ClashOfPawns2 írta:
Any reason in particular you didn't include a link to something about the match? Sure seems like a no-brainer.
Just read the most recent Forum/News/Blog Entry by FM Mike Klein. It is pretty easy to find.
A New Force In The Universe! AI has arrived.
Sort:
Dec 7, 2017
0
#23
Have a look at the "snopes" website, it gives lots of good info about the 10 % myth
ClashOfPawns2 wrote:
RMChess1954 wrote:
When Gary Kasparov lost to Deep Blue he lost to a database of knowledge gathered by mankind for over 500 years, administered by man made algorithms that applied that knowledge. Today Google's DeepMind learned the game of chess and a few hours later had created it's own knowledge base. Then it crushed the current world champion chess engine. Now the computer DeepMind was running on may have been 1000 times faster then the one Stockfish was on but still the implication is a whole new way of computing. One in which the computer may find new approaches to the problems we face. It may be better, much better at things like how to create a fusion reactor, or how to cure cancer. Think about that for a minute.
Any reason in particular you didn't include a link to something about the match? Sure seems like a no-brainer.
What website? Chess.com main page has the story. Maybe you want Deepmind's page. HERE
Do you think you could have beaten Deepmind in its very first game after having the rules programmed in? Did it literally turn from an amateur player to the greatest of all time in 4hours? Or was it unbeatable (from a human perspective as soon as it was programmed?)
Genuine question.
TrippyOAFC wrote:
Do you think you could have beaten Deepmind in its very first game after having the rules programmed in? Did it literally turn from an amateur player to the greatest of all time in 4hours? Or was it unbeatable (from a human perspective as soon as it was programmed?)
Genuine question.
Yes. Because it did not have an evaluation function originally.
Yes. That's how fast it learned the game of chess.
Dec 7, 2017
0
#28
I'd love to be able to understand ciarli's posts! I'm sure they're interesting but I never understand them
ciarli appears to be a chatbot with a large amount of randomization. This is Sokal-level poppycock.
Dec 7, 2017
0
#30
Hey RMChess1954, cool topic indeed, thanks for sharing your thoughts. I thought about what you said. it's true that lots of good stuff could happen, but same goes the other way around with the bad stuff, so I'm not sure if we should be happy or not with all these new frontiers. In any case, what I find interesting is that while DeepMind presents a groundbreaking method, it's still something that I'll never be able to comprehend. I mean, we're never able to understand what those machines do, we can only see them play and beat other chess engines. For an average Joe like me, it's quite frustrating. But there's a cool product in Beta called DecodeChess, and they claim their algos explain machine decisions in human language. Now that kind of thing really get me hooked up because at least I can try (!) and understand what those machines do. Anyways, back to the board...cheers.
Dec 7, 2017
0
#31
Anyone who thinks they could learn to draw against AlphaZero should note that is about as difficult as learning to get a consistent 90% score against Carlsen.
Dec 7, 2017
0
#32
RMChess1954 wrote:
When Gary Kasparov lost to Deep Blue he lost to a database of knowledge gathered by mankind for over 500 years, administered by man made algorithms that applied that knowledge. Today Google's DeepMind learned the game of chess and a few hours later had created it's own knowledge base. Then it crushed the current world champion chess engine. Now the computer DeepMind was running on may have been 1000 times faster then the one Stockfish was on but still the implication is a whole new way of computing. One in which the computer may find new approaches to the problems we face. It may be better, much better at things like how to create a fusion reactor, or how to cure cancer. Think about that for a minute.
Not 1000 times.
Stockfish was running on 64 threads, which is 32 Intel cores. AlphaZero used 4 TPUs which are supposed to be about 30 times as fast as a fast CPU (multicore, presumably) for running neural networks (which involves a lot of matrix operations). So maybe 30 times faster (for what AlphaZero needed to do).
Note that AlphaZero can only evaluate 70,000 positions per second, enormously less than Stockfish. It just does it a lot better!
RMChess1954 hat geschrieben:
Deepmind running on a supercomputer but without an opening database. Using algorithms it created itself.
I find this bit to be the most important.
I always felt like normal chess engines were cheating when they used an opening database - after all, human players aren't allowed to bring their collection of opening books to the table.
Deepmind sems to have gotten rid of the need for such cheating, which is a significant step forward in technology.
Dec 7, 2017
0
#34
Elroch wrote:
RMChess1954 wrote:
When Gary Kasparov lost to Deep Blue he lost to a database of knowledge gathered by mankind for over 500 years, administered by man made algorithms that applied that knowledge. Today Google's DeepMind learned the game of chess and a few hours later had created it's own knowledge base. Then it crushed the current world champion chess engine. Now the computer DeepMind was running on may have been 1000 times faster then the one Stockfish was on but still the implication is a whole new way of computing. One in which the computer may find new approaches to the problems we face. It may be better, much better at things like how to create a fusion reactor, or how to cure cancer. Think about that for a minute.
