Linas

Some clarity and thoughts...

________________________________
From: Linas Vepstas <[email protected]>
Sent: Tuesday, 19 February 2019 9:06 AM
To: AGI
Subject: Re: [agi] Some thoughts about Symbols and Symbol Nets

Hi Robert,

I'm waiting for unit tests to pass, and it's like watching paint dry.  So I 
write spurious emails as I wait. Spurious response follows.

On Mon, Feb 18, 2019 at 11:45 PM Nanograte Knowledge Technologies 
<[email protected]<mailto:[email protected]>> wrote:
Linas, Mike and Jim

I find this to be a most-interesting conversation. Primarily, because it 
suggests that the development of AGI may not only be challenged by the 
development of competent theory, but also by programming capabilities to put 
the theory into practice.

Yes. The people who know some of the theory usually don't know how to program, 
and v.v. and getting both to meet up is hard. That, plus the fact that there 
are 1001 theories, and there is very little (almost no) consensus about the 
right approach.


Evolving such an architecture then, should desired outcomes be for an AGI 
entity to achieve self-theory and self-programming?

Uh, yes? This question seems to be phrased awkwardly.  AGI isn't some thing 
that is just like a self-driving car, but only a tiny bit smarter....

As a human, I have a self-theory. Parts of it are excellent: I really do know 
where my hands are, and what they are doing. Parts of it are terrible: I really 
don't know much about the vascularization of my lower legs, or how a 
black-and-blue spot appeared there.   But hey, self-driving cars have a very 
good idea of what is on the road in front of them, and have no idea at all 
about the chemistry of rubber.  Self-driving cars have a self-model. Which is 
maybe less than a self-theory?

>> What I meant by self-theory was the ability to form a hypothesis and evolve 
>> a theory and test for such a hypothesis, all the while spinning off the 
>> learning as computational functions of programming. I think I have a 
>> very-good idea of what from an agi-service would take. Agreed, it;s not a 
>> smart machine. In my vision, it's a species.

As a human, do I engage in self-programming? Sure. I make new-years resolutions 
every day. I even keep some of them.  Self-driving cars don't, for obvious 
liability reasons.

>> Self-programming would be an ability to code programs on demand, or via 
>> threshold triggers, and so on. Perhaps, as an evolutionary step.

Do I expect an AGI to be equally self-aware, and equally in self-control? Yes.

In its most-simplistic from, a symbol is but a suitable abstraction of a 
greater reality, similarly to how a symbol of a red-heart might be an 
abstraction of a sentient being. Concept? Context? Meaning? Transaction.

Who, or what decides what the symbolic world should look like and its 
meaningfulness?

Is this a rhetorical question?

>> Not really

In the human sphere, the poets, painters, dancers and mathematicians decide 
what the symbolic world should look like, and work very hard to capture its 
meaning in poetry, paintings, movements and equations. I expect AGI to struggle 
with the same issues.

>> Too vague, too generalist. I think symbolism emerges from diversity, or more 
>> accurately, programs of diversification.

But if you mean "who decides whether symbol #4589342472934 should even exist, 
or what it means?" ... heck I dunno. Some algorithm, the same algorithm that 
decided that symbol #11316372398 is meaningful.

>> The designer decides the system of symbolism. The agi entity has the 
>> existential prerogative to symbolize.

In between the symbols generated by algorithms and the symbols generated by 
poets are the symbols generated by insurance companies, zoning commissions, 
safety regulators and lawyers. These symbols have names like "penal code 
#234241;para.4.b-addendum 62" and are almost as boring to read as reading 
software.

Do I expect the middle layers of AGI to look like something an insurance 
company or zoning commission might write? Sure, why not?  I expect a Chinese 
Room hard at work, in the middle layers of the AGI.

>> Linear and alinear systems contribute to holistic systems. I think it was 
>> Checkland who made the point most clear. The discussion flux between linear 
>> and alinear systems, and considers something in between. Why nto simply 
>> extract the fractal value of an instance of linear and alinear contexts and 
>> symbolize that in an agi, as a knowledge node, and so on. That was my point 
>> on fractals, the search for a pattern of meta superpositions (the 
>> paradoxes), which may well hold a key to finite-infinity within systems. 
>> Gell-Mann provided theoretical essence in his discussions on scalability (in 
>> the sense of boundary-independent form - my words) and his notion of 
>> intermediate.

Having said all this makes me realize that there is a lot more accepted theory 
than we could imagine. What is required then are the appropriate frameworks to 
put those theories to work in context of an agi service.


The global state of social evolution may cause terrible confusion in any 
learning entity. The learning objectives should be specific, not generalized. 
Isn't learning incorrectly worse than not learning at all?

