Two academic research studies have revealed that “friendly” chatbots aren’t always having the intended effect.