News and Updates

Meet the faces behind conversational AI

Share this post:

Building an AI assistant is different from building an FAQ bot. AI assistants require interdisciplinary talent to properly build, scale and optimize. Teammates can include conversational engineers, computational linguists, UX designers and brand representatives.

So, who are the faces behind successful AI assistants? Meet one of them: creative writer for conversational AI Maya Hausammann, who crafts responses for Autodesk’s Virtual Assistant, AVA.

Course of study

Maya has a BA in creative writing and literature from the University of Michigan. With a strong background in social and developmental psychology, she is particularly interested in how language influences and reflects human behavior and emotions.

As a creative writer for conversational AI, Maya’s job is to teach AVA how to communicate like a customer service employee through the assistant’s conversational flow. She works to ensure that the Autodesk brand is incorporated into the virtual agent’s “personality.”

A day in the life

Being responsible for an AI personality puts Maya at the forefront of one of today’s newest professions. Her daily work focuses on analyzing AVA’s conversation logs to identify miscommunications between the assistant and the customer. After a behavioral analysis of similar miscommunications, she formulates a response strategy to avoid future miscommunication, using her creative writing skills to ensure that each proposed response is aligned with AVA’s voice. In doing so, she looks for personality and linguistic patterns that can better represent human speech and improve recognition of what customers are trying to communicate.

A typical tool Maya uses to ensure her writing is data driven and research based is data classification: grouping responses and/or customer utterances that express similar intents. This helps her formulate responses that help the overall conversation flow better and more naturally.

How to make a virtual assistant sound like a company’s representative without being creepy

As the official voice behind AVA, Maya pays special attention to ensuring that AVA is on brand, which means AVA needs to sound like Autodesk. Engaging in over 100,000 conversations per month, AVA has a huge impact on what customers first hear, see and experience with the brand. Maya’s job is to infuse AVA with the special attributes that make Autodesk Autodesk.

To do this, she needs to have a complete understanding of the traits that make up the Autodesk brand. How do you make a machine sound like it’s part of a company’s workforce? Your machine must have a personality with a distinct voice that isn’t exactly human but also not mechanical. That voice has to be on brand, consistent and emulate emotional intelligence—your machine may not be able to feel, but it must communicate a level of empathy and consideration for the customer it’s interacting with. For example, in order for AVA to be classified as “smart,” one of the Autodesk brand traits, she must be able to recognize what customers want and respond appropriately. If AVA knows that she won’t have the correct response, she should be able to connect customers with a human agent. To be read as “real,” or, as Maya adapted it, “authentic,” AVA has to convey her witty sense of humor, showing that she doesn’t take herself too seriously.

Although AVA has her own personality, which integrates the Autodesk brand attributes, it’s extremely important that she makes the customer aware that she’s an AI assistant, not a human. She does this at the start of each conversation.

This job description doesn’t include only writing and research

Autodesk has partnered with Soul Machines, whose vision is to “humanize artificial intelligence to better humanity,” to design a voice and video beta version of AVA that responds in a more human-like fashion.

With this project, Maya identifies which types of emotions should be attached to specific AVA utterances, and how to emulate human reactions, making her background in psychology quite useful. Using emotional markup language developed by Soul Machines, Maya programs AVA with micro expressions, small facial movements associated with different emotions, and reactions to try to convey empathy.

Unique parts of the job

Maya finds that the most interesting thing about being a creative writer for an AI solution is that any little change can shape an entire conversation.

There are many ways to get a desired result, allowing her to be truly creative with the way she designs conversations. She is also excited to see AVA expand into voice and video to meet Autodesk’s goal of creating the best possible customer experience.

In episode 3 of the Watson Masterclass, you can hear firsthand from the Autodesk Machine Assistance team. Hear the AVA business owner and strategist, creative writer for conversational AI, product manager, creative director, computational linguist and UX designer discuss how their team works to build and optimize AVA.

Register for Episode 3 of the Masterclass

Content Manager, IBM Watson

More News and Updates stories
September 21, 2018

Announcing the AI solution with voice for contact centers

IBM Customer Care Voice Agent is easy to set up and connect to existing call center infrastructure or telephone networks. That's part of the reason Forrester named IBM Watson Assistant the leader in their report, “The Forrester New Wave™: Conversational Computing Platforms, Q2 2018.”

Continue reading

September 19, 2018

It’s time to start breaking open the black box of AI

At IBM, we’ve seen our customers use AI as a catalyst to reimaging their workflows – transforming how customer call centers operate, how people complete their taxes, and how legal professionals make data privacy compliance decisions. However, many organizations still continue to struggle in deploying their AI into production environments across their existing applications. For AI […]

Continue reading

September 10, 2018

Clearscope makes it easy to improve organic search with Watson AI

Clearscope is able to use IBM Watson to reverse engineer the search engine process so content marketers can understand the key concepts surrounding the topics they want to write about, without having to duplicate content or resort to keyword stuffing.

Continue reading