Slashdot Mirror
How Windows FreeCell Gave Rise To Online Crowdsourcing
TPIRman writes "In 1994, a physics doctoral student named Dave Ring assembled more than 100 math and puzzle enthusiasts on Usenet for what became one of the earliest online 'crowdsourcing' projects. Their goal: to determine if every hand in Windows' FreeCell solitaire game was in fact winnable, as the program's help file implied. Their efforts soon focused in on one incredibly stubborn hand: #11,982. They couldn't beat it, but in the process of trying, they proved the viability of an idea that would later be refined with crowdsourcing models like Amazon's Mechanical Turk."
I can figure out how to solve Free Cell...
(scrambles back to Spider Solitaire)
In real life, with real mines. Terrible results. While we did find most of the mines, it turns out that people are terrible at safely locating them. Lots of dead bodies, limbs, etc, everywhere.
It doesn't look like he ever proved that the hand in question was not solvable. It only claims that by having many human players try to solve it and several different AI approaches, that it was never solved.
The article ends by implying that this was a victory, because the outcome of all 32,000 hands is now known. But, as far as I can tell, one hand is still undecided!
Free unix account: freeshell.org
Unless the exploration of the game space was exhaustive, there's no proof.
Wikipedia claims that exhaustive search has been performed, assuming that the same hand numbering is used:
http://en.wikipedia.org/wiki/FreeCell#Impossible_games
Based on my own results, I'd have to say that thirty-one thousand, nine hundred, and ninety-nine hands are not winnable, and one is.
As a corollary result, I seem to have proven that I really suck at FreeCell.
No, the first large distributed project is the Cunnigham project:
http://books.google.com/books?id=udr3tHHwBl0C&lpg=PA375&ots=s4GNA3LkQo&pg=PA375
that started in 1949 on the ENIAC !
And this project is still ongoing.
In fact, this search started with human efforts, so it was already heavily crowd-sourced since a least 3 centuries.
The culmination of the manual effort came in 1903, when Frank Nelson Cole showed that:
193,707,721 × 761,838,257,287 = 2^67 - 1
It took 3 years of Sundays to discover.
http://www.rutherfordjournal.org/article030105.html
This was also done at about the same time in the UK, by a group of people on Cix (a CoSy conferencing system), with the same conclusion, except we found two more unsolvable ones that I suspect the American team didn't look at: -1 and -2. For what it's worth, I invented the notation we used to document solutions, and one of the team produced a solver that exhaustively checked the game space for 11 982 and indeed found it impossible. So give or take formal proof of the solver's correctness, it is proven that not all games are solvable.
Quidnam Latine loqui modo coepi?
Here's a program that does it.
http://kurage.nimh.nih.gov/tomh/public_html/archives/patsolve-3.0.tgz
The program generates a list of axioms, followed by a list of transformations chosen from a finite set.
After a finite number of steps, the proof reaches a conclusion that that game (and that's the only one
out of the original 32000) is unsolvable. This is a real, valid mathematical proof. It's just very long
and hard to read. But it is of finite size, and follows all the normal rules of mathematical proof.
You're welcome to try to come up with a shorter proof.
The article seemed to imply that it was proved unwinnable, but never absolutely stated it..
I found something a little better: http://www.solitairelaboratory.com/fcfaq.html
11982 has now eluded solution by probably thousands of human solvers, and at least eight independent computer programs I am aware of (most of which are designed to search exhaustively for a solution), and I am confident in calling it impossible.
I really wish the FAQ had been linked from the slashdot summary, it's far more interesting, and better written than the gaemological.com article.
As I understand the quote above, there is at least 5 different programs (ie. more than half of "at least eight") that have solved hand 11982, and all arrived at the same solution: #11982 is not winnable. This has persuaded the FAQ's author to call [winning the hand] impossible.
Free unix account: freeshell.org