Originally posted by tvochess
Isn't it so that computers go entirely nuts when they are sure the position is completely lost?
I'm not sure about that.
Think about this though. You are in a lost position.
There are three moves to choose from.
Move 1: You simplify to an ending a bishop down that should lose by force. The computer has you at -2.79 or something (not looking all the way to the end).
Move 2: You withdraw from the attack and lose a pawn or two more. The computer only has you at -2.5 , but you no longer have any counterplay whatsoever.
Move 3: You make a dodgy rook sacrifice. The computer has you at -4.80. Two moves into the sacrifice, your opponent has 3 choices - one clearly loses, one leaves material even but your opponent on the defensive, and the third has a deadly trap in it but does win if he finds 3 more correct moves.
The computer will almost always choose line 1 or 2 even though they really give the losing side no chance. It has calculated out choice 3 and sees that it loses. Unfortunately, humans sometimes don't have the mental accuracy (especially in the heat of battle) or the nerves to find all the correct moves to survive the rook sacrifice line. Even though, it's not the computers best choice, it is still a human's best chance (provided he is playing against another human).
Computers are able to walk on razors with variations 5 moves deep or more with only a single line that survives and come out on top. I guess that's a real advantage of being able to calculate thousands of positions a minute!
I guess this is just a more general repeat of what I already posted, but I thought another example may be helpful to some.