Language - Describing things
tomcat
I am creating this thread to hold anything related to language. NARS
input/output language is succinct and powerful, but may be not very
human readable. As I am learning nars, I have come across a couple of
issue so far that is hindering me.
I am trying to describe a situations in logic but not clear how it is
done in nars. Having looked at the syntax, I have been able to enter
some simple scenarios, but systems are a little more complex than
that. I am sure there is a way of describing what I want to do, but I
haven't figured it out. Does anybody know
1) How do you show like/dislike (i.e. Jo likes apples) using the
notation?
2) How do we express actions such as living, working, building etc in
this notation? i.e. “Jo works in town”, “bob lives in town”, "
Another example :
jo likes apples,
jo likes oranges,
jo does not like banana
grannysmith is-a-kind-0f apple
Goal - Find out if Jo likes grannysmith
does Jo like grannysmith?
3) How do we describe relationships such as "is" or "kind-of" (i.e. Jo
is human, Jo is a man)
4) How do we express goals without asking questions?
thanks
tomcat
Pei Wang
I have to say that once again I wonder if you have read any of the
materials I mentioned, because "is" or "kind-of" is usually
represented as Inheritance, which is discussed in almost every
publications on NARS. Other relations you mentioned, such as "like",
"living", "working", etc., are handled by the "ordinary relation"
mechanism introduced in NAL-4, which is also discussed in many of the
publications.
If you think you can start to work on concrete examples, and solve the
problems you run into by this kind of email discussion, I'm afraid it
is not going to work. Even though I (and the others, I believe) will
help, the issues are simply too complicated and time-consuming to be
solved in this way.
Pei
tomcat
I appreciate what you are saying, and furthermore, I did look at the
material I was pointed towards. However, I encounter the same
situation.
I am very interested to use this for my work, and looking for some
level of confidence that it can do what I want it to do. What I cant
do, is to spend month learning something which will not do the job.
This would be a total waste of effort and I am sure you appreciate
that.
The example I mentioned is fairly simple. The issue is translating it
to the NARS syntax. This seems to be the problem (from what I can
gather so far).
Having looked at the papers mentioned, the solution seems to be still
elusive and getting there is even more elusive as there doesn't seem
to be much help to get there. I really don't want to be a
mathematician to use NARS. Once I figure out the solution, no doubt I
will help writing some documentation to help others.
As the answer doesnt seem to be forthcoming, what would you suggest is
the best thing to do (apart from reading all the documentation in
NARS :-) )?
cheers
tomcat
cdrick
> What I cant
> do, is to spend month learning something which will not do the job.
> This would be a total waste of effort and I am sure you appreciate
> that.
not totally as you would have learned stuffs ;)
>
> The example I mentioned is fairly simple. The issue is translating it
> to the NARS syntax. This seems to be the problem (from what I can
> gather so far).
I'm porting NARS to squeak and I'm not familiar with logic expression
especially when it becomes complicated (see third links). But NARS is
really expressive.
Have a look at:
http://code.google.com/p/open-nars/wiki/InputOutputFormat
then
http://code.google.com/p/open-nars/wiki/SingleStepTestingCases
then
http://code.google.com/p/open-nars/wiki/ProceduralExamples
Also, for the overview: http://code.google.com/p/open-nars/wiki/Overview
and especially links at the end.
In my understanding, the basic idea behind term logic (which is
different from usual preidacte logic) is the simplest expression are
"is a"..so
bird --> animal mean bird are animals
John is a man would be John --> man. Actually, in this case, this is a
bit different as John is an instance (instance and properties are
dealt differently)
{John} --> man is the same as John {-- man {} denotes sets
I'm not (yet) interested in logic expression as I focus on control
right now and reasoning. I'll see later :)
But you should read and try to understand. I suggest you propose logic
sentences that allow to express your simple problem without confidence
and truth first, then I'm sure Pei will help you correct them, and
it'll help me too.
