For a while I'm aware that from semantic tables like these for boolean stuff

¬ T -> F
¬ F -> T

T & T -> T
T & F -> F
F & T -> F
F & F -> F

T | T -> T
T | F -> T
F | T -> T
F | F -> F

you could by total induction completely automatically induce axiom rules
like

¬¬ x -> x
x & y -> ¬(¬x | ¬y)

[link]
[link]

What is NuPIC?

NuPIC, the *Numenta Platform for Intelligent Computing*, comprises a set of
learning algorithms that were first described in a white
paper<[link]>

Ivan,

Thanks for showing ACE, I've been aware of it, but never tried a demo.

Natural language has some features that require "abductive" interpretations
(abductive meaning "finding the best explanation"). For example, how we
resolve the pronoun in "John tells his son the story of his life". Or how

What about OpenCyc and WordNet?

But I think you would be better off if you could figure out the
algorithm for learning natural language and training it on plain text.

-- Matt Mahoney, mattmahone...@gmail.com

I think that starting set should be rich enough to attract enough students,
so I'll probably do a year or two on filling data before attracting
students to the site.

I hope that many free already coded ontologies on web would help me, so if
U know about some quality resources, please let me know.

I think you might be asking the crowd to provide value to you without
providing much to them in return.

However, if your application wrote papers for those students from a
meager dialog with them for things like topic, depth of research,
etc... then you might have something people will use - not because

When NL is inputed, the program can ask questions on ambigous problem, i.e.
some dialogbox can pop up. The more rules are entered, the less dialogs pop
out.

Also NL is live, it changes over time. Then some favors of NL can be
optioned on ambiguity input.

Nevertheless, the most of definitions should be inputed from crowdsource,

Maybe you can add common sense rules to help you parse, like:

Sausage is a pizza topping.
Chopsticks are eating utensils.
Bob is a person.

But how many rules do you need to add? This was the approach used by
Cyc. But they had no idea how many rules would be needed. They have
been adding rules for 29 years and still have not solved the problem.

I guess that would be problems I will have to deal with.

Maybe some concrete verbs bind to subject, other to object. There are also
problems of ambiguity. This grammar book I've mentioned has some answers,
it is pretty thick and has 500 pages. A century of reading is comming.

Also conditional parsing rules will be in my way. For example, if first

I mean that translating natural language to *any* formal language
requires that you have solved the natural language problem first. For
example, who does "he" refer to in the following sentences?

Jim punched Bob because he cheated.
Jim punched Bob because he was mad.

At what point do you solve the problem, during the translation or

NL definition should look like this, but a lot more complicated:

Sentence (
Subject -> { I, You, He, She, ...},
Predicate -> {am, want, pick, ...},
Object -> {apple, apricot, food, ...}
)

Now we can parse sentences this way:

Sentence(~I pick food~);

and parser distributes words onto properties Subject, Predicate and Object.

I plan to translate NL to Synth (a programming language I developed for AI
purposes - some sort of
BNF<[link]>language
enhanced with EventActions). When it is translated by universal
parser, the rest is easy, it is just about running queries and transforming

How do you plan to translate natural language into ACE? Translating
natural languages to formal languages is an advanced skill in humans
that happens only after they learn natural language. If your
translator already knows natural language, then you have solved the
problem.

--
-- Matt Mahoney, mattmahone...@gmail.com

Here is the proof that AI exists already, an "ACE<[link]>"
reasoner: *[link]<[link]>
(Click program->load example to play with the beast :))
Right now I'm onto *intelligent encyclopedia* development.

Nirgal @GSS:
Is there place for evolution or mutation in what you describe ?
I'm not sure...

Me @GSS:
I think that generalization algorithm is still. The catch is that function
has id, parameters and result and they can be mutually combined to match
data. And here, functions can have whatever parameters and whatever result.

Nirgal @Global Survival Group:
[link])

If this is what you have in mind, an AI would not be able to do a better
job than a system that enhances and collects and interlink our individuals
intelligences.

Me @Global Survival Group:
@Nirgal
Something is moving in my mind in direction that in nature, rows from many

I've just got a answer to my question.
How much bytes takes an AI behavior algorithm? Let's say 1000 bytes. It is
like we expect from error to line up 1000 bytes of unknown AI code in a
sequence on error and to continue working. I estimate probability of 2 to
the power of 1000. Freaking zillion and more.

Nirgal @GSS group

Maybe you would feel better by actually feeling you are part of a ecosystem
through your body and your actions ?
By eating plants you've watched grow and have taken care of, and by
planting much more than what you eat, you would feel of service of the
planet, while taking from her what you need to support her even more ?

You may be both relieved and disappointed to learn that what you propose
won't actually work. Sure, mathematically speaking search algorithms to
which you refer could do all sorts of things... if you had an infinitely
powerful computer to run them on. But if you try to use them in real life,
the heat death of the universe will interrupt your program long before it

@Matt
godlike i'm still afraid of.
scientific? I'd damn could!
The question is: is it safe enough?

2013/4/5 Matt Mahoney <mattmahone...@gmail.com>