Maia: A Good Computer Opponent for Humans, at Last?
While I will sometimes encourage students to play out certain positions against their engines, I don't suggest they play full games against them. The reason is not that the engines will crush them; on full strength, they'll crush Magnus Carlsen and everyone else, too, and in any case there's a workaround: lots of engines can be set to play more weakly. The problem with "lobotomizing" an engine so it plays to the human's approximate rating is that the engine won't play anything like a human. Setting an engine to play like a 1500, say, is likelier to mean some combination of good moves alternating with senseless ones, with a sprinkling of clear tactical errors. In other words, moves that are either stronger or weaker than those the human would make, and with relatively little resemblance to what a human would do.
This has been a problem from the beginnings of computer chess, but thanks to the wonders of the neural network approach clever programmers may have found a way around this. Go here (HT: Tyler Cowen) to learn more about Maia, which was trained not by playing itself or other engines, but on a library of human games played on Lichess. It's an interesting project, and I wish it well.