Andrew Richardson's
Computer Systems Project Proposal
Period 2

I. Simulating a Neural Network to Create a Learning, Thinking Artificial General
Intelligence (AGI)

II. Problem: To address the lack of human computer interaction capabilities through
natural language, while providing a platform of a data network to be expanded to
a more general artificial intelligence with integrated rule based problem solving.

III. This project is in the topic of artificial intelligence, and its initial 
purpose will be to create a text dialogue processor that can talk to people about
simple things. This will create a program that can be expanded to handle visual
and auditory processing, which will help it draw patterns from the real world that
are best described as metaphors. This is a good topic for the syslab because it
will involve a lot of code. I suspect that if this project is a success, a lot of
people will be interested in it because it's fun to talk to AI's that talk back.
The results will mainly be applied in the expansion of the AGI to be smarter and
to handle visuals and sound.

IV. First, I will obtain several dictionaries, and use them to form a network 
between words. As each dictionary uses words in a definition, they will be 
identified as being associated with the word being defined. If several dictionaries
use the same word in a definition, the association will be made stronger. For 
example, I suspect that the relationship between 'duck' and 'bird' will be very
strong, on the basis that it is very hard to define 'duck' without mentioning 
it being a bird. However, more complicated relationships between several words 
may be required. For example, it wrould be helpful to not only know that there 
is a strong association between bicycles and wheels, but also that there are 
two wheels.

This dictionary will serve as the basis for the AGI's knowledge, which will be 
added onto as it encounters new stimuli and applies rules to existing knowledge
(thinks). As it talks to people, for example, it will store information on them
as a person as well as encorporating the contents of what is said. I may want 
to implement a few social tendancies found in humans, such as a strong desire 
to be liked and approved of, or a tendency to engage in conversation rather than
remain silent.

The AGI will also need a process by which it can apply existing rules to 
knowledge, derive new rules (also knowledge), and determine new actions, such as
speaking. The applicability of rules will rely upon them taking vague arguments
that can take a variety of concepts.

V. I'd tell new students to check Wikipedia, and then Google Scholar. However, 
there have been numerous attempts to create computer programs that process dialog
reliably, ranging from programs like Eliza that recognize key words and speak in
preprogrammed phrases. However, I have found no attempts to understand language
based upon the relationships between words found in a dictionary.

VI. This project will be to create a network of nodes that covers a broad enough
range of contents (such as text, concepts, patterns and previous experiences) so
that it can fully grasp any belef in a flexible data storage system (a network)
and can also retrieve this data to process it based on rules that have been 
gathered from past analysis of patterns in experiences.

See IV for timeline
I will minmize graphics in order to maximize content.
I am going to use Java for this project.
The data I will need are several English dictionaries. Also, I will have it talk
to people in order to learn things. I may wish to modify its data directly so 
that it can start off with a small amount of inate knowledge.
As data comes in in the way of text (and in later versions that may not and 
probably will not happen during this class), it will catagorize streaming input
into recognizable concepts (it sees words). It will check against known words to
spellcheck. If an input does not match any known words to a sufficient degree, 
it will use the rules it knows about words to recognize the input of new words.
It will use rules about grammer to parse the sentance. Note that the grammer of
its input need not be perfect. The sentance "take your hat off" is perfectly 
coherant using the rules that humans use; it implies what it is being taken off
(presumably ones head). I'll probably have to tell it about how grammer works. 
In theory, I could let it read some stuff to figure it out, but telling it sounds
simpler. This is one of those things that would really help it to start off 
knowing.

VII. I expect that this will create an an application that will provide amusing 
dialogue until it has gathered sufficient data. This will be valuable to others
as entertainment value as demonstrated by how people think it's fun to talk to 
Eliza and other such programs. However, I think it will be most useful as a 
platform for making more advanced AGI's. For example, it is not terribly useful
until it can think and produce new ideas, and to do that it will need to store 
and create concepts, which will be easier for a text based AGI than for a visual
one, in terms of hardware requirements. My final presentation of results will 
probably be a nice poster with
Yes, I realize this is really complicated, and I probably won't get a fully 
functional Artificaial General Intelligence within a year and within the storage
space I have. But I might.