The two links are highly informative regarding the future of programming.
Will computers ever solve chess?
Sort:
Jun 19, 2017
0
#1063
"These abilities let ??? convert local regions of influence into coordinated global advantages."
Sound familiar or remind you of anything?
Jun 19, 2017
0
#1064
On a lighter note...
"We accomplished this by taking the historical data that had already been collected by thousands of sensors within the data centre -- data such as temperatures, power, pump speeds, setpoints, etc. -- and using it to train an ensemble of deep neural networks."
Sound familiar or remind you of anything?
Grandmaster games?
Database?
Jun 19, 2017
0
#1065
"If you want to solve it in a strict sense of the word, then yes, you need a complete tablebase."
Disagree
There is no need of 99.99 % of possible positions that clearly are incorrect, the path to reaching them can be eliminated well in advance.
Jun 19, 2017
0
#1066
99.99% of all possible positions will never lead to best possible play, a criteria for solving chess. It's that simple. There is no need to consider 1. Nf3 2. Nf1. 3. Nf3 by example after 2 good moves by Black. The position can be discarded. Did anybody read the links from Alphago and the Deepmind program? Obviously not.
Jun 19, 2017
0
#1067
PP, your reasoning is faulty because there is no way to reliably reject positions based on heuristics. Any rule that is not associated with a direct calculation is either going to be very specific (eg an abstraction of analysis of a very specific class of endings) or plain wrong. Moreover, there is no way to know which heuristics are right for SURE without testing them thoroughly. Here you are claiming that, based on the positional understanding of average players, a certain opening is surely inferior. You may be right (or it may be that the line is good for a draw and therefore best!) but the idea that this is reliable enough for a proof is laughable.
For example, you might come up with a rule: "If you have K+QRRBBNNPPPPPPPP v K+N, you are surely winning. Wrong. In some positions you are losing. There are some weird blocked position problems where very unlikely results take a long time (the longest problems are absurdly wrong - hundreds of moves!) And again, even if you come up with a rule that is so intricate and precise it happens to be right, you have no way of knowing this without exhaustively checking it.
The only rule that is useful and valid has been mentioned above: if a win can be forced, you don't need to worry about alternative tries by the winner. The practical consequence of this is that if you suspect a position is winning, you can look only at the more plausible ways to achieve this. If you succeed, that's good enough. This does not help at all in positions where the result remains the draw it very likely starts as!
Jun 19, 2017
0
#1068
I have written a lot of code for chess over the years.
I have designed many financial systems and databases for global banks over the years.
I solve very complicated problems and make things simple in practice, that is my actual job.
This is what you are going to need, just to get this thing started.
Note that the strategic objective of this design is not to "solve" chess, it is to create a solid foundation for data analysis, this is something that is lacking with chess.
Programming
C or C++
I have discarded all languages except for C and SQL, I do not care for abstraction in code.
Data Files
.pgn
.xml
.txt
.bin
A lot of the work has to be done outside the relational database with C and data files.
RDBMS
MySQL
It supports UINT64 directly unlike some other relational databases.
Designer
Some smart alec that knows what he is doing.
That's what you need just to get the database system started.
PGN --> TXT --> BIN --> XML --> RDBMS
B+ TREES and other data structures.
UINT64 Bit Boards.
Do you see?
ProfessorPownall wrote:
99.99% of all possible positions will never lead to best possible play
Great, that leaves only 10^42 positions left to go.
btickler wrote:
ProfessorPownall wrote:
99.99% of all possible positions will never lead to best possible play
Great, that leaves only 10^42 positions left to go.
Is that a big number for computers (or whatever we call them because they will probably do far more than compute) 100 years from now? I think the answer was never if, only when.
s23bog wrote:
Actually, to be thoroughly "solved", all continuations in which white can force a win should be found. Not just one (if it exists).
It does not exist. Hard to find something which does not exist.
Jun 20, 2017
0
#1073
#155512 hrs ago
PP, your reasoning is faulty because there is no way to reliably reject positions based on heuristics.
