Humans v Houdini chess engine (Elo 3300)

Everything. And I mean your assumption, not the ELO system.

Yeah, I've just looked into it a bit more (than not at all). I was way off. Apologies.

Factly, the ELO system was already suffering from inflation quite some time before engines made it to the top. The statistical model on which it is based is due to a radical revision since a long time ago.

I'm sorry if my english isn't good enough to explain it my i'll give it a shot:

The Elo system is based on confrontation between members in a given group. If you have 2 separated groups you will have 2 different Elo systems.

If computers never play with humans they can't be assigned a FIDE rating, they have a rating on their own.

If I start a new federation tomorrow morning just with me and other guys who never played chess and I use an Elo system to calculate ratings i will soon have an incredible Elo rating but that won't make me a stronger player.

BTW I believe computers should have much higher ratings if they played more with humans

While this is a commonly seen assertion, this paper disagrees with it: http://www.cse.buffalo.edu/~regan/papers/pdf/ReHa11c.pdf.

Looks like an interesting reading, thank you!

My "feeling" about ELO inflation is based on estimates by people who know way more than myself, and have contributed to the evolution of the rating system since at least 30 years ago. This might be a good new thread for discussion.

 That paper confirms what I suspect (and not just about chess, but about most competitive occupations), a tendency for the avergae skill level to get better over time. I think due to things like learning from past masters (standing on the shoulders of giants) a larger player (population) pool to draw talent from, better training methods (computers for example, Lasker did not have access to CT-ART!).

My guess is any of the top 5-10 players today, if they could step into a time machine, would become world champion in any year prior to like 1980 or so.

My guess is any of the top 5-10 players today, if they could step into a time machine, would become world champion in any year prior to like 1980 or so.

Especially if they could bring their laptop with them !

 I think if I could bring my iPhone into the time machine, I could be world champ too. I even have my explanation ready:

Em. Lasker : "Hey...whats that your looking at during our game???"

InvisbleDuck: "Nothing."

I think this would work.

That's interesting! The investigators in this paper analysed 1000s of master games with a super-engine (Rybka) and recorded how often players made the same move as the super-engine's recommendation. They found that the stronger the player, the more often their moves matched the super-engine's moves.

Their data shows that as a players strength increases, the probability that they will play the same move as a super-engine gets closer to 1. Therefore they assumed that the super-engine's moves were almost always the best available.

if i played a correspondence game against houdini useing houdini to analyze, of course i would win .so correspondence chess doesnt really count when it comes to gaugeing an engines strength because the poor engine doesnt  have any entity to help it.the top engines have good positional understanding but not as good as a gm of course but the fact is ,unless you play correspondance against the engine without the aid of another engine its pointless to to easily dismiss the awesome playing of engines like houdini. id love to see a match with the top dog with out the aid of a computer! Play houdini with full 6 man tables uesing a duel socket hex core  like the computer in this utube vid

http://www.youtube.com/watch?v=hLBhvzeE_TE&feature=plcp&context=C37a91f4UDOEgsToPDskKGsD7Gd8NOY7ZwBmh7Ojil

now that would be a match! id put my money on houdini

 Yes, but this also is OTB and not correspondence.

We dont have a correspondence test available ( that I know of) of a super GM playing Houdini at a time control of 3 days/move.

Some very valid points in here. Houdini is a combinational monster, but regarding abstract strategical planning and endgame technique its play leaves a lot to be desired. This is why the strongest centaurs are regularly fairly strong players (say IM level) which when planning in their correspondence games just turn Houdini off, and AFTER they have decided about the overall game strategy turn it on to check for tactical flaws in their findings.

Not that sure a super-GM will win hands down on long-time CC, but he does have certain advantages against a machine. The problem is that we will have no concrete proof about anything because there is no motivation for super-GM's to play such games.

It's strange to hear that, pfren. I'm only going by my individual experience, so it doesn't mean anything, but I remember when I switched from Fritz to Houdini, and I couldn't believe how human-like and positional it seemed to be. Of course, it's all relative, and today, as much as I respect it, I can still tell it's a computer :)

Still, Fritz is a commercial engine, and I think Houdini is way better, so I think you should maybe give it more credit. With that said, I don't have any experience with other engines.

its awesome

This QUESTION ASKED TO VISHWANATHAN ANAND... NOT FOR SWIDLER.

For all of the people singing Houdini's praise, I own Houdini, I use it for some things, and it's spectacular at calculating and finding moves I wouldn't normally see, but not much use for anything else.    For all that it can do and how strong it is, I'm still amazed at what it gets wrong and how it's really not an excuse for human thought.  For example, when I play the moves 1. c4 e5 2. g3 Nf6, Houdini immerses itself in deep calculations and then proceeds to tell me that black is up 1/5th of a pawn.  It operates on a formula based mostly on brute calculation, to the point that when it doesn't recognize something and there's nothing tactical in the position it'll just spout out a random number based on who it thinks is better developed. Houdini also frequently fails to understand positional domination - in a position where one move "stalemates" a piece and changes the ending into an eventual win and another wins a pawn, Houdini will take the pawn 10 times out of 10.  When Houdini is outside of a tablebase and calculation range in the endgame, it doesn't understand how to push for a win in a drawn position or how to plan ahead and improve your pieces so that your position is at its most optimal point before eventually taking up action.  I agree with pfren in that people vastly overestimate the capabilities of engines - we're intimidated by their impeccable tactics so we overlook a lot of flaws.  

What you say applies to any chess engine, though. As bad as Houdini may be, I think Fritz is even worse -- when I switched from Fritz to Houdini, Houdini seemed so positional and human-like. Of course, now it doesn't quite seem that way, but I like it a lot more than Fritz. How do you know Houdini only judges based on development? It likes square coverage, but are you not just guessing on what it's basing its evaluations on? There are some positions though where it doesn't have a plan -- I have seen blocked positions where it likes one side (because there are more pieces), but the other side is the one with pawn breaks. As you start to play out the plan for the latter side (pawn advances on a wing), it can often change its mind.

On the other hand, a lot of people use the fact that sometimes the computer is wrong as an excuse to disregard what it says. I think the computer usually has a pretty good feel for positions simply because, if you can't be broken down, then your position is probably pretty good, even if it looks ugly. That's why computers are sometimes ok with making weakening pawn moves -- if you can't take advantage of the weakness, what good is it?

So I think in one sense you shouldn't trust everything it says -- but you should also have a lot of regard for its opinion too -- most of the time -- not all of the time -- it is seeing a (critical) variation that you aren't. If it makes an anti-positional move, it's usually because it doesn't think it can be exploited, perhaps because there is too much dynamism in the position.

What I find interesting about Houdini is the difference of opinion it sometimes has compared with DR4, even by so much as half a pawn. It's good to have both perspectives. In the particular position I was evaluating, it looks like Houdini was way off and that the DR4 choices were much better.