As with other innovations, there’s reality amid the chatbot hype, as I discovered when I posted several chatbot-related questions in a community forum. Among the responses were two accounts that prove that when it comes to chatbots, it’s not just talk.
Oracle ACE Leon Smiers, Center of Excellence lead for Oracle PaaS at Capgemini, based in the Netherlands, reports that his organization has already implemented chatbot projects using a variety of technologies, and has recently started a new chatbot project for the Dutch police that will make use of Oracle Technology.
According to Smiers, designing a chatbot project differs considerably from designing a standard user interface for a Web application or a mobile application. “The focus is on conversation rather than on transaction,” Smiers says. “Chatbots are another channel in the interaction with customer, with the primary focus on delivering a fast track for providing answers or enabling transactions.”
Smiers cites two unique challenges in developing chatbots. The first, and this should come as no surprise, is the complexity of language. As he explains, there are a variety of ways one might report that one’s bicycle was stolen. “My tandem is gone,” Smiers suggests, or “somebody nicked my iron horse,” amid nearly infinite variations. “With bots you need to able to understand the intent of a sentence, in singularity and in context of a conversation,” Smiers explains.
The other challenge? The sky-high expectations of end users and companies. “Chatbots are expected to deliver neuro-linguistic programming [NLP] and artificial intelligence in every possible way,” Smiers explains. “So we need to tone down the user expectations.” To that end Smiers’ team has created a Chatbot Maturity model, “to provide a roadmap and make it clear what can be expected in the first and consequent releases.”
Smiers and his team rely on a variety of tools and technologies to support chatbot development, including Apache Open NLP and Stanford NLP. “These can be implicitly available in tooling, such as the Intelligent Bot Services, but also explicitly used when calls via API’s are made to enrich the conversation,” Smiers explains. “These API calls need to be fed with the proper information from the question sentence in order to provide the right answer.”
Oracle ACE John Sim, a consultant with Fishbowl Solutions in the UK, reports that the Fishbowl team has built Atlas, a chatbot that was initially developed as an entry in Fishbowl’s annual hackathon event. “We are now using it internally to help to build up its intelligence through machine learning and natural language understanding capabilities,” Sim says. “Once we feel Atlas has built up enough wisdom, we plan to release him as an offering with integrations for SaaS and PaaS applications.”
The Fishbowl team did its homework before kicking off the Atlas project. “Our goal from the beginning was to make an intelligent bot and not simply an if-then-else bot,” Sim says. “We also wanted to write a user flow and look at common tasks where we could abstract the user interface and do the interactions through user intents. Our intent here was to enhance the user experience to quickly access content and get tasks completed effectively without going through a click form filling process stream.”
The Atlas project was fun, according to Sim, but it wasn’t without challenges. “The biggest ones were multilingual support and connecting user credentials so that Atlas only retrieves relevant info for the user,” Sim says.
And the work continues. “We are now looking at areas where we can enhance Atlas, such as an administrative interface to allow users to review mapped credentials and his offerings and capabilities from outside of the chat window on Facebook and Slack,” Sim says. “We are excited about the future for Atlas and the use cases for bots overall. Who knows, bots could replace mobile apps. We shall see.”
How do chatbots figure in your future? Are you doing your homework to prepare to meet the challenges of this revolution in human-computer interaction? Post your comments below.