Trying to understand the inference control problem.
91 views
Skip to first unread message
Ken Kafieh
Feb 24, 2017, 9:10:44 PM2/24/17
to opencog
I want to help.
My impression is that "inference control" is the principle hurdle for opencog, and other AGI efforts. I would like to try my hand at solving/reducing this problem. I have a general idea of the problem but I am sure I need a better handle on it. Is there a good description --maybe with examples-- out there that someone can point me to? Or if anyone is so inclined, maybe they could describe it here? Googling "inference control problem" just returns a lot of discussion on how to prevent a researcher from inferring secret info via legitimate queries.
Ben Goertzel
Feb 24, 2017, 9:33:03 PM2/24/17
to opencog
Ken, I am traveling in rural ethiopia with very limited internet till Monday , and will reply to your query when I'm back to Addis Monday or Tuesday
What is your background regarding programming, CS and logic, btw?
On Feb 25, 2017 05:10, "Ken Kafieh" <kka...@gmail.com> wrote:
I want to help.My impression is that "inference control" is the principle hurdle for opencog, and other AGI efforts. I would like to try my hand at solving/reducing this problem. I have a general idea of the problem but I am sure I need a better handle on it. Is there a good description --maybe with examples-- out there that someone can point me to? Or if anyone is so inclined, maybe they could describe it here? Googling "inference control problem" just returns a lot of discussion on how to prevent a researcher from inferring secret info via legitimate queries.
--
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+unsubscribe@googlegroups.com.
To post to this group, send email to ope...@googlegroups.com.
Visit this group at https://groups.google.com/group/opencog.
To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/ff845e5b-db4e-4d70-b8fc-253dfa9ad97f%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
Ken Kafieh
Feb 25, 2017, 12:42:46 AM2/25/17
to opencog
Thanks Ben,
My background in CS is modest as are my programming skills. I have a BA in computer science. I have worked various IT jobs over the years -- none of which involved serious programming. Sorry if I gave the impression that I would code (although down the road I think I would like to). I was just hoping to put my head together with a buddy and see if we could generate some ideas -- at a theoretical level.
I hope its not too naive to hope that i can make some progress in this area. I just feel that anything that goes on in our brains can be translated into code. But if I am wrong then at least laying out such a daunting problem in clear terms and posting it would help invite others to tackle it.
Nil Geisweiller
Feb 26, 2017, 3:49:05 AM2/26/17
to ope...@googlegroups.com
Hi Ken,
It's not naive, quite the contrary, it means your mind is at the right
place.
I see that the wiki page is lacking in content or links regarding that
matter http://wiki.opencog.org/w/Inference_Control. We need to remedy that.
Here's a synthetic answer till Ben writes a more elaborate one.
In one word: learning.
In one sentence: Mine the history of inferences and ask the system to
come up with inference control patterns that may speed up future inferences.
Of course the problem of learning inference control patterns is
open-ended, ultimately you need an AGI to address that. But wait, we
have one! That's what we're building!
Nil
place.
I see that the wiki page is lacking in content or links regarding that
matter http://wiki.opencog.org/w/Inference_Control. We need to remedy that.
Here's a synthetic answer till Ben writes a more elaborate one.
In one word: learning.
In one sentence: Mine the history of inferences and ask the system to
come up with inference control patterns that may speed up future inferences.
Of course the problem of learning inference control patterns is
open-ended, ultimately you need an AGI to address that. But wait, we
have one! That's what we're building!
Nil
Jeremy Zucker
Feb 26, 2017, 5:33:13 AM2/26/17
to ope...@googlegroups.com
Hi Ken,
To expand upon Nils characterization of inference control, I recommend reading Garry Sussmans short essay on robust systems. https://groups.csail.mit.edu/mac/users/gjs/6.945/readings/robust-systems.pdf
It introduces the conceptual background (with references) that explain why inference control is necessary and provides a lot of specific examples from biology, engineering and software design for how to achieve it.
Sincerely
Jeremy
It is also pretty inspiring
--
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+unsubscribe@googlegroups.com.
To post to this group, send email to ope...@googlegroups.com.
Visit this group at https://groups.google.com/group/opencog.
To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/952561b4-eb4f-21e5-1a15-25d980bc2a5c%40gmail.com.
Amirouche Boubekki
Feb 26, 2017, 9:02:21 AM2/26/17
to opencog
How does PLN, backward/forward inference has to do with that? Is it the same thing as unification and subsumsion?
Thanks in advance.
Amirouche Boubekki
Feb 26, 2017, 9:03:58 AM2/26/17
to opencog
On Sunday, February 26, 2017 at 9:49:05 AM UTC+1, Nil wrote:
Hi Ken,
On 02/25/2017 07:42 AM, Ken Kafieh wrote:
> Thanks Ben,
[...]
I see that the wiki page is lacking in content or links regarding that
matter http://wiki.opencog.org/w/Inference_Control. We need to remedy that.
Ken Kafieh
Feb 27, 2017, 3:21:58 PM2/27/17
to opencog
In one word: learning.
In one sentence: Mine the history of inferences and ask the system to
come up with inference control patterns that may speed up future inferences.
So you are saying that you should use learning algorithms to pick out potentially useful inference chains. I feel like our brains probably approach it in a similar way so I guess I can agree that is a good starting point. But I am wondering if it may be limiting the set of problems that can be solved to those that we humans can already solve. To solve the hard problems -- cancer, aging, Rubiks cube -- maybe we need wildly unconventional thinking sequences. For something as important as inference maybe it would be warranted to find a few shortcuts instead of relying totally on learning -- at least in the beginning if not always.
