Chess Championship: Humans vs. Computer

by Lev Dymchenko

05/04/2003 10:41 PM

During the last 10 years
computers penetrated into various spheres of human life. In the today’s article
we will try to find out how well computers can play chess and if it would be
correct to say that artificial intelligence is superior to human mind.

It’s now impossible to imagine our civilization without computers: they are virtually everywhere. Chess is one more field where computers do quite well.
With the arrival of chess programs and the growth of computers’ calculating
capacities, they started talking about the dawn of Artificial Intelligence.
That’s what we are going to discuss here: the computer in chess or computer
chess.

Clone Attack

In the last decade computers became an indispensable tool in chess masters training. It was becoming hard to operate
huge databases of opening variations and chess games without personal
computers. Today, a PC with a personal openings library is one of the
attributes of a chess master. The computer was gaining its ground slowly, but
surely. Kramnik is said to have started using computers the last of all super
grandmasters. When his more “advanced” colleagues were all into chess software,
he was still scribbling variants down into an old copybook. But eventually he
couldn’t resist the progress, either.

With the arrival of chess software and powerful PCs, chess openings were studied more effectively. You could make the computer analyze a position and
use the results of the analysis later. A lot of popular openings are now traced
to deep endgames. The theory of openings reaches dozens of moves into the game
and a novelty on a twentieth move is no surprise today, unlike the novelty on
the tenth move, where everything seems to be studied inside out already.
However, we will see that it is not so true yet.

Many known players are criticized for winning games by using computers at home and thus having wider knowledge of game starts. They often use new
untypical moves in the beginning of the game. Kasparov is a player of the kind.
On the one hand, his style means getting advantage right from the start. On the
other hand, he gave cause for this criticism himself having pioneered the use
of computers in training.

But the problems of analytical work of professional players are not so interesting for ordinary chess fans. It’s quite more fascinating to watch a
battle between chess programs and a man. And that’s what we are going to talk
about.

There have been two remarkable events in the last two months: two Man vs. Computer matches with much publicity and a round sum of prize money. First, the human world champion Vladimir Kramnik met Fritz, the
world champion among chess programs, in an eight-game match. Then it was the
turn of Kasparov, who is the leader of the FIDE rating list, to play six games
against Junior program. Junior had also been a world
champion earlier. These two chess programs haven’t met with each other so we
can’t say which one is stronger until the next computer world championship. We
know that these two programs play different chess, though. Fritz is inclined to
positional game, while Junior likes attacking. It was said that Fritz resembled
Kramnik’s style, while Junior – Kasparov’s. So, the grandmasters played against
their electronic incarnations.

As our web-site is about computers, I have to write specifically about the successes of the electronic brain, but I suppose that the reader knows the
names of Kasparov and Kramnik. If not, there is a link at the end of the
article to a literary description of the inhabitants of the chess kingdom

Man vs. Computer: Match History

So, let’s see how flesh-and-bone chess masters fought against artificial intelligence. We will start out with the sensational match Kasparov played
against Deep Blue, a chess super-computer specially built by IBM Corporation.

This happened in 1997. Personal computers were too weak then to compete with human grandmasters, so IBM made a giant multi-processor chess computer. The
first version of the machine was far from perfect and was routed completely by
Kasparov. But the second version managed to beat a human world champion in a
new match. This provoked a big commotion and drew wide publicity. So what
happened then? Let’s figure it out.

There were six games. Kasparov played White in the first one. Kasparov chose to use “queer” openings in almost all the games so that the computer couldn’t
use its openings library. He made strange waiting moves: you don’t play like
that against humans because this is actually a loss of initiative. Anyway,
Kasparov got an advantage in this game, profited by positional mistakes of the
program and eventually won.

The second game was Kasparov’s Black. The computer was surprisingly good this time. It made a series of strong planned positional moves and threatened
Black’s fortifications. We will discuss the operation principles of chess
programs later in this article. For now let’s state that programs are not very
good at positional playing. They process separate moves, not move sequences or
plans. Kasparov tried to save the day. He sacrificed two pawns and got some
chances to attack the White king and achieve a perpetual check. And there was a
miracle: the computer took 15 minutes to consider the next move, although it
had made all its previous moves in three minutes exactly. After that, it declined
the sacrifice! All chess programs are in for material things. They don’t see
positional threats, but they can count up pawns. That’s why one of the common
methods against programs is to sacrifice for a positional compensation, which
cannot be evaluated correctly by the program. Frustrated and confused, Kasparov
moved his pieces around for a while; the computer reinforced its position,
created threats, but somehow allowed Kasparov to check its king and save the
game through a perpetual check. But Kasparov resigned. He immediately learned
from the commentators that he had given up a drawn position – he did have a
perpetual check!

