Though the personal artificial intelligence M, was a complete failure, Facebook is still hopeful about chatbots and its future. In last couple of years Facebook has brought betterment in its AI technology and now it wants to introduce small talk, proper conversation through chatbots.
Now we might assume that if their name is chatbots, their aim is to chat but chatbots can’t chat. Facebook’s Fair lab researcher explained that chatbots have failed to chat at a number of levels.
One issue encountered is that they keep changing the facts about themselves during a conversation. So, they don’t show a “consistent personality”.
Two they keep forgetting what they or the partner they were conversing with had said in the past.
Another major problem with them is that if they come across a difficult question, they don’t quite understand, they go back to programmed answers and reply “I don’t know.”
Though there are some loopholes in the technology, still chatbots are amazing. They do ask questions like “How do you feel now?” but more than interrogation the goal is to achieve conversation. In order to achieve the target, the experts have turned towards deep learning. What this means is that instead of training chatbots through preprogrammed questions and answers chatbots are trained to look at patterns in huge datasets.
Till now this option is the most viable one and is showing better results but the only issue as per Facebook researchers is to get the right data to start with. Many chatbots are trained through movie scripts which sometimes makes conversation flawed. In order to resolve this issue, Facebook engineers are creating their own database to train chatbots. This database is called ‘Persona-Chat’. It has more than160000 lines of dialogue. Moreover, the idle talk of Persona-Chat is not completely random, it has meaning and reasoning.