plik

FIDE Surveys – Miguel Illescas

Miguel Illescas:
Working with computers: does
the machine fool us?

Farewell to glamour?

In recent years, increasingly powerful
computers and programs are already able to
unravel the mysteries of most positions in just
a few seconds. This has led to the trend of
following the major tournaments live
broadcasts with the engine connected, which
judges implacable every movement made by
the grandmasters. The fan, standing on the
shoulders of giants, thinks he is a big shot and
criticizes Carlsen strongly for missing a
tactical blow, or Karjakin, for not being able to
see it. “What a patzer! It was so easy…” He
shouts in disbelief, almost outraged.
For instance, let’s take the already famous 20
... Nf2 that could have given a beautiful draw
to Karjakin in the tenth game, which he ended
up losing. It was tremendously criticized by all
and sundry, even by grandmasters. Was it
really that easy?

Carlsen : Karjakin
WCC New York (10) 2016

XIIIIIIIIY
9r+-+-trk+0
9+p+-+-zp-0
9-+pzppwqnzp0
9zp-+-zp-+-0
9P+-+P+-+0
9+-zPPsN-zPn0
9-zP-sNQzP-zP0
9tR-+-+R+K0
xiiiiiiiiy

In this position, the game continued with 20
...d5 21.Qh5 Ng5 22.h4 Nf3 23.Nf3 Qf3 24.
Qf3 Rf3 25.Kg2, a very natural sequence of
moves that led to a slightly better endgame for
white, which Carlsen went on to win in good
fashion.
While Karjakin was thinking about his
twentieth move, the audience on the Internet

held his breath at the verdict of the engine,
which offered a dead equal evaluation of 0.00,
a consequence of the repetition of moves that
occurs in the curious line

20...Nf2! 21.Kg2 Nh4!

XIIIIIIIIY
9r+-+-trk+0
9+p+-+-zp-0
9-+pzppwq-zp0
9zp-+-zp-+-0
9P+-+P+-sn0
9+-zPPsN-zP-0
9-zP-sNQsnKzP0
9tR-+-+R+-0
xiiiiiiiiy

22.Kg1 Nh3! 23.Kh1 Nf2! 24.Kg1 Nh3 with
equality!

But Karjakin finally played the natural 20...d5,
and before we had recovered from the
excitement Carlsen answered quickly with the
incisive 21.Qh5 (instead of the prudent 21.f3),
which still allowed the capture on f2, which
again was correct. Social networks set on fire!
The audience was pulling their hair out. How
is it possible that the two finalists of the World
Championship do not see something so
obvious? Could be read on the net.
I will rhetorically repeat the question. Was it
really that obvious? How to estimate the
difficulty of a combination? Of course, the
machine sees the truth on the spot and the
lines that it offers do not admit any discussion
whatsoever. But let us review the analysis step
by step, for a better understanding.
20...Nf2!
Is an ugly move, which leaves the knight
pinned and apparently lost after the obvious
21.Kg2
Here, most masters stop the analysis, because
experience tells us that the knight is lost in
these cases, except for a miracle. And the
miracle exists as a sacrifice:
21...Nh4!!
based on the fact that after

FIDE Surveys – Miguel Illescas

22.gh4 Qg6

XIIIIIIIIY
9r+-+-trk+0
9+p+-+-zp-0
9-+pzpp+qzp0
9zp-+-zp-+-0
9P+-+P+-zP0
9+-zPPsN-+-0
9-zP-sNQsnKzP0
9tR-+-+R+-0
xiiiiiiiiy

the white king has nowhere to hide, since the
black knight deprives him of h1 and h3, and
white must return the material with interest.

Once seen, it seems easy, but there are no
referents in the grandmaster’s mind to imagine
the jump to h4, which leads to a very rare
combination. I took the trouble to search the
Mega2017 database for a similar pattern:
knight sacrifice on h4 to give a check on g6,
with a knight on f2. In almost seven million
games, do you know how many I found?
ZERO. Not a single time in the history of
chess competition there have been a similar
combination. Thus, it is not surprising that
Karjakin's mind almost automatically
dismissed the "ugly" catch in f2.
And what about the "missed" chance in the
next move? Logically, if neither of the two
contenders noticed the blow in the 20

, they

wouldn’t see it in the 21st. Because it was
harder, indeed!
After 20 ...d5 21.Qh5