Kasparov said that he hadn’t believed the computer to choose a perpetual check variant when it had declined this very variant a few moves before,
refusing two (!) pawns.

In the next three games Kasparov was pressing on the program, but it always slipped to a draw by means of tactical nuances.

In the last, sixth game, Kasparov played Black and chose a variation where White sacrifices a knight for a sweeping attack. The computer sacrificed the knight
and won easily. After the game Kasparov reasoned that computers didn’t usually
give material for initiative, which they couldn’t evaluate correctly. Some
commentators said that this variant could have been put into the openings
library of the machine. Kasparov somehow dismissed this option. Some said
Kasparov just confused the order of moves. That is, he made a second move of
the variant before the first. This can happen to a chess master, especially, as
Kasparov was quite tired after the previous games.

So, the computer scored 3.5-2.5 in the match. IBM puffed up this success immodestly. I even recall some movie beginning with something like “This was
the year when Computer defeated Man at chess”, a kind of apocalypse. IBM’s
stock went up significantly after the match.

The strange behavior of the computer in the second game made Kasparov’s team doubt the fair play of IBM. They suggested that IBM had invited a grandmaster
to prompt certain strategic moments to the chess program. That is, the helper –
a human – told Junior not to accept the sacrifice of two pawns. This would have
brought the computer under attack, which it couldn’t have calculated to the end
and evaluate right. But when the computer gained a critical advantage, the
advisor lost concentration and allowed Kasparov’s perpetual checking.

Well, the company was very interested in the match result. The computer had lost the first match hands-up, and IBM had put a lot of money into the project,
so there shouldn’t have been any misfire anymore. Another fact: the computer
was dismantled right after the match.

But on the other hand, we could also suspect Kasparov in losing the match on agreement. Why did he resign in the second game without even trying to give a
few more checks? Why did he choose a risky variant in the last most important
game? There are more questions than answers.

The multi-processor computer was also rumored to have hung up a few times, so that it had to be re-booted. The technical staff were
bringing it to its senses like a boxer between rounds. I may venture a theory
that the computer gave out a series of good moves (maybe accidentally,
sometimes a poor player can do a thing or two) and then hung up. And it takes a
while for such a machine to re-boot. And after the re-boot, it wasn’t into the
game yet and played “safe”, avoiding sharp variants, before all its units were
turned on. Anyway, the program did hang up to Kasparov’s astonishment. We will
discuss this game in detail below when talking about chess program algorithms.

Of course, IBM had to dismantle such a reboot-inclined computer as soon as possible. Maybe they would have done this after the first match, but the proud
name of the corporation and the money already spent stopped them. They saw that
it was a hard job to make chess computers and were lucky to wind their project
up with a loud bang.

If we take pure chess into account, the AI showed nothing extraordinary for a world champion: the last game was an evident blunder of the tired Kasparov, the second should be considered a draw, as the
final position was a draw. In other drawn games, the computer showed absolutely
nothing, just escaping to the safe side. It’s even strange that a specialized
chess computer should play so weak.

I’m far from suspecting IBM in “unfair play” as Kasparov does, but still it doesn’t seem right when an average result and escapes to a draw are presented
as a victory of the artificial intelligence over the human one. It seems like
IBM-compatible computers have made a career to the ranks of IBM top managers.
Of course, this result is a big step ahead in computer chess, but more steps
are required for a computer to reach the level of the human world champion.

Newest Matches

Processors for PCs and small multi-processor systems were becoming more powerful. So, chess programs, which are also constantly polished off, can now
show grandmaster-level play without running on a super-computer.

First, let’s discuss the earlier match – Kramnik vs. Fritz. It was eight games long and somewhat split up into two halves. In the first four games,
Kramnik was accurately following a well-thought anti-computer strategy. He
exchanged the queens and went for a complex endgame with most pieces still on
the board. There were fewer tactical variants there, while the big number of
pieces prevented Fritz from considering variants deeply enough to see the
outcome of its decisions. For example, the computer can consider a variant 6
moves ahead, but the outcome of the variant will happen on move 10. In this
case, the computer can make a weak move.

