Learning a skill requires practice. It might even be the only way to learn anything. How would you learn to write without practicing it with your hand and a sheet of paper? How would you learn to meditate without sitting for hours, watching your breath?
But practice can be transferred. When someone has walked the walk, they can precipitate their learning into a transferable form: knowledge. Using knowledge, learners go faster because they don’t have to redo everything from scratch. And knowledge compounds. One of the reasons why our species dominates its home planet.
Practice is also needed in machine learning. The models that have ruled the field for the last decade are based on neural networks trained on millions of examples. Models try things, we tell them if they succeed or not, update their behavior, and do it over and over again. Practice. More clever methods were tried before that, but brute-forcing learning with practice won in the end.