XIIIIIIIIY
9r+-+-trk+0
9+p+-+-zp-0
9-+p+pwqnzp0
9zp-+pzp-+Q0
9P+-+P+-+0
9+-zPPsN-zPn0
9-zP-sN-zP-zP0
9tR-+-+R+K0
xiiiiiiii

21…Nf2! 22.Kg2
Black had to anticipate:

22...Qf7!
which is the only move. And the next one is
diabolical, since after:
23.Kg1

XIIIIIIIIY
9r+-+-trk+0
9+p+-+qzp-0
9-+p+p+nzp0
9zp-+pzp-+Q0
9P+-+P+-+0
9+-zPPsN-zP-0
9-zP-sN-sn-zP0
9tR-+-+RmK-0
xiiiiiiiiy

The only move that keeps the equality is the
quiet:
23...Qf6!! to answer 24.Ng4 (or 24.Nd1) with
24...Qg5!!
Since it turns out that after 25.Qg5 Nh3 the
knight is saved.

Hurray for the engine! But it is difficult to find
this kind of unusual moves when there are
reasonable alternatives like 20...d5.
The title "Farewell to glamour" is more than
justified. With the engine on, anyone believes
to know the truth and presumes to have the
authority to criticize nothing less than the
world champion. A pity, in my opinion, which
makes me feel a bit of nostalgia for those
years in which fans and the press spent several
months in order to refute a sacrifice by
Mikhail Tal.
Lastly, I would like to tell you briefly the
following anecdote. Karjakin has just lost the
tenth game and is obliged to deal with the TV
channels, first. The journalist approaches him
and asks in an excited tone: "What can you tell
me about the blunder on move 20 when you
pushed your pawn to d5?" He asks the
question as if he asks the striker how he could
miss an easy goal. Karjakin doesn’t
understand anything, and he cautiously replies,
"I’m not sure what do you mean, but I
certainly have not lost because of a move as
natural as 20 ... d5." The journalist didn’t
know what to answer.

FIDE Surveys – Miguel Illescas

MATHEMATICAL EVALUATION vs
PRACTICAL EVALUATION

To play chess well, and of course to analyze,
you need to be able to evaluate a position well.
To do this we must consider several factors,
the most obvious being the king’s safety and
the material balance. Almost from the
beginning, engines also computed the concept
of piece activity, which is essential, and
nowadays modern programs also take into
account countless factors, structural and
dynamic.
However, despite all the improvements,
engines are still much more reliable in the
move they come up with than in the evaluation
they offer.
It often happens that, when developing a line,
the engine corrects its initial evaluation by
clearing the calculation horizon.

How does the engine evaluate a position?

For each position the engine adds and
subtracts all good and bad from each side,
besides the material: if your king is not safe,
the evaluation drops, even if there is no
immediate attack. If your knight could be
installed in a weak enemy square, the
evaluation raises. With doubled pawns, it goes
down, again. And so, hundreds and hundreds
of factors. After narrowing down the
calculation to the limits allowed by the engine
and mixing all the factors into the shaker, the
program gives a score to each branch of the
line’s tree, giving a value in pawn units, with
their decimals.

Finally, the engine offers a numerical and
strictly mathematical evaluation of the
position. But this is not always enough for the
human player. What does an equality
assessment mean in a situation of tactical
madness, if we need to find several only and
strange moves, obvious to the engine but
almost impossible for a human made of flesh
and blood? Let’s remember again the great
Mikhail Tal, when he admitted that perhaps
his combination was not perfect, but it caused
insolvable problems to his opponent.

The idea then arises: the chess player needs a
practical assessment of the position.
Not just the acknowledgment that the position
is winning, but whether it is easy or not to
play, taking into account the probability of
either side to make a mistake.
Take as an example the position that occurred
in the most decisive game of the last World
Chess Championship, the third of the
tiebreaker. A critical moment was reached
after the 36th move by black. Carlsen had
sacrificed a pawn for great compensation.

Karjakin : Carlsen
WCC New York (play-off, 3) 2016

XIIIIIIIIY
9r+-+-+-mk0
9+-zp-+-zp-0
9-+-zp-+-zp0
9+P+P+-+-0
9-+LwqPzp-+0
9+-+-snP+P0
9-+-+Q+PmK0
9+-tR-+-+-0
xiiiiiiiiy

Here the engine evaluates the position as
equal, which may be strictly true from an
absolute point of view, but it is obvious that it
is much easier to play for black. White must
keep a stubborn defense, and even in the best
case scenario, exchanging the active enemy’s
major pieces, with that awful bishop they
could only dream of getting a draw.
As we know, Karjakin blundered and lost in
just three moves.
36.Qe1?
According to the engine, white resists with
passive moves such as 36.Bd3 or 36.Ba2.
36...Qb2 37.Bf1 Ra2 38.Rc7 Ra1 0:1.