That is certainly a hotly debated question in Brexit and Trump discussions.  
More accurately, the problem is one of not having an accurate understanding of 
the world, and being unable to get one.

>> the clamor for truth vs big data?

its not so much a case of "learning badly", but one of hallucinating: 
hallucinating that things will be better, or worse, if England leaves the EU, 
etc. The combined sensory-system+political-brains seem to be incapable of 
figuring this out.

>> general relativity - self interest drives what we see, and learn.

I expect AGI to hallucinate, too. Just not like us.  Actually, I expect AGI to 
be schizophrenic, psychopathic, and a zilllion other rather negative things 
that are existentially dangerous to humans. That's the tricky part, the part 
that is unpleasant to face.

>> I think agi, as a concept, already is.

I think, there should be a general agi-architecture, replete with the capacity 
to develop and function within a generic world view. Furthermore, I think the 
real value would be derived from specialized AGI. Maybe beyond that, an AGI 
architecture would - in future - morph via its own social networking and 
inherent capabilities to become more than the sum of its parts.

Well, you used lower-case-agi and Upper-Case-AGI there.  There's no such thing 
as an  Upper-Case-AGI architecture -- claiming to have one is like claiming to 
have the blueprints for a rocket-ship to another galaxy.

>> An interesting observation. I have no idea why the case-sensitivity. Perhaps 
>> it does not really matter at all. How did you come so far on your journey to 
>> another galaxy without a blueprint? Surely, it must all be just luck? No, 
>> it's the result of years of driving passion and vision. The blueprint must 
>> exist in your collective minds.

However,  lower-case-agi-- well, that is more like mountain climbing: you try 
to go one way, until you find that you can't so you try another way, until you 
can.  You explore, looking for routes to get from here to there. So, if 
Upper-Case-AGI is a mountain peak, then we are at the foothills of the 
Himilayas, fumbling and tripping and getting exhausted.  Explorer X says that 
the deep-learning route is promising; explorer Y disagrees.  Everybody's got a 
base-camp, and some are busier and livelier than others.


To do so, would take a lot more than intersections. I agree with the statements 
made about binary/vector theory, but it seems obvious to me that this would not 
be sufficient for this task. You implied fractals.

Heh. Careful with the analogies, there. Fractals are manifestations of 
shift-spaces. Which are infinite-dimensional vector spaces. The last three 
decades have exposed a deep and abstract mathematical theory for "fractals".  
That theory is ... interesting, but has rather very little to do with AGI.  
Like pretty much nothing at all

>> every fractal has a distinct boundary. it might be a symbolic black hole to 
>> some, but I would argue how it's not an "infinite-dimensional vector space". 
>>  A pure object is a fractal too, as is a context. Not in a physical sense, 
>> but in an informational sense, where Physics behave in the role of the 
>> carrier and information in the role of the content being carried. I contend 
>> how fractals has very much everything to do with AGI.

We, as researchers, may not all be using the same terminology, but the concepts 
you are discussing in your latest response to Rob are not foreign to my mind. 
Perhaps, there are different routes via which to discover AGI, and as evidenced 
within Ben Goertzel's treatise on world religions, there exist different paths 
for different people towards achieving AGI enlightenment.

What if we had at our disposal a common language to start putting these paths 
together within a single AGI frame? Imagine, consensus.

PS: I'd think the answer to your compound relational question is: 27. Even so,  
if a rule was being enforced where a singular, existential method of 
association between X and Y entities were being deployed. Hierarchy itself is 
linear. It can flow over an alinear system.


-- linas.

To my mind, that would be the only way to proceed. As such, I think the primary 
issue remains a design issue.

Robert Benjamin

________________________________
From: Linas Vepstas <[email protected]<mailto:[email protected]>>
Sent: Monday, 18 February 2019 10:36 PM
To: AGI
Subject: Re: [agi] Some thoughts about Symbols and Symbol Nets



On Mon, Feb 18, 2019 at 1:17 PM Mike Archbold 
<[email protected]<mailto:[email protected]>> wrote:
I'm not sure I completely follow your point, but I sort of get it.

I tend to think of symbols as one type of the "AI stuff" a computer
uses to think with -- the other main type of "AI stuff" being neural
networks. These have analogies to the "mind stuff" we use to think
with.

Symbol systems and neural-net systems can be seen to be variants of the same 
thing; two sides of the same coin. I posted an earlier thread on this. There's 
a 50-page long PDF with math, here: 
https://github.com/opencog/opencog/raw/master/opencog/nlp/learn/learn-lang-diary/skippy.pdf

roughly: both form networks. They differ primarily in how they represent the 
networks, and how they assign weights to network connections (and how they 
update weights on network connections).

