How To Write an AI

If our goal is to write program, that interprets naturals language and the real world, then we should provide an example how to interpret it. For example, we can take a child's book with a simple story, then provide just enough framework for the program to interpret the text: what each book's character does, what he/she wears, how he/she relates to other characters, what he can and cannot do. All so that the program would be able to build mental model, draw characters, describe them, reflect on itself and simulate the reality of the book, the actions of the book's characters and causal relationships between the worlds events.

NO EXCEPTIONS

Yes but are you trying to make the computer understand, or are you just getting it to parrot?. How hard is it to build responses? getting a computer to essentially think it's own concepts up and form opinions, that's the hard part.

Also it will mean the death of CAPTCHA.

the issue with this is that building that flamework is quite complex

Have your read your AIMA today?

>>3
Yes but are you trying to make the computer understand, or are you just getting it to parrot?
Can you please elaborate on what you mean by "understand"? Are you sure, you really "understand" every book you read? Do you "understand" it as was intended by author? How do you know what was the authors intention. The reality is so that you can never built perfect understanding of texts written in natural language. Still you can interpret them with some margin of error, given sufficiently flexible data format and reasoning framework.

the issue with this is that building that flamework is quite complex
It is not. Something like sparse voxel octrees would already do most of the work on the data structure side. All you have to do would be populating the structure by the interpretation of a text book, then providing some way to do multidimensional searches in parallel.

getting a computer to essentially think it's own concepts up and form opinions, that's the hard part.
I hardly doubt you built anything original in your life. All you do is parroting reality.

>>6
Octrees don't provide enough flexibility to represent real world data. We need hextrees...

"Hextree (plural hextrees) - a treelike data structure each of whose nodes has up to sixteen children. It can be used to partition a four-dimensional space by recursively subdividing it."

>>7
Also, fourth dimension introduces relatively unresearched change to pathfinding algorithms, because our search tree now changes with time. I.e. a door, which is open at time 0, maybe closed at time 1, so even a simple A-Star pathfinder suddenly gains ability predict movements of world entities.

>>7
HAXTREES

Why not icosikaitera-trees, with 24 nodes? Not only does it look Japanese, which, believe me, is important, but could divide a multiverse into searchable areas.

>>9
you don't need multiverse for practical purposes.

A computer cannot understand language with out being embodied. How can a computer understand bodily metaphors like "close" or "in" when it has no spatial awareness?

>>11
Bullshit, prove it.

>>12
Octree provides space awareness. Hextree provides temporal awareness. And the terms like "close" and "in" are ambiguous without precise context. Earth is "close" to Sun, yet it is 149,600,000 km away.

>>13
Several worlds could be hosted inside of a single world, just like you can use an emulator to emulate several virtual computers on your real machine. That way you can run Windows and Linux programs on top of MacOSX.

>>14
I.e. an AI-geared octree memory should provide only relative sizes, without precise metric units.

http://www.asmcommunity.net/forums/topic/?id=9280

This mapping system would be applied to objects in the world. These nodes would represent a system of lines forming shortest routes, in turn these could be applied to AI for pathfinding, or simply to find the nearest objects for weighted need calculus.

The original idea from my youth was to keep three axial binary trees,but a hextree puts it all in one neat neural network.

For any object in the 3d world, we can decide which is the next object in the X+,X-,Y+,Y-,Z+ and Z- directions.
That means, on its right,left,above,below,in front and behind.
Only one object can exist in one place

>>15
MacOSX
eat shit and die

>>18
Linux is dead.

>>14
Depends if a certain ambiguity is a bonus.

I'd think good idea, if the aim is to parse the language in a 'natural' sense, would be a system capable of doing something based on a combination of this sort of thing: https://en.wikipedia.org/wiki/Semantic_primes
With some way of measuring a distance and relation from other symbols in terms of these primitives you'd end up with less of a dimensional tree and more of a weighted and typed network.

>>19
that's why I installed BSD/Gentoo

>>20
Yet sparse dimensional tree should be enough to get the system working and it and can efficiently store huge amounts of real world data and provide access to it at any level (See http://www.youtube.com/watch?v=VYfBrNOi9VM)

And if you want to say that two things are close, just place them in nearby cells. Sparse trees great distances between the objects without wasting memory.

There is already a similar data-structure approach in AI:
http://en.wikipedia.org/wiki/Hierarchical_temporal_memory

Another similar concept for representing complex objects: http://en.wikipedia.org/wiki/Bounding_volume_hierarchy

http://www.youtube.com/watch?feature=player_detailpage&v=QLDlBRAlSAM#t=468