Not 1000 times.
Stockfish was running on 64 threads, which is 32 Intel cores. AlphaZero used 4 TPUs which are supposed to be about 30 times as fast as a fast CPU (multicore, presumably) for running neural networks (which involves a lot of matrix operations). So maybe 30 times faster (for what AlphaZero needed to do).
Note that AlphaZero can only evaluate 70,000 positions per second, enormously less than Stockfish. It just does it a lot better!
32 cores with 1GB for hashtable...
IpswichMatt wrote:
The idea that humans only use about 10% of their brains is a myth
In my case is true.
Dec 8, 2017
0
#36
Optimissed wrote:
It has zero to do with A.I. It's just an interactive database.
Absolute nonsense!
The main thing is that AlphaZero acquires all of its game playing skill by learning from experience. That is AI. Nor is is a database: neural networks learn thousands of parameters so that they can perform a fuzzy analysis of entirely novel data to produce (in this case) a single number rather than storing data and performing programmed operations on that data (which is what databases do).
I have a very old Facebook friend who works for the company that designed this and I was just talking to him about it ten minutes ago.
He works for DeepMind? What is his name?
When I heard about it I was able to guess how it worked in about five seconds and he said I was right.
I am sure you said something that was right. However, I bet you did not say everything that is in the paper on the subject.
That's cos I'm intelligent and this thing just compares patterns very fast.
That part is only partly true: it also learns how to represent, describe and evaluate the patterns in a way that all involve innovation by the AI. The concepts represented in the AlphaZero neural networks have in some cases never been known to humans and may have no close analogy to what is known.
This is not the first time an AI has achieved innovative results. Some nice examples are that one AI generated a novel design of radio antenna, another discovered a new type of star.
instead of looking at hardware, we have to see that stockfish was evaluating 80 million positions vs 700k by alpha zero in one minute. that is a hell lot of advantage for stockfish. I would love to how alpha zero will perform when both will evaluate equal number of positions
Dec 8, 2017
0
#38
Optimissed wrote:
It has zero to do with A.I. It's just an interactive database. I have a very old Facebook friend who works for the company that designed this and I was just talking to him about it ten minutes ago. When I heard about it I was able to guess how it worked in about five seconds and he said I was right. That's cos I'm intelligent and this thing just compares patterns very fast.
The supercomputer is using a neural network to find patterns. Artificial Intelligence is probably a word thrown around a little too much, maybe better is machine learning. Its "learning" through pattern-recognition, much like a self-driving car. Its definitely a step towards an A.I. with diverse learning capabilities. Most modern A.I. are using pattern-recognition based methods of sorting, so just because its not a walking, talking robot doesn't mean its not A.I.
Dec 8, 2017
0
#39
I think there's another forum discussing this subject/subjects too. I just posted in it.
But there - they're talking about jobs being replaced by machines.
Dec 8, 2017
0
#40
NationalPatzer wrote:
Optimissed wrote:
It has zero to do with A.I. It's just an interactive database. I have a very old Facebook friend who works for the company that designed this and I was just talking to him about it ten minutes ago. When I heard about it I was able to guess how it worked in about five seconds and he said I was right. That's cos I'm intelligent and this thing just compares patterns very fast.
The supercomputer is using a neural network to find patterns.
Actually, this is a bit misleading if you mean some sort of pre-specified patterns. Rather, AlphaZero uses neural networks to produce two numbers: one is the evaluation of the position and the other is the probability that a move is best. Most other applications of neural networks involve them being used to identify specified patterns.
Artificial Intelligence is probably a word thrown around a little too much, maybe better is machine learning. Its "learning" through pattern-recognition, much like a self-driving car. Its definitely a step towards an A.I. with diverse learning capabilities. Most modern A.I. are using pattern-recognition based methods of sorting, so just because its not a walking, talking robot doesn't mean its not A.I.
The particular form of machine learning used by AlphaZero (reinforcement learning) is, in my view, a stronger flavour of artificial intelligence than those types that are most widely used (mainly supervised learning).
Surely most people would agree that a computer learning everything about how to play chess successfully from experience - only starting with knowledge of what moves are legal and what the objective of the game is - is very much artificial intelligence. By comparison, a neural network that learns to recognise which photos contain cats is impressive but has done less on its own.
When Gary Kasparov lost to Deep Blue he lost to a database of knowledge gathered by mankind for over 500 years, administered by man made algorithms that applied that knowledge. Today Google's DeepMind learned the game of chess and a few hours later had created it's own knowledge base. Then it crushed the current world champion chess engine. Now the computer DeepMind was running on may have been 1000 times faster then the one Stockfish was on but still the implication is a whole new way of computing. One in which the computer may find new approaches to the problems we face. It may be better, much better at things like how to create a fusion reactor, or how to cure cancer. Think about that for a minute.
Any reason in particular you didn't include a link to something about the match? Sure seems like a no-brainer.