Meanwhile, three new records have been found. The program knows two things. When the spheres (circles) are intersecting and that this is very bad, and that the smaller border cube (square) - the better. Then, the evolution is going on, the survival of the fittest configuration.
The publishing is still a human prerogative and it will remain so for now.
Pretending how insignificant this is, is a kind of amusing to me.
What is this and why is it of interest?
It is the automation of something a few people can take part of. Well, almost anybody can try, but only several will achieve any good result. Things are just too complicated, but the results are easily verifiable.
This class of mathematical problems: easy to understand, difficult to solve and easy to verify - it is a very important class.
So it is a very good benchmark for your abilities or your algorithms.
On the other hand, the packing problems in general, have a lot of practical uses every day. Here is a automated solution relentlessly pursuing unknown solutions, no human an no program has thought about them before .
A humble Windows PC program solving NP problems for you to admire and to study those solutions, if you want to.
What is this and why is it of interest?
It is the automation of something a few people can take part of...
So it is a very good benchmark for your abilities or your algorithms...
These are not justifications for posting this article to Less Wrong, as opposed to somewhere else. The article, while interesting to me, is not about the art and science of human rationality. Downvoted.
Also, note that we have threaded comments - it looks like the parent was intended to be a reply to gwern's comment.
As I've already mentioned here before, our small goal is to built a program which only purpose is to search and submit new, previously unknown dense packing achievements to the Internet without any human intervention except to start it and to provide the hardware, power, internet connection and such. Every solution is the program's creation and innovation.
It has been done. A small program is "scavenging" over the www.packomania.com and reading the best packing solutions there. It tries to find a better one. If it succeeds in 8192 seconds, then the program publishes the result on the http://www.algit.eu/htmlji/Packntile/Packing_Contest_01052010.html and sends it to Eckard Specht http://hydra.nat.uni-magdeburg.de/ by an email.
(Well this emailing has been cut out already as unnecessary. Packomaia is updated from our site, directly, by mister Specht.)
If there is no new better solution in those 8192 seconds, a new random problem is selected and pursued by the program. The program (formerly known as Pack'n'tile) also doesn't care if the current target solution is its own or of a human. Neither if its previous effort on that particular problem was not successful. Or if it was. He doesn't remember it, anyway.
For now, there will be only one instance of the program somewhere near us. It is fast and powerful enough to run on a modest PC computer for a solution per day. It would took decades to populate this large searching space of various dimensions, shapes and numbers predominantly by its solutions, but the impact is already visible. Of course, more and faster CPUs will be provided for the job, eventually.
We don't want to take the fun out of the game. On the contrary, the solutions are the most important and we are providing them.