Media comments on chess

[Brendan O'Gorman]

Komodo, Hiarcs and Stockfish find the draw in an instant.

[Geoff Chandler]

Hi Brendan,

I've not tried it my Fritz 6 yet but I reckon it will never get it.

It does appear that Sir Roger needs to upgrade to a super-duper Carlos Fandango machine.

[IM Jack Rudd]

The thing about that specific position is that 1.cxb5+ and 1.bxa4 both allow a queen to roam free, which means there's probably a mate in reasonably short order. You don't need to analyse to anything like 50 moves deep to calculate that.

[Barry Sandercock]

What's the solution ? I don't get it.

[Geoff Chandler]

Hi Jack,

Most players will get the idea right away. Just keep moving the White King.

But will a computer not eval it +20+ (or whatever) for Black till it discovers it cannot win.

[Matthew Turner]

If I understand Turing's algorithm for Turochamp then it would easily find the correct solution. It is maybe that this methodology couldn't 'evaluate' the position, but does that really show the limitation for the computer when you are setting a somewhat arbitrary challenge of approximating a discrete variable to a continuous one?

[Roger de Coverly]

But will a computer not eval it +20+ (or whatever) for Black till it discovers it cannot win.

That depends on how clever were the programmers who wrote the rules for the engine to follow. These trick positions where making a capture will lose and not making a capture will draw have been around for many years. Certainly long enough that some programmers may have addressed the problem.

[Jonathan Rogers]

Hi Mathew.

(The diagram from the link)

White to play and draw.

8/p7/kpP5/qrp1b3/rpP2b2/pP4b1/P3K3/8 w - - 0 6

It is one of them positions that a reasonable player should see practically instantly how
to draw yet a computer will have to analyse for quite a while before it hits the 50 move rule.

...

Illegal position, since the f and h pawns would promote to light squared bishops.

Ok, back to work, Rogers ....

[Geoff Chandler]

Hi Roger,

As I said I've not tried it yet but I reckon my Fritz 6 will struggle and not get it instantly like Brendan's machines.
and I think this is the class of machine what Penrose is using. Why would he have a super-duper machine?

If they want to trip up a computer how about this one.

White to play and mate in two. (H. Hultberg 1944)

r3k3/1p1p2p1/2pP3p/8/8/5R2/PPPP4/4K2R w Kq - 0 1

Tell the machine that Black can castle.

Yet by playing 1.0-0 that proves that Black cannot play 1...0-0-0
because if White can castle how did the White Rook get to f3.

It must have been a promoted pawn and if a pawn promoted to a Rook
somewhere along the line the Black King or the a8 Rook must have moved.

By castling as White this proves the f3 Rook was a promoted piece and thus
proves that Black cannot castle. A computer cannot see that even though you
have said yes it can castle. 1. 0-0 proves it cannot.

So if White plays 1.Rhf1 then the f3 Rook could have come from a1 so Black can castle.
But if White plays 1.0-0 Black cannot reply 1...0-0-0 that is illegal though computers will play it.

[Geoff Chandler]

Hi Jonathan,

"Illegal position, since the f and h pawns would promote to light squared bishops."

The f and h pawns whilst on f2 and h2 can take a piece placed on g1 and under promote,
or at anytime during the game take from the f-file or the h-file to the g-file.

[IM Jack Rudd]

That diagram can represent one of three different legal positions and one illegal position, and whichever you say it is, the computer will give the right answer:

1. If you say neither side can castle, the computer will correctly tell you that 1.Rhf1 is mate in two.
2. If you say white can castle and black cannot, the computer will correctly tell you that 1.Rhf1 and 1.O-O are both mate in two.
3. If you say black can castle and white cannot, the computer will correctly tell you that there is no mate in two.
4. If you say both sides can castle (which cannot be the case if you got there from the normal starting position), the computer will correctly tell you that there is no mate in two.

[Phil Neatherway]

A very interesting idea, but I am not really sure what it proves.
http://www.telegraph.co.uk/science/2017 ... ciousness/

Roger Penrose was interviewed on Radio 4's Today programme this morning, at about 8:20. His point seemed to be that computers are tremendous calculators but they don't understand chess.

[Jonathan Rogers]

Hi Jonathan,

"Illegal position, since the f and h pawns would promote to light squared bishops."

The f and h pawns whilst on f2 and h2 can take a piece placed on g1 and under promote,
or at anytime during the game take from the f-file or the h-file to the g-file.

ah yes. I was distracted by work

[Geoff Chandler]

Hi Jack,

You got it with No.4.

"4. If you say both sides can castle (which cannot be the case if you got there from the normal starting position),
the computer will correctly tell you that there is no mate in two."

The computer cannot recognise that although you say both sides can castle, they cannot.

The moment White plays 1.0-0 it proves Black cannot castle but a computer will play 1...0-0-0.

It's like this.

White to play and mate in two. I have told the computer that Black can castle.

r3k3/ppp5/5B2/8/8/8/5PQP/6K1 w - - 0 1

1.Qg7 and although I said it can castle it will not do it because is it illegal.

r3k3/1p1p2p1/2pP3p/8/8/5R2/PPPP4/4K2R w - - 0 1

In the above position I have yes Black & White can castle but the moment I play 1.0-0 it cannot reply 1...0-0-0.
It does not recognise the exact nature of the position. We have fooled it.

[Brendan O'Gorman]

Apologies, I mistook the badly drawn bishops for pawns! With the correct position, the chess engines I mentioned struggle.