Let's see another example of incorrect
evaluation from the human point of view, also
from the same match.

FIDE Surveys – Miguel Illescas

Carlsen : Karjakin
WCC New York (8) 2016

XIIIIIIIIY
9-+-+-+-+0
9+-+-+-mk-0
9Q+-+-+-zp0
9+-wq-sn-+-0
9-+-+P+-+0
9zp-+-+-zPP0
9-+-+-+LmK0
9+-+-+-+-0
xiiiiiiiiy

According to the engines working at great
depth, the position is more or less equal after
51.Qb7 or 51.h4. The truth is that defending
this position with white is a nightmare, and it
is clear that with that colossal knight and the
mighty pawn on a3 Black has no risk of
losing. The most likely outcome is for white to
end up winning, for instance, bringing the
white king to support his pawn after a series of
countless checks, a journey far beyond the
horizon of calculation of the engine.
As we know, the game ended quickly after the
blunder 51.Qe6? h5! 52.h4 a2!
And Carlsen resigned, because even after
capturing the pawn he remains in a mating net
53.Qa2 Ng4 54.Kh3 Qg1 55.Qb2Kg6!

We find the same evaluation problem in
endings like those with Q+P : Q. Many of
them are draws, but the strong side usually
ends up winning, because the defense needs an
extreme precision.

XIIIIIIIIY
9-+-+-wq-+0
9+-+-+-+-0
9-+-+-+-+0
9+-+-wQP+-0
9-+-+-+-+0
9+K+-+-+-0
9-+-+-mk-+0
9+-+-+-+-0
xiiiiiiiiy

Black to play.

Evaluation, according to the engine: 0.00 total
equality. But the vast majority of players
would be unable to save this ending!
It is a sequence of six only moves in order for
Black not to lose:
1...Kf3! 2.f6 Qg8!
2...Kg4 3.Qe7! Qb8 4.Qb4;
2...Qf7 3.Kb4 Kg4 4.Qe7 Qa2 5.f7 Qb2 6.Ka5
Qa2 7.Kb6 Qb3 8.Kc7 Qc4 9.Kd8 winning.
3.Kc3 Kg4! 4.Qe7 Kh5! 5.f7 Qg3 6.Kc4 Qf4!
And black follows the narrow path that leads
to the draw.

Would it be feasible for the programs to offer
a practical assessment?

The player that is oriented by the engine
would do better to look not only at the
evaluation of the best move, but also at the
next two or three best, and deepen the forced
lines to check if one of the sides is "on the
brink of the abyss”. In that case, even if the
engine announces equality, it is advisable to be
suspicious of the line.

FIDE Surveys – Miguel Illescas

10 TIPS ON HOW TO ANALYZE WITH THE ENGINE

1. Always look first at the position objectively, thinking for yourself, without being
influenced by the module.

2. Enter the move that you would play, even if it is not among the suggested ones, to see the
answers offered; develop the different lines in order to understand the real value of your
move.

3. Ask the program to show you the best three or four moves, to get a quick overview of the
position.

4. Do not take the engine evaluation for granted, for it is only a guidance. It is more important
to guess the practical assessment and whether the position is easy to play or not. Small
differences in the evaluation should be dismissed as a real advantage.

5. In general, do not let the engine think for more than two or three minutes per move. It is
necessary to go on by making the main move, especially if its evaluation is significantly
superior to the one that follows, and to develop the different lines.

6. Stop the analysis when both sides start to have a lot of moves of equivalent value. It doesn’t
make sense to spend a lot of time with the program in quiet positions.

7. In tactical situations, the evaluation is not reliable until the position stabilizes. We will
know that an evaluation is stable when several of the best moves offer a similar value, and so
does the opponent's response.

8. When the engine takes a long time to step up to the next level of depth, it is a good time to
move forward or stop the analysis and come up with a conclusion.

9. It is a good idea to write the lines, reviewing them slowly, trying to understand the meaning
of each move, until we are able to write down our evaluation into words or symbols,
discarding the rest of the moves.

10. Do not do anything else while analyzing, because the concentration is interrupted and we
could forget what we have learned in the process. It doesn’t work to leave the engine
analyzing to spend time on Facebook!