Creating Conversations Between Humans & AI’s Panel


Elizabeth Arredondo is a writer focused on creating compelling characters for television and interactive mediums. She is currently designing the personality, backstory, and the conversations for a robotic wellness coach named Mabu. Mabu is the latest effort in social robotics from Cory Kidd, formerly of MIT’s Media Lab. After earning her MFA in Writing for Screen and TV from USC’s School for Cinematic Arts in 2005, Elizabeth received a feature film writing fellowship and participated in NBC’s “Writers on the Verge” program. Elizabeth worked as a staff writer on the primetime CBS drama COLD CASE. She has also worked with a network to develop an original pilot.

Daniel Padgett is a Conversation Design Lead for Google Assistant, creating engaging user experiences for Home, Pixel, Wear, and more. For the past 15 years, he has been leveraging language technologies to develop user-centered solutions for major brands like Allstate, Nike, Target, and Cisco. Just prior to Google, he led service design efforts as Director of Customer Service Experience at Walgreens, deploying proactive and personalized contact solutions for roughly 100 million customers at more than 8200 stores.

Cathy Pearl is the author of the O’Reilly book “Designing Voice User Interfaces”. She is VP of User Experience at Sensely, making healthcare more effective and accessible with a virtual nurse avatar. During her time at Nuance and Microsoft, she designed VUIs for banks, airlines, healthcare companies, and Ford SYNC, and at Volio she built a conversational iPad app that has Esquire Magazine’s style columnist advise users on what they should wear on a first date. She has a BS in Cognitive Science from UC San Diego, and an MS in Computer Science from Indiana University. In her talk Cathy looks at these points…

1. Design for how people actually talk, not how you want them to talk.
2. You can still design great conversations without strong AI/machine learning .
3. Designing for speech is different than designing GUIs.

Mariana Lin has been a writer for the past 15 years. She previously worked as a principal creative at Siri, developing personality and voice internationally, and currently consults in AI writing. Her writing has appeared in publications such as New York magazine, GQ, Picture, The Huffington Post, the BBC, the Mississippi Review, and the New Guard Literary Review, and she has won awards for poetry and voice branding. She has varying levels of proficiency in French, Chinese, classic Greek, ASL, LAMP (Language Acquisition through Motor Planning), and is interested in the interplay of language, culture, and AI identity. Mariana discusses these points…

1. Each AI personality should have a distinctive voice. It’s important to consider word choice in writing for AI dialogue. There are a hundred ways to express the sentiment of a positive state. “I’m quite all right,” “Everything’s peachy,” “All’s good in the neighborhood” all present a different nuance of tone.
2. Consistency of tone over hundreds of thousands of spoken lines are what creates a strong character and identity. It’s easy to choose a crowd-pleasing response in the moment, harder to create a more cohesive bigger-picture portrait of a character and its speech. But this is what people will respond to over time. Hire writers who focus on economy of language and word choice.
3. AI is combination of product and character design. We need designers and engineers, but we also need writers and creatives. People don’t just care about functionality, they care about relationships. We have relationships with brands, and we will certainly have relationships with AI. Writers and creatives understand what’s entailed in crafting deep, positive relationships. Character voice is different from brand voice. They might be related, but they should not be the same.

Erik Vinkhuyzen is a senior researcher at Nissan Research Center in Silicon Valley, where he is a member of the Human-Centered System group. He brings a social scientific perspective to the development of self-driving vehicle technologies. His studies have spanned a range of work settings and technologies from call center work and CRM systems, to clinicians working with electronic medical records, from copy shop employees using copiers and printers. His current work focuses on the interaction of autonomous vehicles with other road users, and the interactional challenges for non-drivers of autonomous vehicles. Erik examine these points…

1. Autonomous Vehicles must communicate through movement.
2. AVs must be able to read the behavior of other people as intentional actions, and understand how its behavior communicates.
3. Interactions on the road is based on vast amounts of background knowledge, including a shared perception of the environment, the rules of the road, and local norms.