Jazz music and conversations have a lot of things in common -- both require improvisation and attention to what others are saying or doing. So, in some respects, the decision by All About Jazz publisher Michael Ricci to implement a chatbot to interact with users of his website was a natural one.
"With the bot, you are engaging, and it directs you to pages that interest you," he said. "The idea is that if you really like something, you're getting the results you're looking for quickly."
Like many others who see potential business uses for chatbots today, Ricci said he's excited about what the bot can do to make the online content he publishes more interactive. And, along with some experts in the field, he would like to see chatbot applications continue to develop and improve their conversational abilities in order to make them even more useful.
The All About Jazz agent takes the form of a Facebook chatbot; it was built for that platform by San Francisco-based bot vendor Gupshup. Ricci said users on Facebook can ask it to bring up articles about particular artists or the latest album reviews. Over time, the bot learns users' preferences and automatically delivers relevant content.
A chatbot with a personality
Of course, one of the key differentiators between talking with a person and chatting with a bot is that the latter traditionally lacks personality. Ricci said he put a lot of thought into this because it's important for users to feel like they're having a real conversation to truly engage with chatbot applications.
"The idea was [to] make it sound fun," Ricci said. "Then at least your initial interaction would sound conversational. It wouldn't sound stilted like a typical AI response."
To accomplish this, he personally scripted responses to common queries and tried to inject some personality. The bot had been live for just a couple weeks at the time of this interview, so Ricci said he doesn't have much information on how users are interacting with it.
At first, he expects it to deliver the scripted responses that send people straight to the content. But as the bot learns more about people's content preferences, and the underlying machine learning algorithms have a chance to build a deeper knowledge base, he hopes people will be able to have more back-and-forth, improvisational conversations with it.
Conversing with chatbot applications
This kind of conversational ability is the holy grail of chatbots right now. While AI-fueled assistants are popping up everywhere, from home assistants to internet retail sites, the ability to engage in a natural dialogue remains evasive.
At the AI World 2017 Conference & Expo, held in Boston in December, Forrester analyst Mike Facemire said today's chatbot applications function best as order takers. When you give them a straightforward request, they're pretty good at returning an intelligent result. But if you try to engage them in a true conversation, things can go south.
"If you have a conversation, it will fail," Facemire said. "If you treat it like an answering machine and it just comes back with an answer, it works. The more back and forth, the more it's likely to fail."
Facemire said the best approach to chatbot design today is to think about your brand and how your bot's voice or personality can fit into it. He thinks chatbots should also be applied to simple, low-stakes use cases rather than with large purchases that require lots of consultation, for example.
As machine learning algorithms get better at interpreting and generating natural language, they could theoretically be applied to deeper problems, Facemire said.
Still room for chatbots to grow
Getting at deeper ideas is what conversation is all about, and this will be the next frontier chatbot applications need to conquer, said Rob High, vice president and CTO in IBM's Watson group. Speaking at the AI World conference, he said he doesn't like applying the term chatbot to today's technology because it generally works by returning a mostly scripted answer to predefined query topics.
But true conversation is about understanding a person's intent, not just their literal words. For example, he said when a person checks his bank account balance, the true purpose for doing it is probably to see if they can afford to buy something. Chatbots should be better at understanding these types of queries, he said.
"There's something deeper in there," High said. "A conversational agent should work through your problems."