When there are no queens, time slows down in chess. Everything goes on less fast and it’s often necessary to make a number of simple, but plan-bound, moves
to achieve success. To cut it short, Kramnik won two games, and in one game the
computer managed to draw thanks to Kramnik’s inaccurate use of his advantage.
In the first game of the match Fritz had a better endgame, but to no effect: it
just couldn’t play it right.

The second half of the match was quite contrary. Kramnik must have been bored to play and win again and he lost a knight and resigned immediately. In
the next game he was going to attack. He sacrificed his bishop for a pair of
pawns and drew the Black king to the fore. But he didn’t manage to end the
attack. Computers are strong in calculations-heavy positions, while Kramnik
must have made a mistake somewhere. Nevertheless, the initiative turned to be
enough to get an endgame with some chances for a draw. He could build a castle
when the Black queen couldn’t pass a white rook’s defense – the computer can’t
see such things and allows them. But Kramnik followed a kind of tradition of
human players in such matches and resigned in a nearly drawn position. After
that he lost all his vigor. The last two games were insipid and ended in draws.

I could try to represent the match as a boxing event. Kramnik had his opponent on the floor for three rounds. In the next round he got pressed
against the ropes, missed a simple blow and went down. In the next round he
rushed to attack and nearly knocked the computer down, but overlooked a
counter-attack. Although he could stand on his legs still, he preferred to
report a technical knockdown. The last two rounds were spent in a clinch. In
boxing, Kramnik would have won by points: the iron opponent was not ready for
the fight. 

This was an agreed match meaning that both parties just agreed to play it. If there had been any elimination contest, this version of Fritz wouldn’t have
got to meet Kramnik. It has an evident weak point: complex endgames. The two
games Fritz won were Kramnik’s unprovoked mistakes, rather than Fritz’s strong
play. Those grandmasters that are at the bottom of the chess Olympus,
wouldn’t just play chess to Fritz, they would compete: win a game and then wait
for a mistake of the computer.

In fact, an earlier version of this chess program played with grandmasters at a super-tournament in Dortmund.
At first, humans didn’t know what to do against the machine,
they just tried to defend themselves. Then one of them won a game and thus
uncovered a weak spot of the program. It was weak in closed positions. After
that, they all tore the machine to pieces every game.

Kasparov’s Return Match

Now, we have come to the last event: Kasparov vs. Junior. Although this program has nothing to do with IBM’s creation, this match was considered to be
a kind of rematch.

Game 1. Kasparov won the first game. This time he didn’t go for irregular openings and stuck to classic
schemes. Although the chess program could use its openings library, it couldn’t
solve the problem. Junior had a worse position right after the beginning and
Kasparov had no problems winning the game.

Game 2. Kasparov (Black) played a sharp Sicilian defense and got a better position. At one moment Junior agreed
to a sacrifice for a strong attack, but got a lost position. Kasparov was too
hasty, though, and gave the computer an opportunity to draw the game by
perpetual check.

Game 3. Kasparov was attacking again. But the position was opened and the program stood its ground. Anyway,
the machine had nothing more than a draw. And here Garry forgot the rules when
you play against a computer: he was too engaged into the attack and left his
king undefended. To make the things worse, he missed an opportunity to announce
a perpetual check and was checkmated the next move. Human attention often tends
to concentrate on one thing, this time it was the attack. The computer, on the
other hand, has no such problems: it can easily switch from attack to defense.
I play chess with computers quite a lot and it’s a usual thing for the computer
to threaten my king with its queen and the next move take the queen to the
opposite side of the board to help its own king. Such a defensive move can be
easily overlooked by a human player.

Game 4. Kasparov wanted nothing more than a draw in this game (he played Black) and got a rook endgame without
a pawn. Junior agreed to this variant, not realizing it was a dead draw, in
spite of his advantage in pawns. So, they drew the game.

Game 5. Kasparov was ready to attack, but quite unexpectedly Junior sacrificed (!) a bishop on square h2 for
an attack of his own. Kasparov took a long time to think over his next move and
declined the sacrifice. Later he explained that he could lose such a
tactics-heavy position to the fast calculating program. So, the game was drawn
by repetition.

