The New Narrative 

Want to know what makes us different? This is a great place to start. See how we’re innovating and experimenting, constantly. And get our take on the ideas of the moment— and where we think tomorrow will take us—across the industry and around the world. 


Report Excerpt: Making A.I. Truly Conversational

Almost since the birth of modern computing, humans have viewed conversing with a machine as the pinnacle of both artificial intelligence and computing in general. Take the Turing Test, the most famous method for measuring AI, which is determined solely by a machine's ability to convincingly hold up its end of a conversation.

Progress in this area has been historically sporadic and experimental, but several recent developments have made the goal of conversing naturally with a machine feel attainable and valuable. These developments include: technical advances in speech recognition and text-to-speech, industry efforts around building SDKs for “chatbots,” widespread deployment of “Intelligent Personal Assistants” like Alexa, and enterprise adoption of practical AI applications such as Natural Language Generation (NLG).

To understand how to take A.I.-powered conversations to the next level, it's important to first understand (leaving machines out of this for a second): what is a conversation? Many interactions we have on a daily basis (ordering coffee, asking for a colleague’s availability) are not in fact conversations. Here is a framework to help you think about what makes something truly conversational:

Conversations are contextualized

Conversations do not occur in a vacuum. Rather, they are contextualized by the participants and previous interactions. In order to have a fluid, dynamic, and intelligent conversation, you need to have a memory of the previous conversation and an understanding of who is participating in the conversation.

Conversations are flexible

Conversations can’t be defined or scripted out ahead of time. All participants need to be able to steer the conversation and introduce information or ideas that haven’t been explicitly solicited, based on a knowledge of the others’ underlying goals.

Conversations are cooperative

A conversation may involve disagreement, but they are certainly not zero-sum: no one “wins” or “loses” a conversation. Participants in a conversation have aligned interests, and a conversation advances those interests. If goals aren’t aligned, the interaction becomes a debate or disagreement, not a conversation.

Using this definition of conversation, it’s fair to ask: Have you had a conversation with a machine? Most likely, the answer is no. You may have had many useful interactions with Siri, but nothing that was an actual conversation. In fact, most of the “Intelligent Personal Assistants” we interact with (e.g. Siri, Alexa, Google Assistant, et al.) are explicitly designed to not support conversational interactions. Their role in our lives is that of a lackey, a simple agent that can gather some information or complete some task on our behalf, with as little effort as possible from us.

How does Quill power contextual, flexible, and cooperative conversational experiences?

Going back to our definition of conversation, let's examine how our NLG platform, Quill, can help power conversational experiences that can drive value for the enterprise, such as empowering a divisional manager to ask why their profits are down YoY or helping a recently discharged hospital patient to understand why his insurance plan only covers a portion of his bill.

Conversations are contextualized

Recognizing and understanding a person’s goals and intentions requires understanding her business and what’s important to her, tasks that computers have struggled with historically. Quill provides context to conversational interfaces, as it is backed by a knowledge base, with the ability to be taught about new domains and the appropriate analytics and expressions to convey the concepts of that domain.

Quill can be taught a new domain (such as “sales performance”, “franchise restaurants”, or “anti-money laundering”) by specifying the entities in the domain (restaurants, salespeople, etc.), relationships between the entities, typical values and important thresholds, sentiment (is a large value good or bad?), and the language to use when expressing that domain.

Conversations are flexible

Conversational participants need to be able to understand what is most interesting and important to articulate, instead of engaging in dialogue that fails to meet any participant's goals.

Users configure a narrative in Quill by telling it about their communication goals; in other words, what does the reader want to know? Quill’s model of narrative generation is focused on satisfying the user’s communication goals, and then structuring the resulting ideas and expressions into a coherent, cohesive story.

This foundation of a goal-oriented approach lays the foundation for AI-powered conversations, as it enables the system to have the context necessary to engage in relevant and valuable dialogue. By telling Quill your goals, the system then knows:

  • The types of data necessary to satisfy the goal
  • The kinds of analysis that should be performed
  • What constitutes an especially interesting or unusual outcome
  • What is most important in meeting this goal

Conversations are cooperative

When relevant, conversational participants should offer explanations of the past and guidance for the future. This makes the conversation much more valuable to the participant.

Quill understands a wide array of different ideas and situations common to data-driven reporting or conversations. These ideas include both quantitative statements (“Mary was the top salesperson this quarter”) and qualitative assessments (“Mary carried the team this quarter”).These valuable insights that make up Quill’s reporting today will make up Quill’s conversations tomorrow. And as we continue to develop our library of ideas and assessments, they’ll serve naturally in both reporting and conversational use cases.

artificial intelligence conversational cta


Privacy Policy | © 2018 Narrative Science