The problem is similar to bots in gaming. Banning bots predictably gives the attacker (bot builder) clear success/failure signals for evading detection, then it becomes a detect-vs-evade arms race that the bot builders will likely win.

Detecting and monitoring the bots provides a possibly similar outcome, but t with the advantage that he platform knows the bots. If there are bots with way, the platform would probably prefer the option that allowes them to know who the bots are.

The middle ground is basic auto detection/bans for simple abuse and infrequent ban waves for the rest, where bots are detected and all banned together. This makes it hard for the writers to understand which signals have them away and gives them the disincentive of uncertainty. This is popular in gaming, where bot and cheat farms are detected and studied by they platform for up to several months before bans are handed out.