Computers vs Humans

[Batmaz_The_Unsinkable]

Do you guys think people like Carlsen, Kramnik, Aronian, Caruana, etc. could stand up to some of the super computers of today if one was purpose built to process chess? I'm really beginning to doubt whether a person could win considering the speeds of today's computers. The Deep Blue that "beat" Kasparov was made of 120Mhz dedicated chess chips. What if you throw in a few 32 thread 3 Ghz dedicated chip from today, along with some of today's graphics processors? I understand it would take an immense time to create the software and faster hardware would also be available. I also understand that building a dedicated machine like this is not as simple as taking hardware and throwing into a machine then loading software onto it, but still. Use some imagination.

I'm just trying to start a conversation here. I think it's an interesting topic to consider.

[BoredErica]

=I'm just trying to start a conversation here. I think it's an interesting topic to consider.

Actually, I'm a computer chess guy, so I can answer this.

 

You don't need a supercomputer to beat these players. (Not that our engines today would even run on those.) Top GMs have said more than once that they can't even beat their cell phone. Stockfish or Komodo on a cell phone is stronger than Deep Blue. This is not due to increased computing power, this is primarily due to better search, better software. It's a big misconception that computer chess is primarily about hardware.

 

On one side you have Stockfish, which is an open-sourced engine developed by a lot of people and tested by even more people. I help test Stockfish by running a simple program, in essence, folding, but for computer chess development. Many, many hours of CPU time goes into development to get a large enough sample size to figure out if a change is good or bad or neutral.

 

Then you have closed-sourced Komodo, which has to be purchased. That team is being run by a few old guys - but very experienced and dedicated guys. Right now Komodo is leading by a small margin.

 

Grandmaster Nakamura has agreed to play Stockfish and Komodo on multiple occasions in the past. The games are always handicapped so Nakamura has a chance. The trick is to find the right handicap to make the match interesting.

 

Nobody has found a way to make GPU chess work well. We are still using CPUs, and the struggle to make multi-core support better and better is real. 2, 4, 8, 12, 24, 36, 36+ cores, there is still a lot of work to be done. Whether hyperthreading increases or decreases strength is still a point of debate. In most applications that use many cores, HT is just better than no HT, but chess is an exception.

 

And finally as a side note, just like how there are AMD and Intel fanboys, there are Stockfish and Komodo fanboys. Imagine me sitting there, listening to both sides yell at each other over pieces of chess code. ._.

 

EDIT:

Typos like you wouldn't believe, lol.

[DXMember]

Do you guys think people like Carlsen, Kramnik, Aronian, Caruana, etc. could stand up to some of the super computers of today if one was purpose built to process chess? I'm really beginning to doubt whether a person could win considering the speeds of today's computers. The Deep Blue that "beat" Kasparov was made of 120Mhz dedicated chess chips. What if you throw in a few 32 thread 3 Ghz dedicated chip from today, along with some of today's graphics processors? I understand it would take an immense time to create the software and faster hardware would also be available. I also understand that building a dedicated machine like this is not as simple as taking hardware and throwing into a machine then loading software onto it, but still. Use some imagination.

I'm just trying to start a conversation here. I think it's an interesting topic to consider.

there are strategies to beating Computers at Chess, Kasparov was one of the first to come up with that

basically nowadays because the hardware is imnesly fast the competition is really the Grand Master versus the wittiness of the programmer

 

i suggest you read up on some chess algorithms for Computers, basically because it's proven to be impossible to calculate all positions on the chess board, algorithms just cut off the illogical, dumb, losing moves and keeps the calculations going with only the winning moves in mind. Those algorithms don't expect that the opponent will willingly make a move that is harmful to them, like sacrifice the queen on 3rd move. From the standpoint of algorithm that seems absurd, so it cuts off those posibilities and doesn't look in to it, if the opponent actually does a stupid move like that on purpose then Computer will have to recalculate everything again, most likely run out of memory and freeze. That's basically how Kasparov and other Grand Masters win at chess to Computer nowadays - they find flaws in the programming because they know how Computers "think"

[paprikman]

You don't really need a powerful cpu to play chess: almost all combinations are known today (lol but there are bilions of them) and you can put them in memory. The chess are about making mistakes in strategies.

