Hyperon brainstorming documents

[Ben Goertzel]

Hi all,

If anyone wants to dig deeper into recent thinking by myself, Alexey
Potapov and multiple colleagues on an in-progress (still very early
stage) new OpenCog version, OpenCog Hyperon, we have put a bunch of
docs on the wiki linked from here

https://wiki.opencog.org/w/Hyperon#OpenCog_Hyperon

Bear in mind these are mostly early-stage raw discussion documents not
refined specs or anything like that. We are sharing aspects of our
thought-process-so-far with a view toward soliciting feedback and
potentially participation...

thanks
ben

--
Ben Goertzel, PhD
http://goertzel.org

“The only people for me are the mad ones, the ones who are mad to
live, mad to talk, mad to be saved, desirous of everything at the same
time, the ones who never yawn or say a commonplace thing, but burn,
burn, burn like fabulous yellow roman candles exploding like spiders
across the stars.” -- Jack Kerouac

[Curtis Michael Faith]

So... the Open Cog project is run in academia and not really good software land. AI is too specialized and HUGE you can't expect someone who is successful to have time to do it right.

I suggest you forget the name OpenCog. Forget the code. Forget Ben's first prototypes. Think Think think

we really all need to sit down together someplace in Europe or Korea... perhaps

--
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/CACYTDBfrZjp7gE4qE7fUF0fLK%3DFvP3nM3hnrUA%2Be8eXTx8pJnQ%40mail.gmail.com.

[Curtis Michael Faith]

The name and project need a ground up rewrite using lessons that Kenny Kasajian knows better than anyone. I am trying to get Kenny to help me build a team from the OpenCog project but I've been busy with trying to stay alive as I lost my wallet in Manchester, NH and IDs and haven't been able to open a new bank account since I left Portland 10 months ago. I have checks piling up I can't cash and had to really get clevver to live...

[Ben Goertzel]

Sorry to hear about your logistical hassles -- but good to hear from
you and know you're still around...!

We are doing a from-the-round rethink and redesign ... whether this
leads to a from-the-ground rewrite or not is still an open question...

As for the name of the hypothetical in-progress new version it's
"OpenCog Hyperon" for now... I am reserving "OpenCog Tachyon" for the
future closed-timelike-loop based version ;)




On Tue, Sep 15, 2020 at 4:47 PM Curtis Michael Faith

> To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CAJzHpFr1Wjjvatktqq519aOarD05Mwt4kxv_JZgNGQt_gBeu2Q%40mail.gmail.com.

[Noah Bliss]

To this end, has any thought been given to other implementation languages? Julialang seems to have a very strong userbase in the research field, is easy to pick up like Python, JIT so long compile times are generally a non-issue, and it executes with near-C levels of performance.

Just my 0.02, curious to see how this discussion develops. 

[Ben Goertzel]

We have been seriously considering Rust for new distributed Atomspace
code, assuming the latter gets written...

> To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/6517ab72-dcc2-4e3d-9403-6298458ae84en%40googlegroups.com.

[Noah Bliss]

Rust will be great. Especially for the less-frequently-modified components like the atomspace.

I don't know if it would fit the bill as well for the other more volatile components.

Opencog is not distributed as a binary and is rapidly modified. The primary development seems to be in the source code going in, and changes to that require lengthy recompile times. Making humans wait wastes their time and development potential. By having a highly-readable source code, we can make this development more rapid from development, experimentation, and execution perspectives. Julia can be static compiled if needed, it's not efficient from a file size perspective, but it's just as fast and I think rapid development and readable code offsets that easily, especially for developers just joining the project. 

Julia has phenomenal GPU support. It can leverage GPU memory and processors easily using native data structures. Allowing us to very easily scale out into high parallelization. Both for storage and processing.

Julia's images library shows great potential for a future ML vision system as it can store entire images as a vector data structure, allowing us to apply very flexible analytics to the input information.

ML libraries have already been built in Julia. https://github.com/FluxML/Flux.jl is the first that comes to mind, but others certainly exist. 

I don't want to be in denial. Fantastic code in rust would probably be more distributable and efficient than Julia, and if opencog ever went to a "release" type system it would be a great idea for all the components, but from my observations it seems Atomspace 2 will be the main "OK now we're serious" development area. It seems likely to me there will be an atomspace 3 once the science is perfected, but if rapid development is a priority for this stage of the project, it might be worth an evaluation. 

Can't wait to see the next evolution of Opencog!

[Ben Goertzel]

I like Julia for ML and numerical analysis stuff, definitely ... big
improvement over python...

We are in the midst of designing a new version of the "Atomese"
language which will be a functional language w/ gradual typing &
dependent higher order types ... but this will play for Hyperon more
of the role that {Scheme + Atomese} play in the current OpenCog
ecosystem, whereas Rust (if we go that direction) will play for
Hyperon more the role that C++ plays for the current OpenCog system

> To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/52c81a33-3e42-4ef4-879d-45846577658fn%40googlegroups.com.

[Ivan V.]

Considering the new version of Atomese being a functional language, may I ask what happens to URE (unified rule engine)? URE was one of the concepts I liked very much in OpenCog.

To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CACYTDBceJbWhLdEAWmvXD_%3DdUJReKnz%3DajF1vru4bx4FuwUtbQ%40mail.gmail.com.

[Ben Goertzel]

The functionality provided by URE is very much needed, but the current
"division of labor" between URE, Pattern Matcher, {scripts written in
(usually) Scheme invoking Pattern Matcher, URE and Atomspace API
directly} , and code written to ground GroundedSchemaNodes and
GroundedPredicateNodes -- seems suboptimal in various ways. All
these functions are needed but we are exploring organizing them
differently...

> To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CAB5%3Dj6WAQWJEa-Nf6BKipqgB62MnTcrtUiWbzz42dV1R06B7_Q%40mail.gmail.com.

[Michele Thiella]

Hello everyone, i'm Michele, a master's degree student in computer engineering at the University of Padua,
fascinated by the point of view of OpenCog on the idea to build AGIs (GPT-3 doesn't seem the right way).

I don't think it's the correct section to ask,
but I would like to approach this project in some way (master's thesis to begin with), but the link https://opencog.org/cgi-sys/suspendedpage.cgi returns "Account Suspended"..
What happened?

[Ben Goertzel]

We're fixing it!

Welcome ;)

> To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/6edb111f-c036-4e3e-90d6-908d75ceb962n%40googlegroups.com.

[Michele Thiella]

Thanks! Great!
Currently I have built the opencog repo thanks to the readmes,

even if some tests fail (for example https://github.com/opencog/atomspace/issues/1748, where I still don't understand how to skip the database part or to configure it).

To say that I am a beginner in this field is an understatement, but I am well motivated to do something even insignificant or just to learn.

I will read and try to learn as much as possible about this project.

I hope not to disturb and, indeed to be able to make some contribution (if i can).

[Jennifer Elizabeth]

Thank you Ben! :) PS check your fb sometime regarding consciousness architecture and see if there's anything in the group chat that may help contribute towards hyperon (warning: it's a very very long chat, but also feel free to lurk still for any future "rosetta stone" moments!)

Virus-free. www.avg.com

To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CACYTDBfPr0N%2B3YWOkJ-S%3DCAXz8dpzBhyKLaw8D79%2BQavfC_MvA%40mail.gmail.com.

Virus-free. www.avg.com