This game is considered to be a success of the artificial intelligence. A computer exchanged material for initiative. Only humans were supposed to be
able to play like that. But we shouldn’t be hasty with our judgment. The
sacrifice occurred in early opening and may have been put into the program
before the beginning of the match. We shouldn’t forget that a few strong chess
masters were involved into the development of the program and they do know what
openings Kasparov likes to play.

Game 6. In the last game Kasparov once again outplayed Junior, got a complex ending with a big advantage. Some
commentators said it was a technical win. Kasparov preferred to draw it,
though. It was more important for him not to lose the match than win it.

Once again, let’s picture it in boxing terms. The computer was beaten the first round and was very lucky to escape in the second. In the third round it
was cool-headed in defense and punished his over-enthusiastic opponent in a
counter-attack. The next round, it was Junior that couldn’t break Kasparov’s
defense. The fifth round, Kasparov was ready to show his best, but got a nice
punch somewhere at the liver. Before he made up his mind to venture an attack,
the round was over. The last round, Kasparov was sharp and accurate, but took
no chances and let the opponent remain on his feet.

So, the match ended in a draw. But did Junior play stronger than Deep Blue? Deep Blue had a higher computational power, but in the last seven years PCs
have been growing, too. Chess algorithms have also improved over time and thus
make up for a lack of power.

Let’s consider the overall result of Man vs. Machine matches. The computer won 5 games, lost 4 and drew 11. There is no decisive victory among the five
wins, only evident mistakes on the grandmasters’ part. As for losses, they are
real defeats, exposing serious flaws in play. So, it’s quite unclear how a
computer can win without a human team behind it.

You know what kind of a player computer looks to me? There are people who play chess for money at cafes and public parks. They have a lot of learned
opening variations and a number of standard traps. They try to confuse the
opponent with provocative moves and catch him on small tactics, but they don’t
really understand chess.

So, how comes the AI plays just like a chess cheater?

Key Principles of Chess Programs

So, how do the chess programs work and manage to beat the world chess champions? Every chess algorithm is based on a simple search of all
possibilities and selection of the best among them. So, we have a position for
the computer to make a move. It studies all the possible moves one by one.
Every move means a new position where the computer again studies all the
possible moves and so on, to a certain depth.

In the end, it evaluates the positions and chooses the best move. In the case shown on the diagram, it is going to choose the first variant, as any of
the opponent’s moves would give the computer a small advantage.

How does the program evaluate a position, rate it with a number? It bases its decision on factors, which can be obtained analytically. First of all, it’s
the number of pieces on the board. Then, there is the activity of the pieces,
which is the number of their possible moves, control over the center, advantage
in space, and the number of squares under control. There are also simple
factors concerning pawns: double pawns, weak isolated pawns, passed and
retarded pawns. The safety of the king can be calculated according to the
number of squares around it controlled by friendly pieces.

But you should bear in mind that these factors are calculated purely mathematically, mechanically. A computer program can be viewed as a book of
rules in this respect. And this book reads: a double pawn is a weakness. But as
for other factors, they can’t be easily calculated: an isolated pawn can be
either a strong or a weak point of the position, depending on the given
situation, so that all the variants should be taken into account. As for pieces
activity, it is also not always clear if this activity is real or fake, as the
computer simply states the availability of a big number of theoretical moves.

Considering all this, it’s no wonder chess programs are strong in “computed” positions, where they can usually gain material advantage by tactical
maneuvers. Programs just look through all the variants and choose one where
they have more pawns. Human players may turn this against the program itself,
like in the above-described games.

Now, let’s try to estimate the calculation depth achieved by using this simple searching method in modern PCs.

Imagine that we have a 1GHz CPU and 100 seconds per move. That is, the program has 100 billion clock cycles per move. Let’s take that one position
requires 1000 cycles to be evaluated. This is an approximate, and even lowered,
number. We take into account certain optimizations of the search algorithm. For
example, when a position can be achieved by different moves, it’s only
evaluated once. Of course, biggest contribution is made by the variants on the
last level of the variant tree.

Now, let’s estimate the number of possible moves in a typical chess position. When most pieces are on the board, it’s about 30-60. For simplicity,
let’s think it’s the square root of 1000.

