Cogito Patris

Random Thoughts for Random People

The Un-Theory

This is an unusual situation for me.  Usually when I come up with these wild theories I try to discount them before I commit them to writing.  This is one that I was actually able to discount.

The Theory:

I thought it would be possible to create an artificial intelligence with all of human knowledge.  Not in some unknown future, but right now.  If you think about it, human thoughts can be broken down in to words.  Dictionary.com has every word in the dictionary (redundant?).  So, first you can teach the AI to look up words in the dictionary.  So, for a word like “Button” the AI would know that it was either an interface between a piece of hardware and a human operator (keyboard buttons or a power button) or a piece of solid material, usually in the shape of a circle, that is used to hold two pieces of a garment together.

The next part of my AI was to link the words to a category.  So it would need to know if a particular word is a verb, noun, pronoun etc.  That is also in the dictionary.  So it would know that a “Button” is a Noun.

My next step was to give “meaning” to the words. 

With nouns it seemed pretty easy.  Have the AI search on Google for images related to the noun and link the two. So now the AI would know that not only know what a button does through the dictionary, but what a button actually looks like.

For Verbs it would be a tad difficult.  But Google has a video search and since a verb is simply an action a video search should accomplish the same for verbs that an image search would for nouns.

Pronouns and other such “shortcuts” could be handled with sentence structure analysis that any grade schooler can do (though many adults can  not).

This would need to be a self updating program.  Every time it “learned” a new word it would need to add it to it’s own database and/or add a programing module to call.  So it would need to know how to write the code that it is made of.  This isn’t as difficult as it sounds.  I have written many programs that generate code on the fly.  Object-Oriented programming practices rely on this functionality.

Finally, have the AI respond to a question appropriately using the above information.  Though this is more difficult than it may sound, it has already been done in a limited way with Oliverbot

The Discounting of the Wild Theory

The problem with this approach is apparent if you spend some time with Oliverbot.  The system is powerful in it’s simplicity, but it is limited by understanding.  A computer ultimately understands only what it has been told.  Having it search for information over the Internet doesn’t really help because although it may know the definition of a word, and may know what it looks like the concept is completely lost on it.  This would create an artificial intelligence with a massive memory and no understanding.  While writing this I realize that what I have described above is very much what Google actually does.

I still don’t know how my son knows what a “button” is.  How did he make that leap between the specific instances of a button and the general concept?  In many ways the fact that he is Autistic means that he has some trouble with this leap.  But maybe the study of Autism will help to determine how the human brain develops “concepts”.  So maybe this theory isn’t as wild as I think.

Advertisements

April 11, 2007 - Posted by | Wild Theory

No comments yet.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: