I mentioned previously that I’d decided to try to solve the Eternity II puzzle game and write a ‘solver’ for it, possibly using some kind of AI.
Of course, I haven’t yet found a solution, though I do have a couple of promising approaches. The ‘final’ deadline is December 31st and it won’t be solved by then, though it would be nice to see them extend it for another year (no-one has yet won, and it would be much more satisfying from both a company and buyer perspective to have a winner, rather than it just be abandoned!) but I don’t know if this will happen.
See what happens when we get a few days off over Christmas!!
You’d think trying to solve a (seemingly) intractable problem (the general class of problems like Eternity II must surely be NP-complete, so an AI approach might come up with a “very good” solution but won’t guarantee to find the exact solution, though it may do of course…) would be quite frustrating and frankly a waste of time but in fact there’s a certain stubbornness and determination that “no-one can solve that, I can solve it!” amongst people like me that makes problems like this almost more worthwhile to work on than mundane things like a ‘solver’ for those sliding number puzzles.
What bothers me about all this though is that although an AI algorithm might “happen” across an exact or at least very good solution (which I’d then be tempted to tweak by hand) it’s a fuzzy approach by nature in that a solution evolves or is stumbled across; what I’d really like is to truly understand the puzzle to the extent that a solver could find all possible solutions based on a deep structure or rule. Where I’m starting to look is the number of solutions they mentioned that there are, and how else this number can be generated.