[manikyath]

---

an interesting read. would you consider the prediction of future moves to be something possibly multithreaded?

(theres many paths to think about, so theres potentially many threads to do this on)

or is predicting just such a processing sink that it makes zero sense?

[theereal]

You dont need a supercomputer to beat a human In Chess. Humans make errors, computers do not. However, when It comes to what Is more advanced, Computers or Humans, Humans come out on top. Unlike Computers, Humans can learn new abilities and adapt, unlike Computers which need to be told what to do. And who tell them these things? Humans A dog Is more advanced than a computer.

[Headshot_]

-snip-

[DXMember]

You dont need a supercomputer to beat a human In Chess. Humans make errors, computers do not. However, when It comes to what Is more advanced, Computers or Humans, Humans come out on top. Unlike Computers, Humans can learn new abilities and adapt, unlike Computers which need to be told what to do. And who tell them these things? Humans A dog Is more advanced than a computer.

you'd be surprised what we can achieve with modern AI

[theereal]

you'd be surprised what we can achieve with modern AI

Youre going to need to elaborate...

[BoredErica]

 

an interesting read. would you consider the prediction of future moves to be something possibly multithreaded?

(theres many paths to think about, so theres potentially many threads to do this on)

or is predicting just such a processing sink that it makes zero sense?

There are multiple ways to do multithreaded search. That stuff pretty much goes over my head. I do now that the latest trend right now is called Lazy SMP, and we recently ditched Young-Brothers-Wait. https://chessprogramming.wikispaces.com/Parallel+Search  (Although, I find that wiki to be a hard read...) You can have a large speedup in positions per second but not get a corresponding increase in strength. A good example is with hyperthreading. You get a net gain in speed but a net decrease in strength (or so the argument goes). But yeah, there is definitely multithreaded search. Right now I think the top engines can use 8 cores quite well, and it drops off from there. Still, from time to time I see tangible improvements in strength from improvement in how the engine uses the cores even with less than 8 cores. There's still a lot of improvements possible, I think. I used to think that chess engines used each core almost perfectly, but that's not the case.

 

The engine tries to play the best moves, assuming the opponent will also play the best moves. When it's the opponent's move, the engine may try to think during that time (called pondering), but that's usually not used for engine vs engine matches.

 

When it comes to even more specific things than that though, most of the technical details are too complicated for me. I just watch the live engine tournaments and do folding for Stockfish, basically. Some of the most ballin' guys have 36 cores @ 2.8ghz... I took my 6600k @ 5ghz out to play with my benchmark tryhard hat on, and those guys just blew my score away. Check it out actually, I'm on the list (ctrl f Dark_wizzie). It's one way to benchmark CPUs, too. https://sites.google.com/site/computerschess/stockfish-chess-benchmarks

 

 

You dont need a supercomputer to beat a human In Chess. Humans make errors, computers do not. However, when It comes to what Is more advanced, Computers or Humans, Humans come out on top. Unlike Computers, Humans can learn new abilities and adapt, unlike Computers which need to be told what to do. And who tell them these things? Humans   A dog Is more advanced than a computer.

Computers make errors. They just make errors less often and (usually) on a far smaller scale that would take very deep calculation to take advantage of. Computer chess engines isn't this monolithic thing which spews out perfect moves. There are many chess engines of many different strengths.

 

The funniest thing is watching chess engines play a blocked off, drawn position. Computers have no idea what's going on and will shuffle pieces endlessly.

[DXMember]

almost all combinations are known today (lol but there are bilions of them) and you can put them in memory

The Nalimov tablebases, which use state-of-the-art compression techniques, require 7.05 GB of hard disk space for all five-piece endings. To cover all the six-piece endings requires approximately 1.2 TB. It is estimated that seven-piece tablebases will require between 50 and 200 TB of storage space.

