A program that can analyze its own code and re-write itself to be more efficient.

This is an archive of the mabination.com forums which were active from 2010 to 2018. You can not register, post or otherwise interact with the site other than browsing the content for historical purposes. The content is provided as-is, from the moment of the last backup taken of the database in 2019. Image and video embeds are disabled on purpose and represented textually since most of those links are dead.

To view other archive projects go to https://archives.mabination.com

Kueh wrote on 2012-04-06 23:56

It isn't brand spanking new, but I first learned about this stuff today.

Basically, as long as the program can prove that the change will be more useful than its current code, it will change itself to adapt.

A significant development for artificial intelligence researchers: to write a program that can not only learn, but also improve and adapt.

It's a difficult read, though.
http://www.fhi.ox.ac.uk/__data/assets/pdf_file/0014/16421/Ultimate_Cognition_a_la_Godel.pdf
TA wrote on 2012-04-07 00:11

If that's true then depending on its rate of adaptation then it could possibly be the technological singularity - the last invention we would ever need to make.

Or the start of Skynet. Pray for the former.
Sekwaf wrote on 2012-04-07 00:12

So an actual "Falken's Maze" (I believe it was just made up in the movie War Games). I would kill to have my hands on something like this, it would make programming a hell of a lot easier I bet.
Xemnas wrote on 2012-04-07 01:51

I always thought Osay would be starting the groundwork for AI.
Rin wrote on 2012-04-07 03:04

I believe something similar was used to beat the game of Civilization I, by learning, adapting, and countering the game's AI.
Kueh wrote on 2012-04-07 04:44

Quote from Rin;832856:
I believe something similar was used to beat the game of Civilization I, by learning, adapting, and countering the game's AI.

Code that allows a program to learn has been around for about a decade by now, if you count the very primitive forms.

But code that can actually re-write itself, has never been produced before.

The amazing part of this code is that it does not require a teacher.

Typically, learning code has some kind of user-defined goals (such as defeating an enemy AI). This code can allow programs to learn without ever being told what to do. In a simulation, a robot learned to balance a pendulum on its body without being told what to do.

[Also, and arguably the most fantastic part of this kind of program, is that it can be used as a model of many human behaviors, including curiosity, aesthetics, and sensation patterns like eye movement.]