A distributed design ought to be language independent. Start by
deciding how the data and processing will be distributed and what
information the processors need to communicate. Then design a
communication protocol and figure out the hardware requirements. How
many processors? How much storage? How much network bandwidth?

2012/1/25 YKY (Yan King Yin, 甄景贤) <generic.intellige...@gmail.co m>:
I doubt it. Solving a specific instance is usually easier than solving
the general problem.
What about [link] ?
-- Matt Mahoney, mattmahone...@gmail.com

On Thu, Jan 26, 2012 at 5:22 AM, Russell Wallace
Yeah that's what I think, Genifer could use a distributed database for the
KB, but then the querying / fetching operations have to be custom-made, and
existing DB systems may not allow doing that, or it'd be a hassle to
implement.
Guess I'd go with the message-passing approach...

Alas, automatically distributing a serial program across multiple
machines is still an unsolved problem. For doing it by hand, Erlang is
said to be a good way to express message passing computation, though
bear in mind it's not designed for numbercrunching, so if your motive
for distributing the computation is speed, it's possible a different

I have figured out the basic algorithm for clustering the KB, and am now
working out the distributive version of it. It's not that difficult, but
I'm wondering if this can be done automatically by existing software? IE,
I write the non-distributive version of the program, and a software
platform will automatically turn it into an efficient distributive version?

Yeah, I would do that first. It will help clarify what else needs to be done.
I posted on OpenCog a list of things that I think AI ought to be able
to do. Something like this:
Language:
- Predict text with human-level performance.
- Answer questions in natural language.
- Predict user interest in email or web pages (relevant/spam).

Forecast for ~ next month or so...
1. The online book is badly in need of revision, but I didn't have the
time. I will add a to-do section to the online book, summarizing what is
to be revised, as a quick remedy to this problem.
2. I've just worked out more details of the inference speed-up algorithm

2011/12/27 YKY (Yan King Yin, 甄景贤) <generic.intellige...@gmail.co m>:
Probably so. Humans don't break down on input like "this sentence is
false". We just assign a truth value of 1/2.
The problem doesn't completely go away, because sometimes we need
truth values of 1 to think about math, logic, and programs. But I

In Godel's proof, he considered logics where a statement is either true or
false. I don't think his proof works if some statements can have
intermediate truth values.
Let's look at a formal statement of Godel incompleteness:
*Any effectively generated theory capable of expressing elementary
The definitions of "true", "provable", "complete", and "consistent" all

2011/12/25 YKY (Yan King Yin, 甄景贤) <generic.intellige...@gmail.co m>:
Godel incompleteness is better understood as the halting problem.
Fuzzy logic doesn't help.
Maybe you have seen the discussion on the OpenCog list about
distributing AtomSpace?

Godel incompleteness only affects binary logic. If the logic is 3-valued
or more (such as fuzzy), it may be able to escape Godel incompleteness.
We would be using a network of nodes to perform distributive and massively
parallel learning. It's just that the nodes would be using predicate logic
rather than propositional logic (equivalent to neural networks).

2011/12/18 YKY (Yan King Yin, 甄景贤) <generic.intellige...@gmail.co m>
Because logic is learned natural-language grammar. It is a set of
rules for manipulating symbols, like "if P then Q" = "Q or not P". We
can make up rules if we want, and teach them to others. But Godel
tells us we can't ever make up a complete and consistent set of rules.

*1. The ultimate universal logic*
The current situation is like this:
[image: ultimate-logic-1.png]
Both Genifer and Opencog (and others) probably have a logic that is capable
of near-human-level intelligence.
But currently all our AGI logics have flaws. For example, the inability to
deal with some paradoxes (eg Russell's and Curry's paradoxes). Notice that

I dunno. I kind of liked the synthesized voice. I mean, it's
appropriate for an AI, isn't it?
2011/12/11 YKY (Yan King Yin, 甄景贤) <generic.intellige...@gmail.co m>:

By the way, someone complained about the female computer voice, saying that
it's annoying. I used it because I tried narrating it myself but it came
out badly (I need more practice). I'm now finding an English-speaking
female student to narrate it, but she's having final exams now. Also, I
will eliminate the first-person perspective to avoid confusion.