 

how exactly are you planing to do that?

[DXMember]

Youre going to need to elaborate...

look up machine learning for new skills without programming new skills

and the adaptation for Computers is a thing of past it's been done since 1980's how do you think ABS Works - it adapts to the evironment

also voice recognition would be impossible without adaptation

[theereal]

look up machine learning for new skills without programming new skills

and the adaptation for Computers is a thing of past it's been done since 1980's how do you think ABS Works - it adapts to the evironment

also voice recognition would be impossible without adaptation

But who created this adaptation program for the computer? We are able to create computers, but try to put together a sentient being and you WILL fail. A computer Is not sentient, that was what I was getting at.

[BoredErica]

But who created this adaptation program for the computer? We are able to create computers, but try to put together a sentient being and you WILL fail. A computer Is not sentient, that was what I was getting at.

Once upon a time, man built the fastest computer ever and asked, 'Is there a god?', and the computer replied, 'There is now.'

 

Today we watch computers play chess, in the future computers will watch us play chess.

 

[theereal]

Once upon a time, man built the fastest computer ever and asked, 'Is there a god?', and the computer replied, 'There is now.'

 

Today we watch computers play chess, in the future computers will watch us play chess.

 

Actually just left the forums for a little break but I had to answer you Do you actually know If thats true? I believe thats an old "folk tale" by Eisenhower, I dont think that ACTUALLY happened But yes computers are becoming more and more advanced

[BoredErica]

Actually just left the forums for a little break but I had to answer you Do you actually know If thats true? I believe thats an old "folk tale" by Eisenhower, I dont think that ACTUALLY happened But yes computers are becoming more and more advanced

I don't think that actually happened.

[DXMember]

But who created this adaptation program for the computer? We are able to create computers, but try to put together a sentient being and you WILL fail. A computer Is not sentient, that was what I was getting at.

what is sentient?

send an AI back a couple thousand years it will be praised as a divine being

 

You just think like that because you know how Computers work, if you knew how dogs work or how humans work you would think the same of them that there isn't really anything inteligent about them.

 

Biological organisms are just a bunch of random bio-Chemical processes and underneath there is nothing inteligent at all.

Artificial Inteligence is just a bunch of random binary strings and underneath there is nothing inteligent at all.

 

but when you take millions and billions of these unintelligable things and put them together it Works wonders.

 

A single ant can't do much, but there are billions of ants in a nest and together they can act as a complicated being.

A single neuron in a human brain can't do much, but there are billions of neurons in human brain and together they can do unimaginable things.

A single human can't do much, but there are billions of humans on the earth and together we can do wonderful things that would be uninmaginable a couple of thousand years ago.

A single Computer can't do much... but when we will have billions of them connected...

[Batmaz_The_Unsinkable]

I think the drawback of humans playing is that they have emotions, and they have the ability to become nervous. That is what Kasparov said after he played Deep Blue. Computers don't care about that is going on around them, or how complicated a situation on the board is, they just go on doing their calculations. He also said that humans fail in very dense and complicated situations, as they can lose track of what happens after each subsequent variation and eventually make a mistake. That won't happen in computers.

 

ON a separate note, deep blue was said to have been able to do 200 million calculations per second. How would that compare to in today's machines? If we are doing billions or trillions of calculations per second, how long would it take for every important variation to be completely played out and a victory found for the computer, assuming that the computer could differentiate between good and useless positions. Humans would have no chance of winning.

 

I don't know about computer being dumbed down for grandmasters, have never heard of it. Is that actually true?

 

About AI, I don't know if it is true or not, but it is expected that within the next ten years, we will have computers that will operate faster than the human brain  and will likely contain the code necessary for self learning. It is a very frightening concept. Look at the Google Car, or Tesla's self driving system. We are already almost there, basically. I don't think the complete automation of life is very far off. 

 

Wall-e, here we come. Prepare for blended food and floating chairs.

