Humans v Houdini chess engine (Elo 3300)

Well I came back to this topic a few days later hoping to have

• Pfren give some justification to his hypocrisy
• Pfren give some examples
• A lot more discussion into the topic

It seems instead we've gotten a dozen comments from people trying to point out Pfren's crappy debate tactics and a half dozen poor attempts to dodge the issue.

Pfren should look up Ad Hominem. Simply put, the first step is to directly insult the opponent instead of their argument. For example, calling them fools. The second step is to claim that the opponent is wrong because they are whatever the insult was, i.e. that the fool will always look at the finger.

• You are a fool; you are therefore wrong because fools are wrong.

Lol at the pithy sayings book for christmas.

I would vote to close this thread unless someone besides FirebrandX wants to bring forth some evidence. His is quite insightful to be honest and has slightly shaken my belief that computers are incontrovertibly stronger than humans.

I still believe correspondence chess is absolutely the only way to have a snowball's chance in hell against a computer. Human + engine is a different matter entirely because the computer's occasionally dumb play (see IBM's Watson on Jeopardy for something frighteningly similar) can easily be spotted by a human who can use a limited amount of judgement to apply the tactical genius way beyond him/her.

I still stand by the belief that the small positional advantage a human can gain is far outmatched by the fact that the computer will play tactically sound chess from a slightly poorer position and will pull through. FirebrandX showed me a game where he is quite sure a computer would not have managed to come back but I'm just not convinced.

I should mention a probable cause for the computer playing more poorly with longer analysis times is the way it decides which move is best. I know this is vague but I will explain.

If it is comparing three moves after which white can make one from a choice of five and it has determined the "value" of each of the 15 lines, how does it project the "values" onto its own three choices. For example, if in line number 1, move a is slightly losing, move b is drawn and moves c d and e are slightly winning, does it say, probabilistically that line 1 is winning on average, or does it say instead that at worst, line 1 is slightly losing?

The whole Qe5 story might be because Houdini saw that out of white's however many possible moves, only a single one is slightly losing and every other move is instantly winning and therefore Qe5 is a 90% (or whatever) instantly winning move on average in a probability sense.

This is why the engine outputs are so critical (and very good for tactics analysis). I've heard of examples (possibly here) of an engine missing the best move and then evaluating that very move (after being specifically instructed to, I suppose) and finding that it is in fact better than anything else it suggested.

This means the engine's programing is flawed. Some bad shortcut made the engine throw away that move without considering it. An engine that thinks like this (throwing away moves, that is) is not brute force, specifically. Brute force is more like looking at every legal move in a position, looking at every legal move after each move from the previous position and so on to figure out which move from the considered position is best. [If a computer worked like this, whenever the book ending finished, the computer would spend a long time (many years with today's computing power) to figure out the best move and the would play the rest of the game instantly as it would have considered every line its opponent had to offer. In a sense, it would have solved chess. Basically, you could mimic this by creating a database of every legal position and associate a best move to each and have the computer search through it. I'm sure someone is doing this. THAT is brute force.]

True brute force cannot be beat. Period.

At first I was convinced by YawMawn but then I read the lasr part, lol

@YawMawn I don't think that's how computers search (of the five moves available to opponent four are winning). The main point of mini max is that the evaluation of the line is equal to the evaluation of the best move, alternating for min and max depending on turn.

pfren is a forum troll that has something against engines in general.

Let's not forget the fact that in a 3 day corresspondence, the computer can think 6 days straight - that is, 24 x 6 hours per move. The human being will be hard pressed to analyze at maximum efficiency after 5 hours. (You've gotta eat, sleep, drink, right? That's assuming you devote this entire period of your life to this game. At 3 days per move, it ain't happening.)

With Houdini 3 out now with a large ELO jump, I'm willing to challenge a GM/superGM to a 3 day per move match, provided a GM has the time, energy, and interest to do such a thing. (Spending entire months on a game seems incredibly boring.) My computer isn't even top notch - old Q6600 at 2.7GHZ. Wait till' some dude with dual 6-core Xeons walks in, which...what, multiplies the speed of calculation by 6? 8? 10?

Carlson could beat it, he has talent, determination and youthful energy that can overcome the machine. 

Actually, Pfren could probably beat Houdini in a 10 minute per side match because the computer simply calculates whereas a human with a deep understanding can easily find good moves quickly simply because they know chess principles at a subconscious level.  It would be Pfren's deep understanding and wisdom vs. Houdini's sophisticated 21st century brute force algorithms that put even Hydra to shame.  

The flaw with your reasoning is you're simply going by pure calculation, not understanding.  If one has an understanding they can immediately find candidate moves and calculate them whereas a machine calculates everything.  Also, overheating could be an issue.  I have water cooling but not for Chess reasons.  Haswell is coming out soon too and they'll have 8-core models.  

I'm afraid you don't know very much about Houdini, chess engines destroyes humans in blitz chess. 
Understanding is NOTHING compared to pure strong calculation. And Houdini finds candidate moves quicker than a player, just because it only calculates a million moves per second.

