Second Life as a platform is unique in that it is user driven and user built.
There should be no doubt that when approaching an avatar that it's directly controlled by a human.
While it is undeniable that scripted agents provide utility and amusement, they should not exist in any form that allows them to be confused with a human controlled avatar. Ideally there should be no task in SL that needs fully fledged antonymous scripted agent to accomplish and LSL needs to be expanded to remove these edge use cases.
Survey and traffic bots are an annoyance, likely privacy breaching and often operating in ways that break ToS.
Self identification of accounts as bots has broadly been a failure and cop-out on the part of Linden Lab. LL need to be marking bots as bots automatically. It doesn't even need fancy AI to accomplish. Metrics such as teleports per hour, time spent in the same location, viewer channel, etc etc etc.
Confusion over what is behind an avatar has created a trope in SL that unresponsive avatars are in fact bots, that clubs and venues aren't occupied by individuals, the direct implication being that SL isn't a real place made and populated by real people.
No one wants to waste time trying to be be social with a bot.
AI chat bots are novelty and not a substitute for real people.
At the very least name tags should be flagged in some way that makes it obvious an agent is scripted in some way. Participation in local chat should be equally differentiated. profiles should be limited, etc etc.