A.I.M.E.

 

Watching the 4th part of the Godel, Escher, Bach M.I.T. lectures (http://ocw.mit.edu/OcwWeb/hs/geb/VideoLectures/index.htm) I realised how similar my own interests are to the main stream researchers. The issue of Algorithmic Complexity was mentioned – this is exactly one line of what I’ve been questioning. It made me think that I should have continued with the A.I.M.E. project in 1996.

Artificial Intelligence Mechanism Emulation – was a development on a much older and more naive project in 1986 which sort to create a speech engine. It simply looked up each word and if it didn’t know it asked you to select what type of word it was (noun, verb etc). It was then programmed to know how sentences were organised and so it could deconstruct input sentences and reconstruct sentences based on grammar. I never got very far and have never got very far because I have always sort to avoid solutions which require more inbuilt programming than one gets out of the system. There is no glory in my mind in getting out of a machine what you programmed in. The Mandelbrot is fascinating because it seems to provide a lot more than is programmed in. This in the realm of A.I. is the goal…. and I decided along time ago it was impossible – you get out what you put in.

A.I.M.E. was to solve a different problem… an engine that could generate what I can now call algorithmic compression.

An example of the problem:

A computer is given data structures and functions to deal with points i.e. (x,y) pairs of data.

Given a series of points the problem was to generate an algorithm which could generate a data structures to deal with lines i.e. y = mx + c, with start and end point. Thus all the points could be approximated to special cases of a single line structure. That is true compression.

Recursively performing the algorithm to reduce the complexity of the mind and reduce its entropy was the model for thought and I hoped for consciousness also.

What I thought would such an algorithm do when it had its own output fed into its input? It would try to abstract the data into simpler more general forms and this would lead it to an ultimate form which would correspond to the nature of the loop itself and this self-consciousness.

Intriguing idea. I set out to program A.I.M.E. It would have a 2D grid of binary cells onto which the system’s state was to be mapped (somehow). The system in turn had a minimum set of transformation matrixes to use which were instructed by sequences of code. These sequences of code were stored in a hierarchy to recreate structures seen on the screen. These sub-programs strings were then stored in a library and used where possible to simplify recording of the patterns in the binary cells. I was testing it on a simple starting circle to see how it would simplify the circle.

This didn’t actually solve the intelligence problem and I realised it wouldn’t generate self-consciousness either because the system would just enter a meaningless loop and becoming progressively less and less accessible to any outside observer. I shelved the project.

But I may resurrect A.I.M.E.now in search of algorithmic compression (I don’t believe anymore that consciousness resides “in” brains or any machines so its the wrong place to look for it).