Are tactics ever successful against a computer?

In my years, I've played thousands of games against a computer. One thing I've noticed, is that I've never been able to successfully use a tactic against a strong computer. Have you? Think about it.

Has there ever been a single incidence against a strong chess program where you were actually able to pull off a surprise move that the computer didn't see? For example: A Queen sacrifice that leads to a winning position. Or, a Rook sacrifice that actually works and tips the evaluation in your favor?

You haven't.

Reason why, is that the computer sees the tactic loooooong before you. Not only does it see the tactic several moves before you, but it can look much deeper into the position than you can. It also has a plan to trump your fancy tactic just in case you happen to try it. This, by the way, has happened to me many times where I thought my tactic was good....only to find out that the program had swift and effective responses to make the move awful.

There are no surprise moves against a computer. It's going to see those tactics and have plans to strike them down.

So if you're going to be winning against a chess program, tactics are not going to help you in the slightest. It's the long terms strategical ideas that seem to work best. For example, lock down the pawn chain and flag it's clock by playing faster. This seems to be the most popular method of winning against a program. Other ideas like having a 3:2 pawn majority on either side of the board might actually help you beat a program if you make it to the endgame.

The reason I mention all this, is that some new games have surfaced between Alpha Zero and Stockfish. Alpha Zero is playing so well, that it's actually finding tactics that Stockfish isn't seeing. And it's not that Stockfish isn't actually seeing the move, it's just not really taking the move seriously or not evaluating it deeply or effectively enough.

Alpha Zero plays Ng4 clearly offering the knight for a pawn. Stockfish looks at the move, evaluates it, takes the knight, and then only a few moves later realizes it has made a mistake. AZ is successfully executing tactics against SF. That's how strong it is.

Humans rarely, if ever, have that luxury against a strong program.

You need to specify computer's hardware and allocated time or number of nodes. Latest SF at depth 3 or 1000 nodes will massively fail lots of tactics( SF rating  level probably 1600-1700), meanwhile the same SF on 1 billion nodes( SF rating level approx 3500) will see 90-95% of most tactics positions. 

In some difficult tactics, stockfish needs more than 1 trillion nodes( SF rating approx 3700+) to search to get the right answer, where we normally defined as Failed for Stockfish.

That's why I was saying "Strong" program. Not a 1600-1700 level strength. 

There are many positions SF failed tactically even with prolong analysis, like 10 min, 30 mins etc. This is one of the recent positon where latest SF failed at that time of my posting. Try to search with your computer, and let me know in how many hours it take SF to see the solution.

The game was actual human game play in Russia in around 1934.

https://www.chess.com/forum/view/general/so-funny-that-stockfish-failed-this-simple-knight-fork-puzzle

Chess is extremely massive to search. Stronger program means, better focusing in searching for checkmate threads, hanging pieces etc.  SF is a million times faster than human but it will still miss tactics cos it is impossible to search every move in chess. (10^100+ or something) . 

I just ran this position with SF on my laptop. There is no winning solution: only a draw. After 5...f5, white has no way to win even when the Q is sacked. 6.Nc7 loses. SF didn't find a solution because there is no solution.

Tactic does not work.

I made the same observation; you cannot ever execute any tactics on a decent computer, because that is essentially how they play. That is why the best human players all try to play "anti-computer" against any chess engine because they know this is their home turf, and humans just get crushed by them easily if they try to get into any tactical disputes with a computer. So it is very surprising to see AlphaZero squashing Stockfish tactically and making it look so puny.

That is why I told you it was an example of tactics where Stockfish Failed. Human already know the solution even before computers were invented in 1930s.

The solution is

1. Nc8 Qd8 2. Ng5+ Kg8 3. Ne7+ Kf8 4. Nh7+ Ke8 5. Nd5 Qb8 6. Nc7+ Kd8 7. b6 Kc8 8. g4 e4 9. Ng5 Bf8 10. Nxf7 Bxd6 11. Nb5 Bc5 12. Na7+ Qxa7 13. bxa7 Bxa7,

White win, forced win. SF failed to see, 1. Nc8 tactics.

There is no forced win for white. Black can get out of it.

Ok. How?

That position interested me, so i did some analysis.

I think white wins in any variation.

You mean 99,999%?

My program saw the position as a draw after 5...f5. Several moves after that, it still saw a draw even at a depth of 62. For some reason, it just wasn't seeing the winning line. I had to force it. So yeah, white does win. I stand corrected.

Still, I don't really see Nc8 as a tactic. I guess it is, but it's not really my idea of a surprise move like a sacrifice. To me, it's more like an attacking move and also the start of a really long and complex strategical idea.

I think examples like this one are quite rare. I have never really seen any successful tactical blows against an engine until Alpha Z.