Disagree. Here is the simple logic. If chess were to be solved by a computer(which it won't by brute force, ever) and it were to be found that 1. Nf3 2. Any other move but the Knight returning to it's original square, leads to the best possible result, we can dismiss 1. Nf3 2. Nf1 3. Nf3 as a viable search. This is called "logic" which is the goal of AI.
Jun 20, 2017
0
#1074
s23bog wrote:
Does one single processor have to count from 1 to 10^42, or can it be broken up, and the work divided up among many computers?
There is plenty of potential for parallelisation. But even if you use say 1,000,000,000,000 processors (costing more than the world's GDP and using more than all of the world's electricity) you still have 10^30 positions for each, which is too many. As if that was not enough you need to store all of this information (which cost more than a million years of GDP) and share it between the processors.
Jun 20, 2017
0
#1075
ProfessorPownall wrote:
#155512 hrs ago
PP, your reasoning is faulty because there is no way to reliably reject positions based on heuristics.
Disagree. Here is the simple logic. If chess were to be solved by a computer(which it won't by brute force, ever) and it were to be found that 1. Nf3 2. Any other move but the Knight returning to it's original square, leads to the best possible result, we can dismiss 1. Nf3 2. Nf1 3. Nf3 as a viable search. This is called "logic" which is the goal of AI.
No, that's an error. You need to do the same WITHOUT first solving chess using a computer.
What you are saying is "I can get to Japan without using transport. The way I do it is by first going by plane to Tokyo and then getting off and saying 'I am in Tokyo and don't have any transport'".
Jun 20, 2017
0
#1076
No, that reasoning is in error. There may be several paths which need not be solved before embarking on the search. Obviously you can not reach the destination before making the journey, the cart before the horse. But most forks in the road lead to dead ends and can be recognized beforehand. They all need not be traveled. Same for the "chess solving" problem. Analyzing all possible moves and or positions, besides being fruitless, is not the way to proceed. A proper course would begin with implementing previously learned data, eliminating unnecessary efforts and time.
Jun 20, 2017
0
#1077
I reiterate. "If" 1. Nf3 eventually leads to a solution (which is not known ahead of time) then we can safely dismiss 1. Nf3 2. Nf1 3.Nf3 in the search from the very start. I stand by this "logic". This eliminates the need to search all possible positions.
bb_gum234 wrote:
lfPatriotGames wrote:
btickler wrote:
ProfessorPownall wrote:
99.99% of all possible positions will never lead to best possible play
Great, that leaves only 10^42 positions left to go.
Is that a big number for computers (or whatever we call them because they will probably do far more than compute) 100 years from now? I think the answer was never if, only when.
Yes, 10^42 is really big.
If the entire planet were made of pencil lead, and you tried to use it to write all the numbers from 1 to 10^42, and you were able to write really small... so small you'd need a microscope to see the numbers, you wouldn't be able to do it. You'd use up the whole earth.
So let's forget chess for a moment. Computers will never count from 1 to 10^42.
The fastest computer we have right now, if it were to start counting to 10^42, would take more time than the age of the universe, and this isn't even storing the information (as we would need to do with a solution).
So as long as computers (or at least our storage devices) aren't at least planet sized in the next 100 years, then no, that's not enough time
Yes, future technology will be amazing, but it will not be magic!
Well then solving it seems even more likely then. Compare the computers today with the ones 100 or 200 years ago and it's likely the ones today seem like magic.
Jun 20, 2017
0
#1079
They arleady have. A 3400 hundred rated comp is essentially perfect chess.
Jun 20, 2017
0
#1080
Why do you spam so much man?
Trying to keep the topic floating on the list of recent posts for longer?
So that more people see it, so that you can carry out your hidden agenda?
I know what your hidden agenda is.
I know what people do on the forums.
I know how the forum lists work, 3 quick posts like you always do keep it longer in the list.
Stop it now spammer.
One of the great promises of AI is its potential to help us unearth new knowledge in complex domains. We’ve already seen exciting glimpses of this, when our algorithms found ways to dramatically improve energy use in data centres - as well as of course with our program AlphaGo.