Engine games (you can watch them at tcec-chess.com and some are terrific) with the standard starting position are almost always drawn. Engine tournaments are usually played instead with human selected, unbalanced openings designed to produce an exciting game with lower likelihood of a draw. The unbalanced opening intentionally gives one side (black or white) an advantage, so each opening is used twice, with each engine playing one game as black and one as white, to balance things out. The games are usually drawn anyway, with some occasional exciting wins. The top human players now study engine games closely, and play more like engines themselves. Very few mistakes => more draws.

It's unknown what would happen if a top human with serious preparation played against a modern engine. They'd have basically no chance of winning and a fairly low chance of a draw, with the unknown being "how low is fairly low". My wild-ass guess is maybe 1 game in 10 when the human has the white pieces and much lower with the black pieces. There is not much chess theory about what happens when a human plays a game from the beginning aiming for a draw, even though they do that all the time in real chess. It is something of a gap in the "literature".

High level computers draw each other almost 100% of the time and beat humans 100% of the time.

In general as skill increases (assuming the competitors are evenly matched), so does draw rate. Human grandmasters draw much more often than human beginners, and engines draw even more than human GMs.

This is one of the intuitive reasons for hypothesizing that chess is drawn with perfect play, even though we can’t prove it.

Yes, there is a limitation. People have to learn high-level ideas and develop their intuition. Computers can have millions of training games (to teach the NN and to do regression tests), terabytes of tablebase positions and an ability to calculate thousands of lines in real-time.