Google AI Computer Now Reigning Chess Champ

Post Reply
 
User avatar
Richard S.
Mayor
Posts: 15183
Joined: Fri. Oct. 01, 2004 8:35 pm
Location: NEPA
Stoker Coal Boiler: Van Wert VA1200
Coal Size/Type: Buckwheat/Anthracite

Post by Richard S. » Fri. Jan. 25, 2019 10:51 am

https://www.chess.com/news/view/google-s-alphazer ... game-match

It taught itself to play chess in a few hours hours and then went onto to beat the top computational chess computer. Here is the key sentence.
Indeed, much like humans, AlphaZero searches fewer positions that its predecessors. The paper claims that it looks at "only" 80,000 positions per second, compared to Stockfish's 70 million per second.
The computer it beat is a dumb machine crunching ridiculous amounts of data to decide which is the most advantageous move. The AI computer is able to focus on calculations that will win the game relying on what it already knows about previous games.

This is why AI tech is so important going forward. It's simply not possible to teach a computer everything nor is it possible to provide them with the resources to utilize that data. e.g. a car cannot drive itself if it requires a computer the size of refrigerator, at least not in any practical sense. As the AI tech improves the resources they require get smaller, not bigger.

 
User avatar
BigBarney
Member
Posts: 1853
Joined: Wed. Feb. 08, 2006 2:48 pm

Post by BigBarney » Wed. Feb. 06, 2019 12:35 pm

The computers that will be in cars will have a neural network connected

from multiple sources to get the latest info for travel , and info

collected from vehicles that have traversed the same road ahead

of them so the path is constantly updated in real time but always

has a fall back of the stored info in case of malfunction .

This is a massive project and will take time to accumulate enough

data to be viable. Waymo seems to be doing good on the geo fenced

areas for a starter , which they can expand as required . Much more

complicated than a chess board with only so many pre determined

moves on a static board.

BigBarney

 
User avatar
Richard S.
Mayor
Posts: 15183
Joined: Fri. Oct. 01, 2004 8:35 pm
Location: NEPA
Stoker Coal Boiler: Van Wert VA1200
Coal Size/Type: Buckwheat/Anthracite

Post by Richard S. » Wed. Feb. 06, 2019 3:51 pm

BigBarney wrote:
Wed. Feb. 06, 2019 12:35 pm
and info

collected from vehicles that have traversed the same road ahead

of them so the path is constantly updated in real time but always

has a fall back of the stored info in case of malfunction .
Local information such as mapping data is the only thing that will important. That's already available through various sources and will drastically improve if every car is gathering it.
Much more

complicated than a chess board with only so many pre determined

moves on a static board.
There is an infinite amount of combinations of moves available on a chess board at least in any practical sense. The best human chess player might be looking 20 to 30 moves ahead. Back in the 90's Deep Blue was able to beat Kasparov the top chess champion of the time, the reason for that is the raw computational power exceeded his abilities and the computer was able to look much deeper down the road. In other words it was able to use brute force for a win.

This computer in the article cited is completely different, firstly it learned to play the game by itself without any instructions on the game itself. The best example I've seen given is you took a car apart down to every last screw and told the computer to assemble it with no instructions as to what it might.

From there we move onto the game itself, while it's certainly using computational power it's significantly less than other chess computers. It's calculating moves for the win based on what it has learned instead of relying on calculating as many moves as it can. Every game it plays will increase it's abilities.


 
User avatar
BigBarney
Member
Posts: 1853
Joined: Wed. Feb. 08, 2006 2:48 pm

Post by BigBarney » Wed. Feb. 06, 2019 5:14 pm

That is all true , but it is still a finite # of possible plays but

still calculatable whereas in an auto self driving program you

have almost unlimited # of constantly changing variables .

Chess is similar to geo fenced program because only a small

number of factors are included , and only prescribed moves

are permissive , and is only played in one dimension .

BigBarney

 
User avatar
Richard S.
Mayor
Posts: 15183
Joined: Fri. Oct. 01, 2004 8:35 pm
Location: NEPA
Stoker Coal Boiler: Van Wert VA1200
Coal Size/Type: Buckwheat/Anthracite

Post by Richard S. » Wed. Feb. 06, 2019 5:59 pm

BigBarney wrote:
Wed. Feb. 06, 2019 5:14 pm
That is all true , but it is still a finite # of possible plays
It's an infinite amount in any practical sense.

still calculatable whereas in an auto self driving program you

have almost unlimited # of constantly changing variables .
AI does not calculate in the way a conventional computer does. Most ten year old kids can identify the letter A displayed using an infinite amount of fonts(e.g"constantly changing variables") , that requires intelligence. For a conventional computer to do that requires you provide it an example of the letter A in every possible font. For an AI computer we provide it with 50 examples of the letter A and then teach it to identify the rest of them on it's own. It can identify an infinite amount of fonts just like a human.

One of the most important things to understand about AI vs. a conventional computer is that the conventional computer is always right assuming the human has not made a mistake programming it and there is no errors in the structured data that constitutes it's knowledge. This however severely limits their potential, trying to structure the entire knowledge of the world and program something that can utilize it is not going to happen. Even if you could structure that amount of data the computer resources required to utilize it will eat your lunch.

An AI computer on the other hand can be fed that knowledge as unstructured data, you feed it Wikipedia and 100 other encyclopedia's, medical journals, newspaper archives and anything else you can think of. The key here is it needs to take that knowledge and figure out on it's own how to use it. It can be wrong and when it is wrong it will go back and adjust the logic/reasoning that made it arrive at the wrong answer.

Post Reply

Return to “Technology”