Absolutely agree, however there will certainly be a time where AIs will swarm #nostr to manufacture hate. Maybe Nostr is just too small at this time to justify poisoning the discussions. At some point maybe Nostr will gain to integrate a #ProofOfPersonhood protocol such as Idena (https://idena.io/) so users can prove that they are humans and not bots. 

 Curated paid relays can stop a lot of bots and spam in the future.

It will just be the cost of having a clean timeline. 

 But how would they remove « troll AIs » poisoning discussions (like those on Reddit)? 

 With curated relays. 

A relay owner could potentially manually curate which npubs gets to write to that relay. That means if you only use that relay you would see a very curated timeline of manually picked people that are not bots or AI. 

Of course it might not be feasible to do it, but it can be done already. 

 It seems that what you describe is a whitelist relay which may have interesting properties but probably not a good fit to act as a regular relay free of AIs. 

 I'm not solving captchas for a token lol that idea is older than Bitcoin it's called fatigue operator and bot operators took turns solving each other's capatchas so their bot could run when they were away. 

I like the model where relays handle the bans and anyone can easily spin up a relay.