More feedback from other people can be found on the incremental db.
Improvement: The paddles should calculate which ball is headed their way that they are capable of getting in front of at their current speed. Currently they only care about the one target ball.
If the player's cursor is within the left 1/5 of the court, allow them to control the paddle's movement, adding interaction to the early slog.
An option to prevent the current target ball from getting stuck in a pattern, randomize the target spot of the paddle for the AI by about 20% of the paddle's width, such that the ball's target spot is in the center 40% of the paddle. (This suggestion assumes original pong's style is used, where the further from center a ball hits the paddle, the more angle is applied to a ball.)
Upgrade option: If the player's paddle AI has a 50% chance to fail to calculate if it can get in front of the ball, and the opponent is 100% correct success rate, an upgrade to improve the player's accuracy to find the correct ball to get in front of, and "improve" the AI's chances of missing. (Re-reading this days later, here is another shot at explaining this: If the first improvement above is implemented, the paddles are determining which ball they should get in front of. If instead of them being perfect in their prediction, they had some chance of being wrong, balls could get past them that they should have otherwise gotten).