Computers

Some starters. If there is a computer section to discuss these things, I apologize. I didn't see it.

Second, if you aren't into computer chess and AI don't bother reading this. It's going to be long and it will bore you.

The thread on old chess programs got me thinking about a couple things. I recall that back when chess programs started, that they were terrible. I couldn't beat them but I knew enough to know that moderately strong players were having no trouble with them. Their elo ratings were always drastically exaggerated in advertising. There was a radio shack one that I think claimed it was around 2200. I wanted it but I never did get one. Anyways, the NM who I hung out with had one at one point. I remember asking him how strong it was and I don't remember exactly what he said but basically it was awful. He had no trouble with it at all and he said on the hardest level it would take forever to make a move ... a terrible move. Time went on and computers eventually started playing GM's and eventually world championships. I remember Kasparov finally losing a blitz game and him stating that it was just blitz. The computer was terrible positionally. Under standard time controls he would beat it easily. Then of course there was the loss to the controversial Deep Blue and it was shortly after that where I stopped following. Now that I'm back into it, it's apparent that computers have surpassed us far enough that we don't even bother playing them anymore. If I'm wrong about this please feel free to correct me but I don't believe the top humans bother with the top computers anymore. I see that they have ratings in the 3300+ range. 

So my first question is... Computer ratings used to be inflated. Are their 3300 ratings inflated ... deflated? If humans and computer don't cross pools ever. How do we know that they are 3300? I'm extremely interested in the return to neural nets. I'm thinking we could be on the verge of another exponential leap in computer chess.

My second question. Kasparov said they were terrible at strategy. I believe I saw a recent GM criticize their positional play. What I don't understand is. How can humans criticize the computer's position if it is so much better than us that we can't beat it? Isn't that like an expert saying a GM is a terrible positional player? If the position seems better but there is a 40 move tactic that proves it wrong? how can you criticize the position?

Thanks for the response. I was a little surprised how little interest there is in this topic. 

I thought I read a quote somewhere from a GM recently who was still criticizing the computers at strategy/position. 

Fair.. perhaps a better topic for a comp sci forum. however, when I read about leela. my first thought was Skynet.

One area you are wrong is that people no longer play engines. Plenty of people still play engines, but when they are handicapped in some manner. The good side of that is they are always available for you to play. The bad side, is that some handicap in manners that result in decidedly non-human like play. Others are much better about that. In general at really weak settings, few of them play in a really human manner.

And the reason this didn't get more discussion is that it wasn't quite troll worthy enough. If you had said something like "Computers are better than all but .... players", you would have gotten a much better response.

It is stupid to compare computer ratings to human ones. Actually even human ones aren't uniform.

You just have to understand how these ratings (ELO, Glicko or whtever) are produced, and have minimal math knowledge.

True. Ratings only tell you relative strength within a pool of players (or computers). So unless you have a mixed pool of computers and players, you really don't know how they compare based on the ratings.

There are places like here that have mixed pools. However, the way that is implemented is very problematic and leads to all kinds of ridiculous results. For example, right now Komodo computer11 is rated worse than Komodo computer10 despite there being strong objective reasons to know that computer11 is better than computer10.

All you said is a bunch of utter nonsense.

Read post #12, your answer is there.

A very simple example, which proves that you understand nothing on the issue:

The strongest centaur player on ICCF is currently rated just a tad more than 2600, but of course he plays better than any non-engine aided human, or any standalone engine. It's the playing pool which decides his actual rating.

Got it now, or not yet?

"A good Human plus A computer is the best combination"

-Garry Kasparov

@DamonevicSmithlov wut

And this is why Garry's quote is so true.

Looks like you've put forward a water-tight case against Pfren - but can you say which bit of his statement the 3 PhDs you mention disagree with?

https://www.chess.com/forum/view/general/15-years-of-chess-engine-development-what-would-be-score-between-super-gm-vs-engines-now

Conclusion from that post! Stockfish 10( SF dev version)  4 cpu vs 2800 GM in 90min LTC in

400 games= 399 vs 1 or 399.5 vs 0.5 in favour of Stockfish. 

You have been told "why" at post #12, pretty clearly, by Madratter7. But apparently your severe ostrich syndrome causes perceptual blindness.

Ah, deleted your RS comment part, but anyway, whatever claim RS chess computer makes can be actually valid if they assumed a normal distribution ELO and give a very low confidence interval/level; technically they can probably say any ELO they want and give a near 0 confidence.

As for your first question, I think there are some crossover in human and computer play because there have been official games (games held under tournament conditions used to establish human ratings) between humans and computers.  Now, once that's done for computer A (here, I'm assuming your "computer" means a chess machine); then computer B plays A, that give computer B a rating, and so on.  Now, how accurate are those computer ratings? Well, how accurate is your USCF rating?

I think you answered your own 2nd question, and I agree with your answer.  In the example, if Kasparov can't beat the computer, then in some ways, he's just not able to see the computer's positional advantages.  Think of it this way, this problems looks like (while I'm typing casually) a prove that needs to go from the end; something like this, if we assume that checkmate is the ultimate position, then the position just before it must be very good, and so on.

Anyway, one of the very important issue WRT chess machines is the resources available such as processing power, memory, whatever.  It is probably totally correct (although you'll have to know the algorithm to say for certain) that a machine which can compute deeper than one that cannot, given some amount of time, should play better.  So, in a way, those RS chess machines, but in the day, even using battery, with barely any computational prowess and must make a move before its 15 yr old opponent runs out of patience ..can lay claim to 2200 whatever (USCF, I assume) is.. should not be believed... just on the surface.

The IPR garbage is well-known to the forum because it was used to attack a young woman for cheating. And that was embarrasing for chess and especially for the young woman falsely accused.

Of course that babbling idiot Regan ran away leaving this letter of stupidity...

http://chess-news.ru/sites/default/files/3.pdf

Too many jokes, but I like the one where he states that her rating is between 2085 & 2750 plus or minus deviations....also dont forget about plus or minus for the young fast improving parameter That meets the bing bang boom Rybkya Chiron move eval to help complete the 'full' crackpot test.

Bottom line: 'My analysis mostly argues against them'...i.e. Only he knows someone's rating. Plus or minus 1700.

Computers have an inherent weakness for strategy, since all they can do is calculate and evaluate based on set parameters. They are not able to form the super long term plans/ideas. That is there weakness. However, it scales with their rating. As computers improve their "strategy" improves given enough time/move check depth.

Imagine it like someone who is has better tactics than positional play. They love tactics but hate studying positional stuff. If they are at 1400 with those level skills, they cannot have the same level positional play as when they get to GM, it would have to improve to. Same thing with computers. 

Thats what I put "strategy" in quotation marks. Its not actually strategy. They cant actually do strategy but they are improving in positions we consider "strategic".