Abstract: Now that we have the tools and algorithms to develop machines that converse with humans (and each other) we should begin to consider features other than utility and popularity. Natural language generation systems are beginning to have a significant impact on society. The underlying ""merit function"" of those systems will determine whether that overall impact is positive or negative, and whether those benefits are unequally distributed. Many chatbots are designed to optimize for the revenue stream of their owners and this tends to favor popularity and sensationalism. If configured for live, online training, this can lead to antisocial behavior, due to manipulation by their audience or just the environment in which they are incentivized to perform. Nonetheless it is not difficult to build machine learning conversation engines that tend to stabilize with more prosocial behavior. If the history of the evolution of organisms is any indication of the likely advancement of chatbot systems, these prosocial conversation engines should out-compete antisocial bots. The short lifespan of antisocial chatbots sets the bar relatively low for ""survival"" in the Darwinian ecosystem of chatbots.
In this workshop we will show you how to combine several python packages and tools to build a prosocial chatbot. We will utilize machine learning information extraction algortithms to give a chatbot the ability respond with desired actions or answer basic questions. Then we will give the chatbot semantic search capability so that a training set of human conversations from movie dialog and online chatrooms can give the chatbot a more conversational personality. We'll show how this capability can quickly tend towards anti-social behavior unless it is configured to ""think before it speaks"" by considering many alternative statements and including prosocial characteristics among it's list of priorities when selecting a response.
Bio: Hobson is the lead author of "Natural Language Processing in Action" to be published in early 2018 and he is the president and CTO of Total Good, a nonprofit that contributes to the common good with Data Science software and education.