The purpose of this article is to introduce a framework and reference architecture that brings together and describes different ideas and approaches to design an intelligent chatbot. We will start by reviewing the proposed intelligent chatbot framework that revolves around conversational intelligence, cognitive intelligence and intelligent authentication, all within an omni channel strategy. We will then study the proposed reference architecture which goes a step further in identifying the required components, their integration and functionalities to implement this framework.
The proposed intelligent bot framework is made up of three main building blocks: conversational intelligence, cognitive intelligence and intelligent authentication as detailed below. It is deployed across all the channels that users interact with, such as messaging applications, mobile applications, phone systems, Web, chat applications and social media. This ensures that users have a consistent experience regardless of the channel they use.
An intelligent bot must be capable of holding a consistent and intelligent, two-way conversation with a user. The bot must be able to maintain context as the user changes subjects or uses colloquial, conversational expressions and words, similar to how any human would. Presently, most bots are not sophisticated enough to do this. Some might be able to successfully respond to questions, but if the user follows up with the conversation, most bots today cannot maintain context about the question.
An intelligent bot should be aware of its environment, make decisions, execute actions and even predict a user’s needs. While legacy speech recognition systems understood what people said, today’s sophisticated natural language processing systems understand what people mean and want to do. Both speech recognition and natural language understanding rely on a large volume of training data, domain-specific knowledge and user information.
Intelligent bot should ideally be using voice biometrics to allow users to easily and naturally authenticate their identity without having to type in a password or PIN and by simply speaking a short passphrase. This eliminates hard to remember PINs or worse, the need to answer a series of security questions such as “What was the name of your best childhood friend?” or “What was your most recent transaction?” Furthermore, voice biometrics significantly improve security over legacy authentication methods and fraud. The bot should also use facial recognition in combination with voice biometrics for tighter security.
An intelligent bot is not just a single machine learning model but is an integrated set of various AI components which facilitate different intelligence functions. Thus the proposed intelligent bot reference architecture implements the framework as detailed below.
Master Bot (Multichannel Integration)
A Master Bot interacts with users via multiple channels while maintaining a consistent experience and context. It understands the user’s questions and responds to frequently asked questions and routes the rest to the appropriate slave bot (retrieval based or generative) for response and facilitates the required intelligence services or capabilities.
Intelligent authentication is speech and facial recognition models that authenticate the user and also recognizes the user and provides a personalized interaction.
Slave Bots (Conversation Artificial Intelligence)
Slave bots are retrieval based, or “helper bots” that collect information on behalf of the master bot via the information extractor or integration services. It can also predict the next best action, product, service, or promotion for example, using the recommendation engine. Retrieval based bots might be domain specific and/or task specific and can chain several together.
Generative is a helper bot that generates responses to the users questions which do not require any external source of information such as general chit chat. It also has other capabilities like negotiation, self-learning and personalities.
Language translation is an AI model that detects the users language and translates languages to make the bot multilingual, allowing it to chat with users in different languages.
This involves three AI models that evaluate the sentiment for each response to understand the mood of the users for managing the conversation accordingly or empathizing with them. One model performs text analytics to identify sentiment in chat responses or voice responses by converting voice to text. Other model performs tone analysis of user’s voice to identify voice patterns which determine emotions in the user’s voice as they speak with the bot. Yet another model performs facial image or video analysis to read facial expressions and determine user emotions. The previously mentioned analysis models can also determine the user’s age and sex that can improve managing the conversation, provide a human touch and augment user experience.
Cognitive Artificial Intelligence
AI Services: information extractor is an AI model that retrieves information from unstructured data sources like documents, big data, and knowledge base including online document sources like Wikipedia and ranks it based on confidence level. A retrieval based bot uses an information extractor to respond to more technical or knowledge based questions. It also uses the recommendation engine which is a prediction based model to provide personalized recommendations to the user based on historic data, context, or behavior.
AI Capabilities: This includes a set of capabilities that the generative bot is trained to demonstrate human-like intelligence. The bot can negotiate with the user or on behalf of the user with an objective to gain something. It also learns by itself new information and styles via repeated interactions with the user. It uses different personalities to interact with different users based on who the user is (Male/Female, Young/Old etc…) also handles profiles like different kinds of professionals and nationalities including individual personalities.
Integration Services is a retrieval based bot that uses integration services to collect information from external systems, services or databases.
We have discussed a framework that depicts the building blocks required to design an intelligent bot. We have also explored a reference architecture that shows how the framework can be implemented using an integrated set of AI components that facilitate the intelligence that the bot requires.