After a few split seconds of thought Qf7 appears! (with a correct evaluation to boot)
On my core 2 (10 year old laptop) running only houdini 1.5 !!!

FEN: 1rr2bk1/3q3p/1p2p1p1/pQnpPp2/3N4/BP2PPP1/P1R3KP/2R5 b - - 0 1

Houdini_15a_w32:
  4/15    00:00           3,077    1,025,000    -0.36    Qf7 Bxc5 Bxc5 Nc6 Rb7
  5/15    00:00           5,377    537,000    -0.29    Qf7 Bxc5 Bxc5 a4 Rb7 Qd3
  6/15    00:00           7,488    197,000    -0.25    Qf7 Bxc5 Bxc5 a3 Rb7 b4 axb4 axb4
  7/17    00:00          12,186    199,000    -0.30    Qf7 Bxc5 Bxc5 a3 Kg7 Qd3 Qd7 a4 Bxd4 exd4 Rxc2+ Rxc2
  8/19    00:00          31,148    432,000    -0.56    Qf7 Bxc5 Bxc5 a3 Rf8 b4 axb4 axb4 Bxd4 exd4 Qe7
  9/22+    00:00          54,506    619,000    -0.50    Qf7
  9/22    00:00          59,886    650,000    -0.39    Qf7 Bxc5 Bxc5 a3 Qe8 Qxe8+ Rxe8 a4 Kg7 f4 Rbc8
 10/22-    00:00          83,105    814,000    -0.45    Qf7 Bxc5
 10/22    00:00         192,605    977,000    -0.38    Qe8 Bxc5 Qxb5 Nxb5 Bxc5 f4 Rd8 Nd4 Re8 a4 Rbc8 h3 Kg7
 11/22+    00:00         268,105    1,094,000    -0.32    Qe8
 11/22    00:00         277,528    1,088,000    -0.35    Qe8 Bxc5 Bxc5 a3 Qxb5 Nxb5 Rd8 Nd4 Re8 a4 Kg7 Nc6 Rbc8
 11/24    00:00         327,274    1,136,000    -0.34    Qxb5 Nxb5 Rd8 Nd4 Re8 Re1 Kg7 Rd1 Rbc8 Rdc1 Rc7 Nb5 Rd7 Nd6 Bxd6 exd6 Rxd6 Bxc5 bxc5 Rxc5
 12/30    00:00         409,633    1,211,000    -0.34    Qxb5 Nxb5 Rd8 Nd4 Re8 Re1 Kg7 Rd1 Rbc8 Rdc1 Rc7 Nb5 Rd7 Nd6 Bxd6 exd6 Rxd6 Bxc5 bxc5 Rxc5
 13/34-    00:00         706,533    1,481,000    -0.40    Qxb5 Nxb5
 13/36+    00:00       1,418,348    1,686,000    -0.28    Qf7
 13/36    00:01       1,755,089    1,715,000    -0.26    Qf7 Bb2 Rb7 Qe2 Qd7 Ba3 Rbb8 f4 Re8 Qf1 Rbc8 Qb5 Qc7 Bxc5 Bxc5 a3
 14/36-    00:01       2,046,352    1,741,000    -0.33    Qf7 Bb2
 14/36    00:01       2,408,045    1,771,000    -0.36    Qf7 Bb2 Qe8 Qxe8 Rxe8 Rc3 Rbc8 Nb5 Red8 R3c2 Rb8 Rd1 Rbc8 Bd4 Nd7 Rdc1 Rxc2+ Rxc2
 15/36    00:01       2,819,158    1,799,000    -0.34    Qf7 Bb2 Qe8 Qxe8 Rxe8 Rd2 Rbc8 Nb5 Red8 Bd4 Kf7 Rdc2 Rd7 h3 g5 g4 Kg6
 16/36    00:02       5,337,485    1,933,000    -0.37    Qf7 Bb2 Qe8 Qxe8 Rxe8 Rd2 Rbc8 Nb5 Red8 Bd4 Kf7 Rdc2 Rd7 h3 g5 g4 Kg6
 16/39    00:03       6,884,119    1,973,000    -0.37    Qf7 Bb2 Qe8 Qxe8 Rxe8 Rd2 Rbc8 Nb5 Red8 Bd4 Kf7 Rdc2 Rd7 h3 g5 g4 Kg6


Sorry but your silly crusade about computers not being upto the human mind standards is just ridiculous.
Every super GMs have admitted the obvious.

You seem to keep hanging onto this really old stupid idea that computers only calculate with brute force... pls update your knowledge.
There are many, many heuristics and "knowledge/understanding/intuition" built into the machines which GUIDE! the calculation (just as happens with human mind!)

Saying a computer not studying Petrosians games is incorrect... the super GMs who have studied and learned from Petrosian built that knowledge into engines as well
(eg grabbing space, not making weaknesses, exchange sacs for positional compensation, computers do all this very well)

You are probably using old bad engines, and think they are all alike.

