Re: Re : Re: CCC Discusses Chess - Posts of Interest -What Is The Future Of Chess?
This is preposterous. It's the first time I've seen such an argument. Computers only look at more than one board at a time, move pieces, and look up opening theory in their "minds". They don't physically set up tables, move pieces, or explore opening theory.
Humans are also capable of doing these things in their minds. This ends all arguments as to whether computers are "cheating". Again, preposterous.
The human mind and the computer mind are obviously structurally and functionally different. Right now, the computer mind is superior at chess because, even though it cannot reason and think, it can calculate much much much faster than the human mind. But keep in mind, the calculations themselves and the methods of doing them are set by human programmers, many of whom are average chess players. If you tell those humans that they can only create software that uses a human-brain-like paradigm, i.e. a neural net, then things change dramatically. The programs become much "dumber" in that they must be trained and they take very long to advance even to rudimentary level. Their training is just like the human brain needing to be trained (with rare human exceptions, where certain prodigies seem to already have the necessary training "built in" even at a very young age, which leads into very philosophical avenues of discussion).
What does this all mean? It speaks to chess itself. It says that chess is a game that is solvable by brute force calculation. Most of us already know that and also know that the brute force search tree is inimaginably large. But not infinite.
If we already know this about chess, then we already know that chess can be solved by computers. Thus the actual solving of chess will change nothing. Just like the fact the speed of light is the fastest way to travel 100 yards doesn't stop us from holding 100-yard dash events at the Olympics. We don't actually think humans might surpass the speed of light via running.
Another point: humans create the computers and the software that drives them. A computer chess program that is the strongest chess player in the world is indicative of the human progress in one particular domain which happens to be chess -- even if the program is a product of a collection of humans rather than one individual human.
The solving of chess by computers is thus really the solving of chess by humans.
I submit that parallel to solving chess by brute force calculation (which would be interesting, to see if chess is indeed a draw by brute force), we should also see if we can produce neural net programs that can (with sufficient training) surpass human brain capabilities at chess. Only then can we say that we have surpassed the incredible capabilities of the human brain, and not by mere brute force calculations, but by actual computer software ARCHITECTURE that mimics the brain. In other words, we could then say that we have created a BETTER BRAIN that can learn and solve not only chess, but any data- and calculation-centered domain.
Just as there can be different human brains, there can be different computer neural nets. And just as one human brain could conceivably respond differently to different trainings (i.e. "upbringings"), one computer neural net could do the same. Thus you can have infinite computer neural nets each responding to infinite trainings. This gives hints into a near or distant future: computer robotics that learn, and can be "brought up" by humans and compared to other computer robotics with different neural nets and different upbringings.
Finally, it has to be brought up again: what about these human prodigies that seem to have so much of the usual "training" already built into their brains? Mozart composed classical music from the age of 5.
This begs the question: can we marry computer neural net architecture with already-built-in "training", thus bypassing to some extent the "learning" phase of neural net programs?
The thing is that we humans would benefit by being able to set up many boards and actually move pieces on each board to physically see with our physical eyes different variations, whereas computers would have no such limitation. They don't need to physically see or hear anything. They just deal with 1's and 0's at (currently) GHz speed. Thus this marriage of architecture and training is really a marriage of architecture with brute force techniques that chess might help teach. Accordingly, we should be able to someday create computer brains that learn at incredible speed.
This is all getting very philosophical, i.e. "could we create God?". Well, I haven't even brought up that we humans can create chess variants, which no computer program, brute force or neural net, could do. So even given a neural net computer that can master chess, we're still eons away from creating God, if you believe in such a concept (which I do).
So I'll stop there and end with this bottom line on Kevin's post:
Computers do not "cheat" at chess. They visualize just like we do, except differently. They store their visualizations in their memory, just like we do except differently. Visualization and use of memory is not cheating.
Originally posted by Kevin Pacey
View Post
Humans are also capable of doing these things in their minds. This ends all arguments as to whether computers are "cheating". Again, preposterous.
The human mind and the computer mind are obviously structurally and functionally different. Right now, the computer mind is superior at chess because, even though it cannot reason and think, it can calculate much much much faster than the human mind. But keep in mind, the calculations themselves and the methods of doing them are set by human programmers, many of whom are average chess players. If you tell those humans that they can only create software that uses a human-brain-like paradigm, i.e. a neural net, then things change dramatically. The programs become much "dumber" in that they must be trained and they take very long to advance even to rudimentary level. Their training is just like the human brain needing to be trained (with rare human exceptions, where certain prodigies seem to already have the necessary training "built in" even at a very young age, which leads into very philosophical avenues of discussion).
What does this all mean? It speaks to chess itself. It says that chess is a game that is solvable by brute force calculation. Most of us already know that and also know that the brute force search tree is inimaginably large. But not infinite.
If we already know this about chess, then we already know that chess can be solved by computers. Thus the actual solving of chess will change nothing. Just like the fact the speed of light is the fastest way to travel 100 yards doesn't stop us from holding 100-yard dash events at the Olympics. We don't actually think humans might surpass the speed of light via running.
Another point: humans create the computers and the software that drives them. A computer chess program that is the strongest chess player in the world is indicative of the human progress in one particular domain which happens to be chess -- even if the program is a product of a collection of humans rather than one individual human.
The solving of chess by computers is thus really the solving of chess by humans.
I submit that parallel to solving chess by brute force calculation (which would be interesting, to see if chess is indeed a draw by brute force), we should also see if we can produce neural net programs that can (with sufficient training) surpass human brain capabilities at chess. Only then can we say that we have surpassed the incredible capabilities of the human brain, and not by mere brute force calculations, but by actual computer software ARCHITECTURE that mimics the brain. In other words, we could then say that we have created a BETTER BRAIN that can learn and solve not only chess, but any data- and calculation-centered domain.
Just as there can be different human brains, there can be different computer neural nets. And just as one human brain could conceivably respond differently to different trainings (i.e. "upbringings"), one computer neural net could do the same. Thus you can have infinite computer neural nets each responding to infinite trainings. This gives hints into a near or distant future: computer robotics that learn, and can be "brought up" by humans and compared to other computer robotics with different neural nets and different upbringings.
Finally, it has to be brought up again: what about these human prodigies that seem to have so much of the usual "training" already built into their brains? Mozart composed classical music from the age of 5.
This begs the question: can we marry computer neural net architecture with already-built-in "training", thus bypassing to some extent the "learning" phase of neural net programs?
The thing is that we humans would benefit by being able to set up many boards and actually move pieces on each board to physically see with our physical eyes different variations, whereas computers would have no such limitation. They don't need to physically see or hear anything. They just deal with 1's and 0's at (currently) GHz speed. Thus this marriage of architecture and training is really a marriage of architecture with brute force techniques that chess might help teach. Accordingly, we should be able to someday create computer brains that learn at incredible speed.
This is all getting very philosophical, i.e. "could we create God?". Well, I haven't even brought up that we humans can create chess variants, which no computer program, brute force or neural net, could do. So even given a neural net computer that can master chess, we're still eons away from creating God, if you believe in such a concept (which I do).
So I'll stop there and end with this bottom line on Kevin's post:
Computers do not "cheat" at chess. They visualize just like we do, except differently. They store their visualizations in their memory, just like we do except differently. Visualization and use of memory is not cheating.
Comment