Hi everyone. I think we’re collectively at our best when we explore new things. As things seem we’re a little fractious right now let me offer up my idea for what might be a pleasant diversion.
It kind of has a genome and fitness criteria, right?
Let’s take just one 3×3 grid. It could have a genome of 9 digits, and competing fitness functions: (sum or product, max or min) for 3 rows, 3 columns, 2 diagonals and of course some rule about using all the digits (or not if we want better mutations). Each of the 9 genes would affect 2 or 3 (or 4 for gene number 5, in the middle of the square) of the fitness functions. Can we create a simulation, with drifting fitness functions and see how organisms evolve. Will this show islands of function and a path to traverse between them? This might be fun because a mutation can help in one regard whilst hurting in another. I’ll leave this here for now, let me know if anyone is interested…
I’ll be interested if you can make something of this. However, to me, evolution and sudoku seem to be polar opposites.
Sudoko is a logic problem. It is logic all the way. Pragmatics are of little use.
Evolution is pragmatics all the way. We can describe it in such a was that it might look like logic, but that isn’t how it really works.
Sudoku might be a good example of “irreducible complexity”.
Rich is proposing modifications to Sudoku so that it isn’t merely a logic problem. Take a look at his last paragraph.
Grid Weasel. Not that there’s anything wrong with that.
That’s two dimensions. You could make it three, four, or N, for even more fun. Biology has many dimensions of fitness.
I’m betting that when you increase the number of fitness dimensions, an evolutionary algorithm will outperform a human. Just as with the Travelling Salesman problem.
Constraint satisfaction — a topic in introductory AI courses.
The first “evolutionary” approach that comes to mind is to make the problem into one of multiobjective optimization. Define “fitness” functions, one for each row, column, and 3-by-3 grid. The fitness of a region might be the number of distinct digits that it contains. There’s a slew of ways to handle selection in multiobjective evolutionary optimization. I generally would not sum the values of the multiple objective functions, i.e., to produce a scalar fitness, but it might be best here to start with just that, and see what happens.
Rich, I hate seeing you with no takers. I’ve been going hard at something else, and don’t have many sparks left that will jump synapses.
Evolutionary computation is really not the way to do this sort of problem. But it would be interesting to see what happens. Recombination would be fun, because there are three reasonable ways to do it, and no reason I can see (when brain dead) to exclude any of them: exchange columns or exchange rows or exchange 3×3 regions of parents. That is, generate 6 offspring by recombination of 2 parents in 3 different ways.
No worries, Tom (& Keiths) – they can’t all be winners, or maybe its time is not yet.
I enjoy the point at which evolution meets games people understand and the solutions are too complicated to be solved intuitively but is not so complicated that it can’t be followed. My hope is that “show me” doubters will be able to follow along and see evolution in action.
Intelligent Design can produce games like sudoku.
Which games have been produced by un-intelligent non-design?
The games you play. Mung.
If I had the time to spend on giving you a dose of your own medicine, I’d produce a bunch of quotations regarding applications of game theory to evolution, and add emphasis to all occurrences of the word game. I mean, game theory wouldn’t work if evolution were not a game — right?
Arguments from improbability (including those that express improbability on a log scale, and refer to it as information) are rooted in the misconception of evolution as a problem solver. The fundamental reason that Marks, Dembski, and Ewert are wrong is that they treat models of evolution as problem solvers.
I’ve got to disagree with you there. Mung’s games are carefully designed to achieve his goals — any response whatsoever.
I’ve been swamped at work but I do find your idea interesting. I wrote a Sudoku solver that was based on the algorithms people use rather than the usual constraint satisfaction approaches. That was an interesting exercise.
Using an EA would be interesting as well. I’ll see if my GA engine can be used for that.
I think I might agree.
It definitely would be interesting to explore
It might cut down on Mung’s seagull commenting if we gave him a dedicated thread in which people dropped by occasionally to say ‘hi’ and validate his existence.
It might be fun to propagate constraints in a genotype-to-phenotype map, and to treat constraint violations in the phenotype as lethal. I’m tempted to chew through the leather straps (restraints are constraints), and give it a try.
That make you what, yet another sucker?
Possibly, depending on what my goals are. Clearly I need to rethink them.
See what it leads to? (See that I don’t have time for Sudoku?)
Therefore the weather is designed because weather simulations are created through iterative debugging and tweaking.
This is yet another example of how intelligent design creationism is scientifically vacuous.