2011/12/10 YKY (Yan King Yin, 甄景贤) <generic.intellige...@gmail.co m>:
They probably feel the same way about your ideas, so I wouldn't worry about it.
Microsoft only did that after Apple proved it would work. (And Apple
stole the ideas from Xerox). Nobody is going to steal any AI ideas
because Genifer and OpenCog are both a long long way from doing

The problem is that they don't have any ideas worth stealing -- we have
similar logical foundations, so perhaps I have already internalized Ben's
ideas from many years ago. One of their special idea is Atomspace, but I'm
still unsure if I want to use it.
The current situation is similar to Windows copying Apple's GUI idea in the

Interesting, especially taking into account that some real Russian
children can mix noun cases for less familiar constructs until the
secondary school :)

On 04/02/2012, A. T. Murray <mentifex@scn.org> wrote:
> 1. Fri.3.FEB.2012 -- An Astounding Idea
>
> For the Russian-thinking Dushka AI Mind, we have perhaps
> stumbled upon a way to avoid the hard-coding of noun paradigms
> and instead to let the Russian AI learn the inflected endings of
> Russian nouns from its own experience. For example, right now
> the Russian artificial intelligence (RuAi) fails to recognize
> the Psi concept #501 &#1041;&#1054;&#1043; in the following exchange.
>
> Human: &#1103; &#1091;&#1074;&#1072;&#1078;&#1072;&#1102;
> &#1073;&#1086;&#1075;&#1072;
> ("I honor God.")
> Robot: &#1058;&#1067;
> &#1059;&#1042;&#1040;&#1046;&#1040;&#1045;&#1064;&#1068;
> &#1041;&#1054;&#1043;&#1040;
> ("You honor God.")
>
> Robot: &#1063;&#1058;&#1054; &#1041;&#1054;&#1043;&#1040;
> &#1058;&#1040;&#1050;&#1054;&#1045;
> ("What is God?")
>
> The diagnostic display reveals that the software has almost
> recognized the word for God.
>
> 559. &#1041; 0 * 1 1 0
> 560. &#1054; 0 * 0 1 0
> 561. &#1043; 0 * 0 1 501
> 562. &#1040; 0 * 0 0 902
>
> Aha! Suddenly it becomes clear that two things are happening.
> The Psi concept #501 is indeed being recognized at first, but
> perhaps the provisional-recognition "prc" variable s not
> being set, and so AudInput calls NewConcet as if the AI were
> learning a new word instead of recognizing an old word.
>
> 2. Sat.4.FEB.2012 -- Learning Russian Like a Human Child
>
> Now in a very rough way we have trapped for "nazad1" in
> the AudRecog module so as to recognize a noun
> (&#1041;&#1054;&#1043;&#1040;)
> with one character of inflection added onto it. Because
> the noun was indeed recognized, the InStantiate "seqneed"
> mechanism tagged the noun in the "ruLexicon" with a
> "dba" of "4' to indicate a direct-object accusative case.
> In other words, the Russian AI learned a new noun-form
> as a human child would learn it, that is, from the
> speech patterns of another speaker of Russian.
>
> Mentifex
> --
> http://www.scn.org/~mentifex/Dushka.html
>
>
> -------------------------------------------
> AGI
> Archives: https://www.listbox.com/member/archive/303/=now
> RSS Feed: https://www.listbox.com/member/archive/rss/303/1649473-08437141
> Modify Your Subscription:
> https://www.listbox.com/member/?&amp;
> Powered by Listbox: http://www.listbox.com
>

Brief reminder, I am not trying to force any of you to evolve or
gradualize, just looking for your ideas on a principled, axiomatic (in
the loose sense) approach to AGI engineering. It would be wonderful if
your dDNA could build both a mind and a body, but mind-only ideas are
acceptable.By the way, body-only ideas are not irrelevant either, as
we've discussed previously with modular robotics, nanorobotics etc.
Cheers,

