They can't. Engines are absurdly strong relative to humans these days, and Stockfish should be fully expected to beat top human GMs 10-0 without much trouble. In fact, they already have - and that's in spite of Komodo being a few tens of elo weaker than Stockfish. Book lines with high draw rates don't matter: either the engine avoids it, or they play it and win anyway because when the human inevitably makes a mistake, the computer capitalizes.
- The engine needs to be properly configured. Stockfish comes with a "contempt" setting that needs to be turned up. This makes the program avoid quick draws.
- The other top engine right now, Leela, does not have a contempt setting. This means humans can indeed score draws against it (this was against AlphaZero, which is similar to Leela and probably weaker than the latest versions). However: humans still cannot beat Leela/AlphaZero.
- The engine needs to run on reasonably strong hardware. These days engines will beat humans even on modest hardware, but if you try very very modest hardware such that the engine only manages to search a few thousand positions per second, humans might still win.
Edit: for the strategy you suggested, this was what happened when GM Michael Adams tried it against Hydra in 2005. Link to game if you want to see the moves.
Game four was another bitter pill for the human player. Adams, who had the white pieces, did everything he could to draw: exchange pieces to reduce material, simplifying in every way he could. By move 19 he had traded down to rooks and a bishop on either side; by move 27 each side had two rooks and five pawns; and at move 33 it was rook plus three pawns each. This one had to be a draw, but unfortunately the giant machine would have none of it and ground him down for a third victory in this match.
Hydra is, of course, very very inferior to the latest versions of Stockfish, and will probably lose a 100-game match by something approaching 0-100.