A new day, a new step…still far from replicating intelligence, but we are now closer with Facebook’s new chatbot, Blender, claimed to be better than Google’s chatbot Meena, which was released earlier this year. Often confused with virtual assistants, chatbots will help them with their limitations but that doesn’t mean they are the same. Chatbots are mostly used in businesses for customer support or sales, while virtual assistants are meant to help people with their daily duties, creating shortcuts for them.
What is new? Firstly, Blender is a lot larger than Meena because it was trained on more data, specifically 1.5 million Reddit conversations, from which it learned how to respond to different questions. This was just the beginning because Blender also learned from additional data, meant to teach him empathy, knowledge, and personality, the core elements in a discussion. The majority of people chose Blender as a better conversationalist than Meena, and it also tricked 49% of people that they are talking to a human instead of a chatbot.
What is the downside then? Well, after 14 turns the chatbot loses track of the prior information, so it is not able to go into depth. Also, Blender tends to deliver false information, that it made using connections, a flaw of the deep learning methods it uses. Unfortunately, due to the biased data found online, the chatbot can adopt stereotypes and use them into conversations. To help with this problem, Facebook tried to filter out this type of unwanted information, but because of its size, they weren’t able to do so with all of the datasets.
Natural language processing is very important for customer service and it will continue to have a significant contribution to saving companies a lot of money and helping people focus on more valuable tasks.