Nil Geisweiller
Feb 27, 2017, 10:18:38 PM2/27/17
to ope...@googlegroups.com
On 02/27/2017 10:21 PM, Ken Kafieh wrote:
> So you are saying that you should use learning algorithms to pick out
> potentially useful inference chains. I feel like our brains probably
> approach it in a similar way so I guess I can agree that is a good
> starting point. But I am wondering if it may be limiting the set of
> problems that can be solved to those that we humans can already solve.
> To solve the hard problems -- cancer, aging, Rubiks cube -- maybe we
> need wildly unconventional thinking sequences. For something as
> important as inference maybe it would be warranted to find a few
> shortcuts instead of relying totally on learning -- at least in the
> beginning if not always.
You're absolutely correct. Learning here is meant very broadly, it's
> So you are saying that you should use learning algorithms to pick out
> potentially useful inference chains. I feel like our brains probably
> approach it in a similar way so I guess I can agree that is a good
> starting point. But I am wondering if it may be limiting the set of
> problems that can be solved to those that we humans can already solve.
> To solve the hard problems -- cancer, aging, Rubiks cube -- maybe we
> need wildly unconventional thinking sequences. For something as
> important as inference maybe it would be warranted to find a few
> shortcuts instead of relying totally on learning -- at least in the
> beginning if not always.
more like AGI/reasoning/learning. But the idea is to enable full
introspection so that the system can devote its attention to reasoning
itself, as a skill.
Nil
>
> --
> 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
> <mailto:opencog+u...@googlegroups.com>.
> To post to this group, send email to ope...@googlegroups.com
> <mailto:ope...@googlegroups.com>.
> Visit this group at https://groups.google.com/group/opencog.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/opencog/6e654517-e88d-49e8-9551-d1bce4ff2384%40googlegroups.com
> To view this discussion on the web visit
> <https://groups.google.com/d/msgid/opencog/6e654517-e88d-49e8-9551-d1bce4ff2384%40googlegroups.com?utm_medium=email&utm_source=footer>.
Nil Geisweiller
Feb 27, 2017, 10:49:14 PM2/27/17
to ope...@googlegroups.com
On 02/26/2017 04:02 PM, Amirouche Boubekki wrote:
> In one sentence: Mine the history of inferences and ask the system to
> come up with inference control patterns that may speed up future
> inferences.
>
> Of course the problem of learning inference control patterns is
> open-ended, ultimately you need an AGI to address that. But wait, we
> have one! That's what we're building!
>
>
> How does PLN, backward/forward inference has to do with that? Is it the
> same thing as unification and subsumsion?
Ultimately inference can be used to extract even more subtle inference
> In one sentence: Mine the history of inferences and ask the system to
> come up with inference control patterns that may speed up future
> inferences.
>
> Of course the problem of learning inference control patterns is
> open-ended, ultimately you need an AGI to address that. But wait, we
> have one! That's what we're building!
>
>
> How does PLN, backward/forward inference has to do with that? Is it the
> same thing as unification and subsumsion?
control patterns.
As for unification and subsumption, I don't think it is the same. Both
the backward and forward chainers rely on unification to build
inferences. And there are subsumptions at different levels, like within
the type system, that is taken into account during unification, or via
atomese constructs like InheritanceLink, to build relationships between
concepts in the atomspace to reason on.
Nil
>
> Thanks in advance.
>
> --
> 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
> --
> 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
> <mailto:opencog+u...@googlegroups.com>.
> To post to this group, send email to ope...@googlegroups.com
> <mailto:ope...@googlegroups.com>.
> Visit this group at https://groups.google.com/group/opencog.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/opencog/39b254a0-c0b6-48b8-9051-dc1efdb16f9a%40googlegroups.com
> To view this discussion on the web visit
> <https://groups.google.com/d/msgid/opencog/39b254a0-c0b6-48b8-9051-dc1efdb16f9a%40googlegroups.com?utm_medium=email&utm_source=footer>.
Nil Geisweiller
Feb 27, 2017, 10:58:19 PM2/27/17
to ope...@googlegroups.com
On 02/26/2017 04:03 PM, Amirouche Boubekki wrote:
> I see that the wiki page is lacking in content or links regarding that
> matter http://wiki.opencog.org/w/Inference_Control
>
>
> In that article, “The rules of Logical <http://wiki.opencog.org/w/Logic>
> Inference <http://wiki.opencog.org/w/Inference> only tell you how to
> take a single step of inference”, what are those single steps?
I recommend looking at some step by step PLN demos like
https://github.com/opencog/opencog/tree/master/examples/pln/amusing-friend
the file amusing-friend-pm.scm contains a scheme/atomese step-by-step
run and the html file a more visual representation of it.
Nil
>
> TIA!
>
> --
> 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
> --
> 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
> <mailto:opencog+u...@googlegroups.com>.
> To post to this group, send email to ope...@googlegroups.com
> <mailto:ope...@googlegroups.com>.
> Visit this group at https://groups.google.com/group/opencog.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/opencog/0819d5d0-b3dc-42aa-8f5d-38a19967840d%40googlegroups.com
> To view this discussion on the web visit
> <https://groups.google.com/d/msgid/opencog/0819d5d0-b3dc-42aa-8f5d-38a19967840d%40googlegroups.com?utm_medium=email&utm_source=footer>.
Reply all
Reply to author
Forward
0 new messages