I recently found what I consider to be a most useful AI paper: Genetic Self Assembly There is some really great information about how to use genetic algorithms to design self-assembling logic gates that can do useful things like act as a general purpose ALU or Multiplier. The possibilities are endless. The self-assembling aspect seems to be the key to evolving worthwhile systems. Unfortunately it is nearly unreadable. 0/10 for style, 10/10 for content. If you present things badly you get no respect and are ignored, no matter how important your discovery. I do know that from my own experience. To save the day I have done a fact sheet for the algorithm: Fact Sheet Can I publicize my "O'Connor Transform" discovery again? Yeh, why not: O'Connor Transform

Quote: Original post by redtea
I recently found what I consider to be a most useful AI paper:

Genetic Self Assembly

There is some really great information about how to use genetic algorithms to design self-assembling logic gates that can do useful things like act as a general purpose ALU or Multiplier. The possibilities are endless. The self-assembling aspect seems to be the key to evolving worthwhile systems.
Unfortunately it is nearly unreadable. 0/10 for style, 10/10 for content.
If you present things badly you get no respect and are ignored, no matter how important your discovery. I do know that from my own experience. To save the day I have done a fact sheet for the algorithm:

Fact Sheet

Can I publicize my "O'Connor Transform" discovery again? Yeh, why not:

O'Connor Transform

The problem with a complex system being formed is the ratio of failures increases geometrically with the complexity. The test which the logic 'organism' has to pass may be quite long with many graduations of failures. That boils down to many hard to evaluate cycles of evolution.

One thing I thought of -- every so often you need to take your candidates (especially near the end) and chop off bits and and retest. Logic could grow with 'do nothing' redundant parts which dont make the whole thing fail the tests , but do add cost.

The main point of the paper is that shows how the solve problems Inductively using a genetic algorithm. If it solves how to design an 8 bit by 8 bit multiplier it has automatically solved how to design a 32 bit by 32 bit multiplier . You just use a bigger board and let more components self assemble (that is to say the complexity and evolution time does not grow with problem size, for problems with some structure in their solution). That is the big break-through.
The fact that the paper claims to have solved how to design a general ALU should be a wake-up call. That has not been done before by any artificial intelligence system before.

As I said the paper is heavily obfuscated with references to chemical and biological systems and has been badly translated , nevertheless it is very important.
It does show that very interesting results can be obtained by taking a multi-disciplinary approach.

At the very least it is a fun idea that seems reasonably easy to implement.

[Edited by - redtea on July 21, 2008 3:37:59 AM]

A Google search on computational self assembly seems to be a good way to start locating papers.
Nice diagrams and a few ideas.


[Edited by - redtea on July 25, 2008 7:00:19 PM]