine, the latter
made for short-term use while waiting for Hyperon MeTTa interpreter to
mature...)

ben

--
Ben Goertzel, PhD
b...@goertzel.org

"My humanity is a constant self-overcoming" -- Friedrich Nietzsche

[Keyvan Mir Mohammad Sadeghi]

I'm having flashbacks from 2012 :)

Great to see those efforts coming back to life👌

We had the Metaverse in OpenCog way back when, and we had them with cute AI agents instead of shitcoiner strangers🙈 

--
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 opencog+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CACYTDBfSYgVXa_vB%3Doq3PGnh1z%2BvAHZymLP6ovQsgT%3DxA-XczQ%40mail.gmail.com.

[Ben Goertzel]

Well wait till our Sophiaverse AI/blockchain metaverse comes out, you
will then have cute AI agents jamming music and creating art together
with shitcoiner strangers -- plenty of opportunities for
quantum-illogical constructive, destructive and transtructive
interference ;D

> To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CAGx5hRmUF%3DAvvnLka%3Dh%2B%2BFuWiKO5g%2Bcy70xn--tFF%3DzTmKazLw%40mail.gmail.com.

[Ben Goertzel]

But yeah, Hyperon-in-Minecraft is basically the return of the son of
the bride of the daughter of the gastroenterologist of the "OpenCog
for Unity3D blocks-world" stuff we were doing back in HK at PolyU ...
the good news is the AI tools work way better now ;) ... and the deep
RL folks still have not solved the problems we were working way back
when

The intercombination of navigating, planning and building that
Alexey's doing now has some resonances w/ Shujing Ke's OpenCog planner
work from way back when as well...

Keep an eye out soon for some similar-ish reports from Nil's group on
their use of PLN pattern mining and PLN temporal inference for simple
Minecraft RL-ish learning ...

These threads of research from Alexey and Nil are going to converge in
some Hyperon MeTTa code later this year, along with some incarnation
of Doug Miles' episodic memory architecture...

This is one of the key applications we're using to test Hyperon as we
develop it, along with the good old bio-AI genomic reasoning
application, and the Grace robot back-end...

ben

[Adrian Borucki]

Ah, so that’s the project that was mentioned some months ago (during some AGI Discussion Forum).

I’m curious as to how the neural net-rule integration works — if it’s not OpenCog then I’m guessing it is based on something like DeepProbLog? Or is it in a different vein altogether?

I remember earlier attempts at using neural nets together with PLN, I wonder if this time there is end-to-end learning of those components with the rest of the system or is there some pre-training that produces “sub-symbolic skills” that the agent later uses.