Message #3297

From: Eduard Baumann <>
Subject: Re: [MC4D] Re: Greetings
Date: Wed, 24 Feb 2016 09:43:43 +0100

I’m very amused to read "human-assisted" or "our obsolescence". This are necessary new humble terms after "godlike humans" and "from human invented gods".

Is there a 4D fan which doesn’t play go? Go alpha is a little bit sad.

Kind regards

—– Original Message —–
From: Melinda Green [4D_Cubing]
Sent: Wednesday, February 24, 2016 2:10 AM
Subject: Re: [MC4D] Re: Greetings

Welcome Thomas indeed! There’s nothing like breaking one of our most cherished records for waking us from blissful slumber.

I’m happy to create new solution categories though I’d like to make sure we have a rough idea how to determine where potential new methods will fall. Tools like reset, undo & redo are forms of computer assistance but they seemed gentle enough to not detract from an otherwise pure human solution. Macros OTOH required quite a bit more introspection and we ended up categorizing them as another valid human solution, though not directly comparable to no-macro solutions.

How should we now consider more sophisticated computer assistance that goes beyond simple mechanistic aid into some forms of real thinking? I definitely want to create a "shortest computer solution" category. So far we’ve only had one by Don Hatch which produces solutions with around 1,500 twists. Perhaps when we have a more efficient solution, we should create a "first" and "shortest" computer solution category pair like the others. I can also imagine a time in which a human only guides the broad strokes of an otherwise completely computer-driven solution. Would that be comparable to this new computer-assisted solution or should we then create a new "human-assisted" category in anticipation of our obsolescence? These are obviously more philosophical questions than practical ones but they might inform our immediate choices.

I’m really hoping that someone will implement an efficient computer solution that can be integrated into MC4D and replace Don’s version which I sadly broke. It was nice having a true solution integrated, though 1,500 twists turned out to be just a bit too tedious to watch. Something around 200 twists seems much more practical, plus it could be merged with the current "cheating" solution such that states involving less than that number of twists will simply reverse all those twists (ideally with some compression which I also broke), and solutions longer than that will use the full computer solution. In fact if we do this right, the real prize for the shortest computer solution could be its integration into MC4D!

And speaking of human obsolescence, are any of you also Go players, and have you been following the news regarding the upcoming human vs. AI match? Go has been a very difficult game to program due to the roughly 200 possible moves at every turn. Google’s AlphaGo AI recently crushed the European Go champion and next month will play against the legendary Lee Se-dol who has dominated the game for the last decade. That match has a $1,000,000 prize and could well be a watershed moment even bigger than when computers became the best chess players in 1996. The match begins on March 9th. Details here.

Happy puzzling!

On 2/23/2016 9:35 AM, Roice Nelson [4D_Cubing] wrote:

Welcome Thomas! 

My vote is that we mark this new solution as a first class record in a new category, "shortest computer assisted solve". I’ve considered trying to use the computer for attacks on the shortest competition in the past, and it’d be great if more folks were motivated to do this. With advances in this area we could, for example, come closer to intuiting what God’s Number for the 3^4 might be. We don’t even have a rough idea of what it is right now, upper OR lower bounds (as far as I know).

Congrats on your impressive solve, and happy to have you posting here.


On Tue, Feb 23, 2016 at 10:21 AM, Thomas Lehéricy [4D_Cubing] <> wrote:

Indeed. I looked at the wiki page after my first solves, but didn’t understand everything and preferred to keep going with the method I was designing - at this point using an analogue of CFOP was just the thing to do, seing how intuitive it is when you know it well. Now that I read it again it looks very clear, and it indeed looks the same as my own method up to the last layer.

  The last 3D face can be done in at most twice as many moves as one would need for the 3D cube. To do that, one can simply &quot;regrip&quot; (rotate) the cube so that the face one turns is always the same. Of course it can be improved, for instance when using URU'R'&#58; a single regrip in the middle and all moves will cancel... So it's only an upper bound. I don't know of any general method to optimize this step, although I would be extremely interested.

  The human Thitlethwaite is not particularly efficient at giving low-move counts solutions, but still better than CFOP. I think you can hope for a 40-50 move counts on average if you know all cases (which I don't), without optimizing it for too long. What is good is that each step is rather intuitive, and it can be optimized and yields extremely good results&#58; Kociemba's algorithm is derived from it.

  Block-building methods seem the thing to do indeed. It seems to me that Matthew Sheerin built his first two layers like this in his record. It is not as optimized nor as flexible as Heise, and it would be indeed interesting to see how well Heise translates into 4D - but that's far beyond my abilities right now.

  Thank you for your answer!