I don't have any favorite, but certainly the ones I admire the most are: Lasker, his contributions to the philosophy of chess are timeless and his style was way ahead of his era, people thought he played 'wrong moves' to confuse his opponents but in reality he only played maneuvers that used the strategy elements in a subtle way like in this game
http://www.chessgames.com/perl/chessgame?gid=1026352 ; Alekhine, for the finest tactics ever in my opinion
http://www.chessgames.com/perl/chessgame?gid=1012498 ; Yusupov, for his great technical approach to strategical chess and clear methodology on training, this classical game of his is one the best I've ever seen
http://www.chessgames.com/perl/chessgame?gid=1259931; and Kramnik, because defeating Kasparov is just not something any mortal does, and the way he plays is just too elegant and precise
http://www.chessgames.com/perl/chessgame?gid=1252049, of course there are many other players worthy of mention like Morphy, Tal, Fischer, Kasparov himself and even Carlsen, for me the greatest of the modern era, but overall the first four I quoted are my most inspirational ones.
I see you like Karpov's style and couldn't be less reasonable because he is also one of the greatest, his recorded games illustrate endgame technique at its best, Ivanchuk has great romantic style and is also a great reference to epic chess
http://www.chessgames.com/perl/chessgame?gid=1060207 this is one of his best games ever that I enjoyed studying. In respect of your criterion on computer chess, I think hopefully that computers can't take away the human element of the game because it will always need intuition to play with, the fact that humans lose to chess modules is because the standard position has been just so monotonously over-studied that their databases cover most opening moves at perfection and any inaccuracy from a human player quickly gets out-calculated in activity because the engine can't fail at calculating an optimal sequence based on the human inaccuracy for that's what the engines have been designed for, but they can't truly understand the value of the pieces beyond the material factor as was shown in the Alphazero-Stockfish match, like in this game for example
http://www.chessgames.com/perl/chessgame?gid=1937930, Alphazero somehow seems to illustrate the development of an 'intuition' but I think that is more like a module-calculation with structure dynamics that belong to a balanced projection of piece activity within its own system of space, like 'playing without blundering but also keeping balance over the territory controlled', how the programmers achieved that I have no idea, but I think it's something that can be rationally studied too and learned by humans' intuitive quality, as that's how I approach my chess training, although for this I think the standard of chess should be reconsidered so as to not to restrict the understanding of the game like Chess960 attempts to do, so that learning the game becomes a study of activity and structure, not just "such pawn-london-hedgehog-fianchetto system opening" which restricts the understanding of the game to positions that can only be achieved in the current standard, while there can exist infinite possible structures in the game by which people can learn and improve, reason why I support the promotion of the 960 variant, which interestingly was designed by a chess prodigy and that couldn't mean less for recognizing it as a great approach to the game, because no matter what the starting position is chess will always be about security of the king, piece development, central control and draw or checkmate, but in my opinion computers in general can't be superior to its creators for there is always humanity to optimize these concepts just like the Alphazero thing, don't give up on romantic chess man! just enjoy the game in life