I'm wondering how much difficulty is caused in AI research because with social media experiments it's impossible to tell a malfunctioning AI from one that's shitposting.