@nayf I think on chesstempo you can set the mode to "mixed" to get a mix of winning and holding a drawn position.
@sheckley666 I can agree with your analogy, indeed just knowing what the computer thinks doesn't mean one can come up with it oneself. I was merely objecting to the statement that computers can't play "strategical moves".
@nayf I wouldn't call "creating future chances" necessarily as strategic though you do raise a point as that a perfect chess player (in the game theoretical sense) might not play very strong actually since he sees chess is a draw anyway so he would just pick one of the drawing moves. (and since he plays perfectly he has no understanding of what is maybe harder to defend for a weak player) However I would put creating chances in a different category alltogether which at its extreme are the famous swindles. These can be in some sense of strategical nature though more often they are variations where one hopes the opponent makes a mistake. (but as a human with imperfect calculation you can argue that since you can't see it all in advance it's not really a tactic for you) This category is very important for human chess since to win a game your opponent needs to make a mistake and this is a good way to lead them into making one. (e.g. I won many games OTB vs 2000+ players after having been down a piece or more) And the very strong engines aren't particularily good at human-specific swindles (e.g. they won't play a speculative sacrifice because they know that the opponent might mess up) which one could perhaps consider an engine weakness especially for analyzing.
But I stay to my point that strategic play is just a sort of approximation because the actual tactics would be way too deep for a human to calculate. Good players can approximate that very well which makes them "strategically" superior, also still calculating variations may give you a better approximation which is why even strategical players will calculate a lot. For computers that is a bit different, they also have a purely approximative part (that is what the evaluation function is for) however often times they don't need it since they can calculate it directly. (though worth noting that modern engines use really aggressive search heuristic to e.g. prune irrelevant lines, those in some sense are approximative too, or to use a human-chess word "intuition")