100 billion cycles divided by 1000 cycles per move = the number of positions at the deepest calculation level. Find the logarithm of the number of possible
moves. The result is 5-6 half moves. That is, about three full moves. This is
not great. The exponential nature of chess moves search kills the performance.

But maybe we can delve deeper using a faster processor and more time? Here is the table showing numbers for 1000 seconds per move instead of 100 seconds.

CPU	Calculation depth (half moves)
1GHz	6
1000GHz	8
1000THz	10

It’s clear that chess cannot be thoroughly calculated. The only hope for chess programs lies in optimization of their algorithms.

Of course, even candidates for master’s title can calculate 5 moves ahead. So, what did they do to this searching algorithm that it proved capable of
competing with the world’s champions?

Variant Tree Optimization

So, the developers re-shaped this variant tree: somewhere they trimmed it up, elsewhere – made it longer. One of the ideas is to cut a variant short when
any of the two sides gets a critical advantage. This is done to save time and
trouble calculating the moves when the queens of both players take defended
pawns from one another. However, things are not so simple here. This pawns
exchange can be the beginning of a checkmate combination, after which the
material and other positional factors are of no importance any more. That’s why
this “cutting- short” should happen not at once but a few moves deeper.

The second idea goes like “if we can’t calculate all the variants, we will consider only important variants but do it in great detail”. The problem is
that it is not quite clear which variants are important. So, first of all, the
computer counts up all captures and checks in every position. So that it
doesn’t stop considering a variant a few steps short of a checkmate.

Chess programs like Fritz or Chess Master have an option of showing a variant they are now considering and its depth. Usually, this depth is 12-14
half moves. But in fact, the computer has considered a number of variants far
deeper – by tens of moves. Thus, the computer can find a checkmate in 100 moves
if this is a forced variant (that is, a check at every move). And it may not
find a checkmate in 10 moves if the variant consists of “silent” moves, without
captures and checks.

It’s the optimization of a chess program rather than the evaluation algorithm that affects the playing manner. There is a kind of balance between
calculations going wide and deep. If the program dismisses variants too fast it
may miss hidden tactical opportunities, but spend more time evaluating
positions.

Now, let’s discuss a few examples of the chess programs work. We will consider an unexpected tactical blunder of the Fritz program, but before that
take a look at a few moments from the Kasparov – Deep Blue match.

Chess Programs Thinking

Second match between Kasparov and Deep Blue.

Before move 46. Ra6?

The computer played Ra6. Instead of going to this absolutely won ending:

This ending is hopeless for the Black.

They have a lot of weak spots and a great number of passive pieces.

...it gave Kasparov an opportunity to set a perpetual checkmate by 45...Qe3 46.Q:d6, Re8! 47.h4 And there’s no way to escape
the check any more!

The White move now. No protection against
the perpetual check.

How come that a chess program commits such childish mistakes? It’s quite clear even to an average player that it’s better to go to a technically won
position than put your king under the threat of a perpetual check.

Still, most chess programs, even after long thinking, move their rook to a6. They can’t calculate this endgame to the win. And they can’t evaluate it right:
the White have a more active rook and an isolated passed pawn, but the Black
have a protected passed pawn. The bishops have about the same number of
available squares. The program can win this end (if pressed to it) by slow
enhancement of its pieces’ positions. During the realization of the advantage,
the Black may become more active, create more passed pawns – all this cannot be
properly evaluated by the program.

And in the variant with the perpetual check, the program has a big material advantage – a bishop – and a lot of positional pluses. As for the check itself,
the program can only detect it when things get worse in all positions or if
there is a triple repetition of a certain position. And you can wait for dozens
of moves before this triple repetition occurs: the white king can go to the
queen’s side, come back again, try to hide behind the bishop, go somewhere
else. There is an ocean of variants, but they all point to a draw. The computer
can’t understand it. I was trying to run away from Fritz program for 30 moves
in this position.

In this situation the program may calculate certain variants dozens of moves
deep,

and still see no perpetual check indicating a significant advantage of the
White instead.

So, we shouldn’t be surprised at the computer’s choice. By the way, all programs I investigated this position with used to hang up and glitch a lot.

Kasparov thought it strange that the program prevented the perpetual check threat in some earlier games of the match. But you can load this position into
a modern chess program and see that the difference in evaluation of the move
made by Deep Blue and the move that leads to the perpetual check makes around
0.1. This difference lies within the fluctuations caused by different program
settings – a slight change in weighting coefficients of positional factors can
be the reason here.

