Three Principles for Building Better Chatbots

View fullsize

by Vagelis H.

Building effective chatbots is challenging for several reasons. 

Chatbots must first understand the user’s information need, as expressed by one or more dialog utterances. Then, the chatbot must figure out how to best respond, by combining available information which may be in the form or flow diagrams, FAQs or documents. An often ignored attribute of chatbots is that they must show empathy and emotionally connect with the user, to maximize their effectiveness. 

A recent article by Prof. Rosalind Picard of MIT identifies three key principles that chatbots must possess to build great relationships with humans, which are the same principles that a human must possess to build great relationships with other humans. Here, we list these principles and discuss how these can be applied to chatbots (the original article mainly focuses on human-to–human relationships).

Principle 1: Continuously collect feedback from users and use this feedback to improve the chatbot.  For example, at the end of each conversation the chatbot may invite the user to take a short survey. Or right after a response that the chatbot thinks that it may be inaccurate, ask the user if this response is what they expected and if not, why. 

Principle 2: Make sure the chatbot correctly understands the user’s information need. For example, if the user says “how much is it?” the bot can reply “are you inquiring about the price of the Botox procedure?” That is, the bot could repeat the user’s question, paraphrasing it and adding missing context (possibly collected from previous turns) to confirm that it properly understands the user’s need.

Principle 3: Try to generate active, constructive responses. Such responses make people happy and help build stronger relationships. For example, if the user shares some good news, respond enthusiastically, instead of just saying “congratulations.”