Machine Learning and AI with Simple Python and Matlab Scripts

11RNNs for Chatbot Implementation

11.1 Introduction

The imitation game, proposed by Alan Turing in 1950 [2], states that if a machine (e.g. a computer) can answer any question put to it using the words that an ordinary human would, then we may call this machine intelligent. The Turing test [97] is based on this game, whose goal is to determine if the answers provided by the machine are indistinguishable from those by a human. The Turing test, where a machine tries to impersonate a human, is considered one of the earliest attempts at defining artificial intelligence, although it is debatable that answers to such questions requires any real intelligence other than simple search techniques based on previous conversations.

A modern implementation of this experiment would employ a chatbot [98], which is a program that mimics a conversation with a human using natural language interactions (e.g. text or voice) [99, 100]. Current chatbots are typically online and capable of maintaining a conversation in a natural language based on input stimulus provided by the user. Recently chatbots have evolved to bring a significant societal impact by potentially replacing humans in many professional settings, including answering services and customer service operations.

In this chapter, we present an implementation of a simplistic chatbot based on an RNN, which was introduced in Chapter 10. This is a fairly difficult problem for other AI tools such as ANNs and CNNs since they do not look for a ...