[BoredErica]

I think the drawback of humans playing is that they have emotions, and they have the ability to become nervous. That is what Kasparov said after he played Deep Blue. Computers don't care about that is going on around them, or how complicated a situation on the board is, they just go on doing their calculations. He also said that humans fail in very dense and complicated situations, as they can lose track of what happens after each subsequent variation and eventually make a mistake. That won't happen in computers.

 

ON a separate note, deep blue was said to have been able to do 200 million calculations per second. How would that compare to in today's machines? If we are doing billions or trillions of calculations per second, how long would it take for every important variation to be completely played out and a victory found for the computer, assuming that the computer could differentiate between good and useless positions. Humans would have no chance of winning.

 

I don't know about computer being dumbed down for grandmasters, have never heard of it. Is that actually true?

 

About AI, I don't know if it is true or not, but it is expected that within the next ten years, we will have computers that will operate faster than the human brain  and will likely contain the code necessary for self learning. It is a very frightening concept. Look at the Google Car, or Tesla's self driving system. We are already almost there, basically. I don't think the complete automation of life is very far off. 

 

Wall-e, here we come. Prepare for blended food and floating chairs.

Positions per second is an inteligible metric for everybody, but searching 100 million pointless positions versus searching 100 million good positions is a very different thing. It's like trying to figure out which GPUs are the fastest by looking at how fast the core clock is. The software on Deep Blue was very primitive compared to today. You can go for higher depth (search less possibilities but search more moves in the future for that line) or less depth (search more possibilities, each to a shallower depth). 1,000 positions searched in engine A is different than 1,000 positions searched with engine B. The same applies to different hardware using the same engine, although to a lesser degree.

 

I was listening to This Week in Tech, and there the tech panel predicted self-driving cars in 5 years (one guy said 10 due to regulations).

 

I believe Kasparov actually resigned in a game that was a draw. I think his emotions had a big part to play in that instance. If he had remain perfectly calm, calculating and thinking until the very end, he could have probably drawn the match. The entire match could have turned out very differently. We have a sample size of only a few games, but it looked like Deep Blue wasn't far superior to Kasparov, maybe even just his level sans emotional factors. Kasparov hadn't lost in a standard time control game in like, ever, back then. It must have been tough psychologically to see just how far engines have come.

 

Engines also have an advantage in opening with opening books. You can analyze the possible (good) openings and pull up what grandmasters have done in the past, analyze that with a computer (or computers) for hours on end and find many new things. And you can give this opening book to the engine, so that the computer doesn't even have to think at the start of the game, essentially memorizing moves better than any human can. In the ending, where engines typically get wonkier, we have what are called 'end game tablebases', which DXMember alluded to. We have already figured out all the possibilities for a game with 7 pieces of less. That can be put on SSDs so that when the engine calculates to that depth, it can look at the tablebase and know whether the position is won or loss without calculating (but we need SSD so that the lookup is fast).

 

The thing engines suck at is positional play. But the large majority of the time, the speed of an engine more than makes up for it. Positional could be things like how much you value your rook, considering the fact that it's trapped and can't do much. Or how much of a weakness have a hole in your pawn structure is. etc. You need a blend of positional and tactical play of course. You can calculate super well, but unless you calculate to checkmate, the engine has to figure out whether the ending position it calculated is good for black or white. So the engine has to evaluate the position. Generally as a human you want to make the game more positional and closed off because engines are weaker in those conditions. As I've mentioned previously, an extreme of that is with a totally closed off game, for example if all 8 pawns are in a deadlocked. Engines typically don't know what's going on.

 

The engines aren't "dumbed down" for grandmasters. By handicap I meant like, engine has to play without the b pawn, etc etc. Here is an example: http://komodochess.com/store/pages.php?cmsid=17

Which pawn the engine gives up matters a lot, even at the start of the game. Then you can come up with handicaps like, the human gets to play two moves on move one, etc etc. When you give a minor piece (like a knight) to the player and the player is a top level grandmaster though, that much be too much, I'm not sure. When you play with a knight down, we call that "knight odds".