>
>> > Another example :
>> > jo likes apples,
I don't know yet how to represent the relation beetween Jo apples and like...
>> > jo likes oranges,
>> > jo does not like banana
>> > grannysmith is-a-kind-0f apple
>>
>> > Goal - Find out if Jo likes grannysmith
>> > does Jo like grannysmith?
>>
>> > 3) How do we describe relationships such as "is" or "kind-of" (i.e. Jo
>> > is human, Jo is a man)
This is the base :)
here I 'll write
Jo {-- human
Jo {-- man
man --> human
Cédrick
Pei Wang
I understand your concern --- we all have very limited time. However,
your example is not really "fairly simple" from AI's point of view
(which is very different from human's point of view). Otherwise it
should have been solved by the existing AI techniques.
I'm sorry I really don't have the time to guide you through your
example step by step --- that, with explanation, cannot be put into a
short email. See
http://code.google.com/p/open-nars/wiki/ProceduralExamples for how
complicate a "fairly simple example" can be. Your example won't be
shorter than that one, and even if I give you such a step-by-step
solution, you still won't understand it, given your current knowledge
on NARS.
Or you can wait for a few weeks for the new version, which will come
with more examples. After finishing the new version, I may be able to
work on your example if I have the time, though I cannot promise that
now.
Pei
tomcat
Thank you for your support. I appreciate that we are all time-poor and
will continue with my reading and try to get through this. I guess the
issue for me is how to translate my "human" example into "logical"
components which can be fed into the system. I will go through the
ProceduralExamples and see how I do.
I am keen to have something with me, when I meet my PhD supervisor end
of Aug as I plan to suggest nars as one of the tools for my work.It
would be great if I can manage to come up with some answer (even if
partial).
In the meantime, bear with me as I will come up with more questions
(which might be very obvious to you, but not so much to me).
BTW, can anybody suggest a readable book on "applied probability" and/
or "uncertain logic"?
cheers
tomcat
> your example is not really "fairly simple" from AI's point of view
> (which is very different from human's point of view). Otherwise it
> should have been solved by the existing AI techniques.
>
> I'm sorry I really don't have the time to guide you through your
> example step by step --- that, with explanation, cannot be put into a
cdrick
> Hi Cédrick , Hi Pei,
>
> Thank you for your support. I appreciate that we are all time-poor and
> will continue with my reading and try to get through this. I guess the
> issue for me is how to translate my "human" example into "logical"
> components which can be fed into the system. I will go through the
> ProceduralExamples and see how I do.
>
> I am keen to have something with me, when I meet my PhD supervisor end
> of Aug as I plan to suggest nars as one of the tools for my work.It
> would be great if I can manage to come up with some answer (even if
> partial).
>
> In the meantime, bear with me as I will come up with more questions
> (which might be very obvious to you, but not so much to me).
There not to me either but just try and propose, this is better than
starting explaining from scratch ;)
>
> BTW, can anybody suggest a readable book on "applied probability" and/
> or "uncertain logic"?
especially that in my opinion, there are few book out there that will
help you undestand NARS. It's my opinion but I find NARS a very
original approaches where I really agree on the basics. Said quickly,
traditionnal AI systems and NARS only common points are the vocabulary
use (inference, inference engine, rules, logic, term and so on). But
the implementation is really original. Maybe the more common aspect is
NARSESE the logic language and even there it's no the usual predicate
logic that is used so... you'd really better read Pei's document, and
if needed get informed about the reference he gives :)
Also, you must that I'm also learning so take my advices for what they worth ;)
I'm in the process of understanding "better" how nars works. I've
started this wiki page to links things together (but it's a personnal
memo that a collective one for now). I've taken what's written in the
overview, just reformatted, added stuff...
http://www.squeakside.com/seaside/pier/OpenNARS/Big+Pictures+NARS?
If you cannot get a copy of the book, I strongly suggest to read:
http://www.cogsci.indiana.edu/pub/wang.inheritance_nal.ps for
understanding the core logic.