Please stop this crusade and rather use the computers to learn .. and also make more interesting posts like this to (minus the computer bashing)

I think you are missing the point -- it's not that the computer doesn't like or "can't understand" ...Qf7 -- it's just that it thinks ...Qxb5 is fine as well and can't see the danger in the ensuing endgame.

Pfren's example does a good job pointing out where engines can fail -- sometimes there is a long term sequence of moves that is intuitive to humans but would be too lengthy for a computer to fully calculate move by move because the plan may take a long time.

Nonetheless, while an engine should have some guidance, I think most of the time, even in closed positions or endings, they can have some idea of the correct evaluation and plan -- despite how mechanical their approach is to the game. I see a lot of times an engine recommending pawn breaks, playing moves like Nc3-b1, almost human looking moves -- its reasoning is probably different from that of humans, but it often still hits upon good sequences of moves. And in tactical positions -- while computers can still falter there, in particular by underestimating a kingside build up due to horizon effect at first (although a lot of these types of positions actually have positional features, like a lot of space on the kingside, not necessarily pure tactical positions), I would say they are much more trustworthy than humans in such positions!

Lol bro are you a nerd gamer or something? It seems you really love your engines..

By the way, in that Qc1 position firebrandx, are you suggesting white is simply winning after 15...b5? Somehow it just seems like forcing a win seems sort of far-fetched. Of course, if it were just a normal example I would probably say "houdini could probably find a defense" but, I guess houdini failed

However, are you sure you didn't just win because you continued to outplay your opponent as a centaur?

Ah, I remember you, Monster. Attacking my blogs and not allowing me to respond to you

The only problem with computer engines are that they are rather materialistic. Also it depends on how deep you actually set it to,

Yah, I'm amused by people who tell me that computers aren't very good. You know, actually, they've gotten really good. I'm quite impressed by Houdini, its noticeably superior to Fritz, and Fritz was quite skilled.

Wafflemaster, I am not a supergm.  I am just someone who has played engines for many years while traveling.  Sorry you and some believe this game to be illigitimate, but it is 1 of a handfull of cases where the engine did blunder which can happen from time to time.  This game I posted is indeed a game I played against houdini 1.5a on the hardware I cited.  Anyone's opinion to the contrary is meaningless.  As I have stated, the engine normally gives me a thrashing as it would almost anyone.  The important thing to realize is that my play showed a clear plan throughout and any legititmate chess player could have executed that plan.  It should also be noted that moves that some are saying "houdini would never make" ARE moves the engine made.  Facts are facts regardless of opinion.  I am not a fraud, neither am I someone to look up to for beating an engine...I am just someone who enjoys chess just like you do :)  That fact is engines do not like closed positions like the one in the game where there are very limited moves and it cannot use it's vast calculation abilities.  Engines can also get into time trouble while calculating in these positions because it searches many lines which will never be played while the human can analyze only a few lines which WILL get played by force.  All of my moves after 19....Ba5 are forcing.  The knight on f4 is blocking the black pieces from any counterplay and there is no time to open the center since the black king is in deep sh*t :)

I'm not really sure what point you are trying to make. It took me less than ten seconds to feel confident that I knew the correct move, 42.Kh2. Then I checked it with Houdini, who strongly agrees. You seem to be implying that this is a difficult position which will 'break' Houdini. Is there something wrong with Kh2? It looks pretty good to me. I can't imagine a GM wasting ten minutes here. Thinking about what, exactly?

I'm sure everyone has different thinking methods, and I'm more intuitive than most, but this position seems quite basic. The pawns on a7 and b6 are clearly going to pin down black's pieces. Meanwhile, White has free reign to wreack havoc on the kingside. In this particular position, Kh2 seems extremely obvious.

Why would I spend weeks analyzing this position??? There is no need to calculate variations here, because Kh2 is the only candidate move.

Yes, in a blitz game I'd play kh2 fairly immediately, it seems like the only move that gives white a serious winning edge, so, is the point that the move actually isn't kh2? Because that would be surprising.

I certainly agree. I've never taken lessons, and never studied master games. My first teacher was Battlechess, then the Kasparov talking board, then Winboard, Fritz, Rybka, and now Houdini is my coach. The only truly valuable input I've received from a human was when Yasser Seirawan wrote Winning Chess Tactics and Winning Chess Strategy, which explained to me the basic themes that I needed to know. Usually, when people are talking about a chess game, I just tune them out and wish they'd put their position into the computer.

I feel like master games are a waste of time. They are vaguely interesting for historical purposes, and their display of skill is impressive, but they aren't anywhere near as educational as using the engine to explore variations on your own.

I think people who criticize the use of computers just don't know how to use them. You don't just sit there and read the little output line and play whatever gives the highest score. Instead, you play around with all the different ideas that come to you, and the computer gives you immediate feedback. Its like having a GM in the room, and he tells you what he will play in any position.