On their own, symbols don't mean anything, of course, and inherently
don't contain "understanding" in any definition of understanding.

Is there a broad theory of symbols? We kind of proceed with loose
definitions. I remember reading the Newell and Simon works, and they
say AI strictly in terms of symbols and LISP (as I recall anyway).

Yes. The "broad theory of symbols" is called "model theory" by mathematicians. 
It's highly technical and arcane. It's most prominent distinguishing feature as 
that everything is binary:  it is or it ain't. Something is true, or false.  A 
formula takes values, or there is no such formula. A relation binds two things 
together, or there is no relation.  There's no blurry middle-ground.

So, conventionally, networks of symbols, and the relations between them, and 
the formulas transforming them -- these form a network, a graph, and everything 
on that network/graph is a zero or a one -- an edge exists between two nodes, 
or it doesn't.

The obvious generalization is to make these fractional, to assign weights. 
Neural nets do this. But neural nets do something else, that they probably 
should not: they jam everything into vectors (or tensors) This is kind-of OK, 
because the algebra of a graph is a lot like the algebra of a vector space, and 
the confusion between the two is an excusable mistake: it takes some 
sophistication to realize that they are only similar, but not the same.

I claim: fix both these things, and you've got a winner.  Use symbolic systems, 
but use fractional values, not 0/1 relations.  Find a good way of updating the 
weights. So, deep-learning is a very effective weight-update algorithm. But 
there are other ways of updating weights too (that are probably just as good or 
better.  Next, clarify the vector-space-vs-graph-algebra issue, and then you 
can clearly articulate how to update weights on symbolic systems, as well.

(Quickly explained: probabilities are not rotationally-symmetric under the 
rotation group SO(N) whereas most neural-net vectors are: this is the spot 
where deep-learning "gets it wrong": it incorrectly mixes gibbs training 
functions with rotational symmetry.)

So Jim is right: discarding symbolic systems in favor of neural nets is a 
mistake; the path forward is at the intersection of the two: a net of symbols, 
a net with weights, a net with gradient-descent properties, a net with 
probabilities and probability update formulas.

-- Linas


On 2/18/19, Jim Bromer <[email protected]<mailto:[email protected]>> wrote:
> Since I realized that the discrete vs weighted arguments are passe I
> decided that thinking about symbol nets might be a better direction for me,
>
> 1. A symbol may be an abstracted 'image' of a (relatively) lower level
> object or system.
>   An image may consist of a feature of the referent, it may be an icon of
> the referent or it may be a compressed form of the referent.
> 2. A symbol may be more like a 'label' for some object or system.
> 3. A generalization may be represented as an image of what is being
> generalized but it also may be more of a label.
> 4. An 'image', as I am using the term, may be derived from a part or
> feature of an object or from a part of a system but it may be used to refer
> to the object or system.
> 5. An image or label may be used to represent a greater system. A system
> may take on different appearances from different vantage points, and
> analogously, some features of interest may be relevant in one context but
> not from another context. A symbol may be correlated with some other
> 'object' and may stand as a referent to it in some contexts.
>
> So, while some symbols may be applied to or projected onto a 'lower' corpus
> of data, others would need to use an image to project onto the data field.
> I use the term, 'lower' somewhat ambiguously, because I think it is useful
> to symbolize a system of symbols so a 'higher' abstraction of a system
> might also be used at the same level. And it seems that a label would have
> to be associated with some images if it was to be projected against the
> data.
>
> One other thing. This idea of projecting a symbol image onto some data, in
> order to compare the image with some features of the data, seems like it
> has fallen out of favor with the advancements of dlnns and other kinds of
> neural nets. Projection seems like such a fundamental process that I cannot
> see why it should be discarded just because it would be relatively slow
> when used with symbol nets. And, there are exceptions, GPUs, for example,
> love projecting one image onto another.
> Jim Bromer


--
cassette tapes - analog TV - film cameras - you


--
cassette tapes - analog TV - film cameras - you
Artificial General Intelligence List<https://agi.topicbox.com/latest> / AGI / 
see discussions<https://agi.topicbox.com/groups/agi> + 
participants<https://agi.topicbox.com/groups/agi/members> + delivery 
options<https://agi.topicbox.com/groups/agi/subscription> 
Permalink<https://agi.topicbox.com/groups/agi/Tcc0e554e7141c02f-M94a2745199b6ca0b1a6a8f0e>

------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/Tcc0e554e7141c02f-M6f5b4a8d2ee3f5bdf718c233
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to