Also you may want to read that to understand a bit how the system works:
http://nars.wang.googlepages.com/wang.computation.pdf
Good readings ;)
Cédrick
ps: and I forgot, try the software. It's available online. First try
the singlestepinference then the prcedural. Then you should be able to
propose and test expressions for your problem. Once there, you'll have
more precise help :)
Pei Wang
If you want something by late Aug, I should be able to find the time
after I finish the new version.
As Cédrick pointed out, NARS is very different from the other "applied
probability" and/or "uncertain logic" works. Even so, I do have some
reference listed in
http://nars.wang.googlepages.com/wang.AGI-Curriculum.html . For your
purpose, you may want to be familiar with the materials under B2, B4,
B5, C7, C8, C9, and C16.
Pei
TechTonics
Hi Cédrick , Hi Pei, Thank you for your support. I appreciate that we are all time-poor and will continue with my reading and try to get through this. I guess the issue for me is how to translate my "human" example into "logical" components which can be fed into the system. I will go through the ProceduralExamples and see how I do. I am keen to have something with me, when I meet my PhD supervisor end of Aug as I plan to suggest nars as one of the tools for my work.It would be great if I can manage to come up with some answer (even if partial). In the meantime, bear with me as I will come up with more questions (which might be very obvious to you, but not so much to me).
BTW, can anybody suggest a readable book on "applied probability" and/ or "uncertain logic"? cheers tomcat
Bayesian Non-Linear Modelling with Neural Networks
"A review paper giving a basic introduction to neural networks
and then describing Bayesian methods, with two case studies."
http://www.inference.phy.cam.ac.uk/mackay/PhD.html#PhD
http://www.cs.berkeley.edu/~jordan/papers/uai.ps.Z
http://www.cs.toronto.edu/~radford/sff.abstract.html
http://www.cs.toronto.edu/~radford/belief-net.abstract.html
"Connectionist learning procedures are presented for "sigmoid''
and "noisy-OR'' varieties of probabilistic belief networks. These
networks have previously been seen primarily as a means of
representing knowledge derived from experts. Here it is shown
that the "Gibbs sampling'' simulation procedure for such networks
can support maximum-likelihood learning from empirical data
through local gradient ascent. This learning procedure resembles
that used for "Boltzmann machines'', and like it, allows the use of
"hidden'' variables to model correlations between visible variables."
Is this what you meant? When I read your request I wondered
if a mentor had suggested to you that NARS was an appropriate
area to start at, when undertaking your PhD thesis, taking into
account your unique educational background.
Regards,
Stephen
TechTonics
Bayesian logic programs demonstrate some similarities
between these logic approaches at a more fundamental level.
[OTOH, Pei Wang speaks to Bayesian limitations.]
http://tech.groups.yahoo.com/group/ai-philosophy/message/6363
The above url is a post made by Pei Wang awhile ago in
Usenet where he recommends some reading resources.
Entropic musings,
Stephen
-- "The close and thoughtful observer more and more learns to recognize his limitations. He realizes that with the steady growth of knowledge more and more new problems keep on emerging." Johann Wolfgang Von Goethe : (1749 - 1832)
TechTonics
which did not appear on Pei Wang's reading list is
Russell,Norvig - AI - Artificial Intelligence
A Modern Approach - 2nd Edition
Chapter 14, Uncertainty
and
Chapter 15, Probabilistic Reasoning Systems
It should be in nearly all University libraries.
I finally realized that to grasp this stuff required
doing the exercises, reading papers helps only some.
tomcat
Thank you for all your support. It has given me a lot of choices now
and I will work my way through it. What I say is that, the more I read
about NARS the more I like it especially the underlying assumptions on
which the system is built. I am also beginning to understand why it is
different to other systems which I was planning to use.
cheers
tomcat
P.S. Sorry if I posted this twice.
> after I finish the new version.
>
> As Cédrick pointed out, NARS is very different from the other "applied
> probability" and/or "uncertain logic" works. Even so, I do have some