Time Theory

[Mike Archbold]

I'm doing some research for the group I organize, the Northwest AGI Forum. I'm trying to get a feel for how advanced time theory is in architectures. Time obviously is important.

I'm familiar with OpenCog in general terms, and over a the years I have studied it, although I don't know the details.  I was browsing through the wiki. I see references to modal logic, PLN, the atomspace, etc.  It's hard to tell though what is speculative theory, what is implemented, what is planned, long term envisioning etc. I know Ben et al has done a lot of work.

So I was wondering if someone could please give a bit of an overview as far as the general state of time representation in OpenCog, or perhaps point me to the relevant literature.

I really appreciate any help! I want to summarize time theory in OpenCog, NARS, and SOAR. 

Thanks Mike Archbold

[Ben Goertzel]

Nil is doing a lot right now with temporal reasoning in PLN for
learning/planning in Minecraft. So temporal representation is an
active area, though indeed what's implemented and what's envisioned
are not fully equivalent. So best to interface w/ Nil on this...

> --
> 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/abe64f28-da56-4ec4-84d5-8c51d3eab78dn%40googlegroups.com.



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

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

[Linas Vepstas]

Aside from Ben's reply (which suggests talking to Nil about his work on temporal reasoning in PLN) I can add in my 2 cents, which is: nothing that I've worked on deals directly with time in any way :-)

> It's hard to tell though what is speculative theory, what is implemented

I can be very precise about this, if you ask the right questions. Very briefly:

* The atomspace is very mature, debugged, performance-tuned and stable. This includes the query engine, the 120+ Atoms that constitute Atomese, the notion of "Values", the distributed database backend, and the "matrix" layer (it can take arbitrary slices of the AtomSpace contents and expose them as a sparse matrix (of rows and columns), thus allowing conventional probability theory to be applied (e.g. conditional probability of row given column, etc.)  The Atomspace has been used/applied in various ways in at least a dozen different projects (that's how it got mature - trial by fire).

* The Unified Rule Engine (URE) is mature, but has only been used in one project: PLN. It was supposed to provide a generic way of running rules, but no one aside from PLN ever used it for that.  (one of half-a-dozen examples from a long time ago was to put the NARS rules on top of URE)

* PLN is a specific collection of rules. Some of these are fully worked out and mature; some are experimental, some are in active development. Nil can say more.

* I'm very actively developing code for generalized learning (currently, language, in general, vision and sound; yes there's a unified theory that covers all this and more.)

* Everything else is immature, incomplete, abandoned and/or bit-rotted. This includes over a dozen different subsystems, one of which is a "spacetime server" that can record things/events/generic-stuff in space and time.  Some of this stuff should be brought back from the dead, some should be allowed to rest in peace.

I've one comment about time and temporal theory, and I suspect you'll hate it. I strongly believe that an AGI system must learn about time, (and how to reason about events in time) instead of having it hard-coded into it. Likewise, it must learn "common sense", and only after that, can it learn about rationality and reasoning. Thus, I spend all my effort on learning; I expect it to (eventually) learn common sense and how to reason about time.  This is why I (personally) don't work on any explicit theories of time.

This differs from older (dare I say "conventional"?) architectures, where all this stuff (space, time, reasoning, logic) is hard-coded in at a base layer.

-- Linas

--
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/abe64f28-da56-4ec4-84d5-8c51d3eab78dn%40googlegroups.com.

--

Patrick: Are they laughing at us?

Sponge Bob: No, Patrick, they are laughing next to us.

 

[Nil Geisweiller]

Yeah, Hedra and I are actively working on temporal reasoning, the
documentation is not well centralized at this point, and we're in
exploratory phase but with functioning code and an agent able to do
planning in uncertain/unknown environments using temporal/procedural
reasoning (it uses temporal pattern mining as well, but that can be seen
as specialized form of temporal reasoning).

