Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!rice!uw-beaver!mit-eddie!media-lab.media.mit.edu!minsky
From: minsky@media-lab.media.mit.edu.MEDIA.MIT.EDU (Marvin Minsky)
Newsgroups: comp.ai
Subject: Re: HI-Q game
Message-ID: <5753@media-lab.media.mit.edu.MEDIA.MIT.EDU>
Date: 29 Apr 91 20:04:29 GMT
References: <1991Apr24.055054.16724@cunixf.cc.columbia.edu> <1991Apr29.192045.1852@lynx.CS.ORST.EDU>
Reply-To: minsky@media-lab.media.mit.edu (Marvin Minsky)
Organization: MIT Media Lab, Cambridge MA
Lines: 22

In article <1991Apr29.192045.1852@lynx.CS.ORST.EDU> tgd@tesla.orst.EDU (Tom Dietterich) writes:
>This puzzle has been solved using an elegant machine learning method
>by Glenn Iba:
>
>Learning by Discovering Macros in Puzzle Solving.  Proceedings of the
>Ninth International Joint Conference on Artificial Intelligence
>(IJCAI-85), 1985 Vol. 1 640-642.
>
>Iba's program selectively learns macros ...

Hi Tom!   Incidentally, Iba figured out some of the macros himself,
first.  For example, if you have the configuration below:
  
0 0 0    where x means empty space, then after six moves all are gone
0 0 0    except for the bottom 0
0        Now, consider that the entire game consists of just 4 of
x        these patterns plus a few other units.  If you think about
         this for a while, you should be able to solve the entire
puzzle in your head!

I remember not believing Glenn when he discovered this, until doing it
myself.