AT

On Sat, Feb 4, 2012 at 3:48 PM, Tim Tyler <tim@tt1.org> wrote:
>
> On 04/02/2012 07:47, Bob Mottram wrote:
>
> Although there are some similarities between culture and genetics, in my estimation trying to implement cultural evolution along the same lines as DNA evolution isn't going to work very well.� At best I think it would result in an incrementalist culture.
>
> As people such as Steven Pinker have pointed out, changes in culture are not strictly incremental but also include programatic features (leaps of logic or narrative elaboration).� This seems likely to be due to the generative nature of language, which can recruit brain regions and repurpose them as class 4 or Turing complete metasystems which can be distributed across minds in space and time.
>
>
> That's really the perspective you get if you only consider what happens
> between minds.� Idea reproduction and selection takes place inside
> minds as well.� What might look "from the outside" as a large leap is
> usually explicable in terms of a generate-and-test process going on
> *inside* the mind.
>
> Cultural evolution still exhibits gradualism.� The idea that it is hard
> to make large leaps through a search space without testing and wind
> up with something viable is a universal one - not confined to organic
> evolution.
>
> Yes, non-gradualistic random search strategies exist, and *can* be used
> to make evolutionary proghress - but random search strategies are
> hopeless and they only work at all for trivially small problems.� Once
> you have a large complex system, you are back to gradualism again.
>
> It's also true that cultural evolution can use forecasting and extrapolation
> to attempt larger leaps through search space this is possible without
> such tools.� However, those larger leaps are still usually pretty small.
>
> I think the bottom line is that gradualism still rules.� Gradualism
> is one of the key hallmarks of evolutionary processes.� Evolution
> without gradualism would be indistinguishable from magic.� So:
> an "incrementalist" theory of culture is *precisely* what is needed.
> --
> __________
> �|im |yler� http://timtyler.org/ï¿&frac12; tim@tt1lock.org� Remove lock to reply.
>
> AGI | Archives | Modify Your Subscription

1. Fri.3.FEB.2012 -- An Astounding Idea

For the Russian-thinking Dushka AI Mind, we have perhaps
stumbled upon a way to avoid the hard-coding of noun paradigms
and instead to let the Russian AI learn the inflected endings of
Russian nouns from its own experience. For example, right now
the Russian artificial intelligence (RuAi) fails to recognize
the Psi concept #501 &#1041;&#1054;&#1043; in the following exchange.

Human: &#1103; &#1091;&#1074;&#1072;&#1078;&#1072;&#1102;
&#1073;&#1086;&#1075;&#1072;
("I honor God.")
Robot: &#1058;&#1067;
&#1059;&#1042;&#1040;&#1046;&#1040;&#1045;&#1064;&#1068;
&#1041;&#1054;&#1043;&#1040;
("You honor God.")

Robot: &#1063;&#1058;&#1054; &#1041;&#1054;&#1043;&#1040;
&#1058;&#1040;&#1050;&#1054;&#1045;
("What is God?")

The diagnostic display reveals that the software has almost
recognized the word for God.

559. &#1041; 0 * 1 1 0
560. &#1054; 0 * 0 1 0
561. &#1043; 0 * 0 1 501
562. &#1040; 0 * 0 0 902

Aha! Suddenly it becomes clear that two things are happening.
The Psi concept #501 is indeed being recognized at first, but
perhaps the provisional-recognition "prc" variable s not
being set, and so AudInput calls NewConcet as if the AI were
learning a new word instead of recognizing an old word.

2. Sat.4.FEB.2012 -- Learning Russian Like a Human Child

Now in a very rough way we have trapped for "nazad1" in
the AudRecog module so as to recognize a noun
(&#1041;&#1054;&#1043;&#1040;)
with one character of inflection added onto it. Because
the noun was indeed recognized, the InStantiate "seqneed"
mechanism tagged the noun in the "ruLexicon" with a
"dba" of "4' to indicate a direct-object accusative case.
In other words, the Russian AI learned a new noun-form
as a human child would learn it, that is, from the
speech patterns of another speaker of Russian.

Mentifex
http://www.scn.org/~mentifex/Dushka.html