When chatting with chatbots, the idea isn’t so much to have a conversation but to fulfill a purpose, such as asking about restaurant table availability, booking a flight, checking on information, and so on. The goal is that these requests can be made using more “natural” language, which means that it feels less robotic.
However we have to admit that chatbots for the most part still feel plenty robotic, but that’s something Facebook hopes to change. In a recently published paper by researchers at Facebook’s FAIR lab (via The Verge), the researchers are hoping to solve some of the problems with chatbots, such as by giving them a “consistent personality”, as well as making them remember conversations that they might have had with you several sentences ago.
To do this, Facebook’s engineers built their own dataset which they will use to train chatbots called “Persona-Chat”, and it will consist of 160,000 lines of dialogue that has been sourced from Amazon’s Mechanical Turk marketplace, ultimately giving these chatbots a “personality” that will make interacting with them more dynamic and less robotic.
Of course whether or not this is necessary, and whether or not it will actually be applied to chatbots being used by Facebook right now remains to be seen, but if anything we guess this brings us closer to what we’re seeing in the movies, where AI can have personalities or their own that makes interacting with them feel more natural.