You CAN understand the engine

People always say that the engine is really bad at explaining why it recommends a very strange positional-looking move, and that an engine can't explain its plans very well.
But I disagree, because 1.as long as you go deep enough then 2.you will find that the computer will always quickly convert an unclear advantage into a concrete/material advantage (which you can understand).

It's not that some move is strong for some mysterious reason and the engine can never explain to you why. The move just so happens to lead to a forced win of material, and if you dig deep enough, you will find the exact reason why.
You also will frequently find that many confusing computer lines line on the backbone of one single move (if this move did not exist then the whole combination doesn't work), that you may not have spotted as a human, so it makes it more forgiving if you didn't understand why a combination worked at first.

This goes hand in hand with one of the major transitions from beginner to advanced chess, which is the transition from thinking in terms of "general plans" to thinking in terms of "concrete variations". There is a specific move/square/reason why the combination works, and you must base your plans on variations that take into account the specific details of the position.

This is probably hard to understand when I'm not showing any examples.

K...

Here is an example that I often use in connection with this sort of issue: After 1 e4 e5 2 Nf3 Nh6 3 d4 f6 4 Bxh6 gxh6 5 Nxe5 fxe5 6 Qh5+ Ke7 7 Qxe5+ Kf7, the machine seems to think that 8 Nc3 is best.

@kindaspongey said in #3:
> Here is an example that I often use in connection with this sort of issue: After 1 e4 e5 2 Nf3 Nh6 3 d4 f6 4 Bxh6 gxh6 5 Nxe5 fxe5 6 Qh5+ Ke7 7 Qxe5+ Kf7, the machine seems to think that 8 Nc3 is best.

After brief analysis with the engine, Nc3 seems to block the e1-a5 diagonal, thus threatening the rook on h8 without running into Bb4+, thus preventing black's defense d5.

I am not confident that very many players would perceive a correct explanation from machine output.

@kindaspongey said in #3:
> Here is an example that I often use in connection with this sort of issue: After 1 e4 e5 2 Nf3 Nh6 3 d4 f6 4 Bxh6 gxh6 5 Nxe5 fxe5 6 Qh5+ Ke7 7 Qxe5+ Kf7, the machine seems to think that 8 Nc3 is best.
I checked, and the reason is very simple. White would like to play Bc4, but d5 is surprisingly resilient, since Bxd5 Kg6 Qxh8 runs into Bb4. With Nc3 white
1: develops more pieces into the attack
2. Strengthens control of d5
3. But most importantly, (since Bc4 would be significantly better according to the above) it concretely prevents Bb4 from being check.
One more thing-I would be curious if engine says the same at a very high depth since that can also change things!

I once saw someone argue in favor of 8 Bc4+ d5 9 Bxd5+ Kg6 10 Qg3+. After 8 Nc3, the reply seems likely to be 8...Bg7.

I don't fully agree because often the engine lines are so deep that you cannot understand them. Over a large depth there are many continuations and it is impossible for a human to go through all of them, only some of them. Most of the time you can understand why it works (though not how you would find it) but oftentimes you can't even do that

Actually it’s most of the time hardly any deeper than a couple of moves. But the machine combines all resources incredibly well...

It thinks Bc4+ is best for me. Ultimately its preference comes down to which move it thinks leads to the winningest position after what it calculates as best play by both sides in some 20 moves. Both moves lead to positions that are pretty winning but not so winning that it can see white forcibly winning massive amounts of material (let alone checkmating black) within its search horizon so far.