On the other hand, the perpetual check threat was more evident in the previous games – the position repeated three times much sooner. Modern programs
usually see this kind of repetition.

Now that we know that programs are prone to cut short all variants with big material losses, we may suppose that there are positions chess programs just
can’t evaluate correctly.

Here is an example (from grandmasters’ match):

The Black move.

I don’t claim a thorough analysis of this complex position, but it’s clear that the White must look for an escape, notwithstanding their extra pawn. The
Black move the rook to e4 and the White can’t take it because the Bc4 would be
taken in this case. Bc3 is responded by a sacrifice of the knight on g5, while
Nc6 can be answered by Qe8! The position is full of tactics. You might think
chess programs would feel at ease here. You would be wrong!

Fritz7, a recent world champion among chess programs, takes an hour on a 1.5GHz CPU to realize that the White are in no good
situation at all. And it takes the program several hours to evaluate the
position correctly. It just doesn’t see “silent” moves and so on.

As for another, more powerful programs, like Fritz8, they are not much stronger players than Fritz7. By the way, old ChessMaster 5000, a 1996-year
release, understands this position quicker than Fritz7. It has got a lot of
other problems, though.

Maybe more powerful processors would be of some help in these positions? Yes, this is possible. On the other hand, in the previous position the program
has to consider tens of moves. This would kill the performance growth
altogether.

Many amateur players like to put a game into the computer for analysis. But we should be careful about such evaluations. The program may say it’s only a
half a pawn advantage for an absolutely won position, or cry advantage in a
dead draw. Here, the high rating of a program shouldn’t deceive you: it doesn’t
necessarily mean the program understands positional playing correctly.

Having played for a while against chess programs, I came to my own recipe: try to make the best move possible in every situation. When you just make a
move that looks like good, without any plan in mind, it may bring you to
trouble against the computer.

Chess Programs Playing

So, chess programs are based on a simple and rather slow algorithm. How do they manage to beat human grandmasters? What are the tricks of a chess cheater?

A chess cheat has an openings note in his left pocket with trap variants marked red.

Yeah, every chess program has a huge openings library to consult with, while a man has none. I might divide the human memory into internal and external. A
paper sheet (a book) is a kind of CD, while human eyes are not far from CD-ROM
heads. Thus, a man is not only deprived of some part of his memory, but of a
specialized part. This is all right when humans compete with each other, but
not fair when two kinds of intelligence are involved.

There is an endgames book in the other pocket of the chess cheater.

Having found no other way to make the program good at endgame, program developers started feeding them databases of common endgames. Without this, the
program would be at a loss even in a simplest pawn endgame. That’s the reason
for chess programs being much larger today than before – they take whole CDs.
It turns out man plays against his own knowledge base
rather than against an AI. This is good for training young chess players, but
we can’t call it an achievement of AI.

Cheaters seldom work alone

The above-described matches were played between a man and a multi-processor machine. The processors were prompting to each other and exchanging ideas. This
doesn’t seem fair. You might go and load a huge chess program into all
Internet-connected computers and make it play against a single human. Wouldn’t
that be right? A man must play against his own desktop PC.

Cheaters catch the opponents

Cheaters like to let their opponent win, grow heated and then beat him. Chess programs also may be tweaked during a match. The technicians change
program settings and as a result create a completely different chess player. I
wonder if they would be happy to suddenly get Kramnik as an opponent instead of
Kasparov, with whom they were preparing to play. 

The chess cheater has another chessboard under the table to check a few variants.

Chess programs have a lot of memory at hand. It’s like they have a million of chessboards to make moves on. And the human has none. If I were Kasparov or
Kramnik, I would come to the match against the computer with my own board and
played all variants on it. The PC can’t see, you know. Who said you can’t touch
the pieces when playing against the machine? There is no such rule.

It’s like you play “blind” chess against a cheater. You are trying to figure it all out with closed eyes, while the cheater sees everything. 

Cheaters win from blunders.

All the games the computer won in the above-described matches were won due to blunders of the human opponents. They blundered
everything: a piece, a checkmate, a draw, an opening. The cheater can’t win
without that.

Conclusion

To good or bad, but artificial intelligence is still very far away from the human brain. The example of chess programs proves it.