Here are a few pointers

- PLN book (chapt 14) http://goertzel.org/PLN_BOOK_6_27_08.pdf
- Wiki page https://wiki.opencog.org/w/Category:Temporal_Reasoning
- Presentation at AGI-21
https://odysee.com/@ngeiswei:d/AGI-21-Temporal-Procedural-Reasoning-Nil-Geisweiller-Hedra-Yusuf:6
- PLN repo https://github.com/opencog/pln (see for the currently
implemented temporal rules and their documentation
https://github.com/opencog/pln/tree/master/opencog/pln/rules/temporal)
- ROCCA repo (reasoning-based agent control in unknown environments)
https://github.com/opencog/rocca

Nil

[Nil Geisweiller]

On 2/3/22 04:15, Linas Vepstas wrote:
> * The Unified Rule Engine (URE) is mature, but has only been used in one
> project: PLN. It was supposed to provide a generic way of running rules,
> but no one aside from PLN ever used it for that.  (one of half-a-dozen
> examples from a long time ago was to put the NARS rules on top of URE)

The pattern miner https://github.com/opencog/miner and a (very
experimental) implementation of the AS-MOSES reduct engine
https://github.com/opencog/asmoses also use the URE.

> * PLN is a specific collection of rules. Some of these are fully worked
> out and mature; some are experimental, some are in active development.
> Nil can say more.

On point.

> I've one comment about time and temporal theory, and I suspect you'll
> hate it. I strongly believe that an AGI system must learn about time,
> (and how to reason about events in time) instead of having it hard-coded
> into it. Likewise, it must learn "common sense", and only after that,
> can it learn about rationality and reasoning. Thus, I spend all my
> effort on learning; I expect it to (eventually) learn common sense and
> how to reason about time.  This is why I (personally) don't work on any
> explicit theories of time.

It's probably true in theory, though in practice it helps to have
readily available temporal representations/rules/procedures.

Nil

> <mailto:opencog+u...@googlegroups.com>.

> To view this discussion on the web visit
> https://groups.google.com/d/msgid/opencog/abe64f28-da56-4ec4-84d5-8c51d3eab78dn%40googlegroups.com

> <https://groups.google.com/d/msgid/opencog/abe64f28-da56-4ec4-84d5-8c51d3eab78dn%40googlegroups.com?utm_medium=email&utm_source=footer>.

>
>
>
> --
> Patrick: Are they laughing at us?
> Sponge Bob: No, Patrick, they are laughing next to us.
>
>

> --
> 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

> <mailto:opencog+u...@googlegroups.com>.

> To view this discussion on the web visit

> https://groups.google.com/d/msgid/opencog/CAHrUA37sWFaCMssDKHzfLnYuv%2B4sM%3DR_F2HfR-X%2B4uH9ou48BA%40mail.gmail.com
> <https://groups.google.com/d/msgid/opencog/CAHrUA37sWFaCMssDKHzfLnYuv%2B4sM%3DR_F2HfR-X%2B4uH9ou48BA%40mail.gmail.com?utm_medium=email&utm_source=footer>.

[Linas Vepstas]

On Wed, Feb 2, 2022 at 11:40 PM Nil Geisweiller <ngei...@googlemail.com> wrote:

On 2/3/22 04:15, Linas Vepstas wrote:
> I've one comment about time and temporal theory, and I suspect you'll
> hate it. I strongly believe that an AGI system must learn about time,
> (and how to reason about events in time) instead of having it hard-coded
> into it. Likewise, it must learn "common sense", and only after that,
> can it learn about rationality and reasoning. Thus, I spend all my
> effort on learning; I expect it to (eventually) learn common sense and
> how to reason about time.  This is why I (personally) don't work on any
> explicit theories of time.

It's probably true in theory, though in practice it helps to have
readily available temporal representations/rules/procedures.

I think I know how to make it work in practice, but my ideas on this remain very speculative, and at the current rate of progress, will remain so into the foreseeable future. I will try to write them up in a paper, because why not. Everyone else is doing it :-)

-- linas