In the debate between the AI doomsayers, who are convinced AI will automate away most jobs, and the AI optimists, who believe this technological revolution, like all of those that preceded it, will create new jobs we haven’t yet imagined, the optimists are winning.
Many of the “jobs of the future” are here today, and they didn’t exist as recently as five years ago. One such job is AI Interaction Designer. At x.ai, that position falls to Diane Kim.
Vertical AI like x.ai is very narrowly focused on completing a single task. In x.ai’s case, Amy and Andrew’s success in scheduling meetings is partly the product of well-designed interactions, plotted by a human like Diane. It’s the job of the AI Interaction Designer to make sure that anything their AI agents say is helpful, sensical, and coherent and that each interaction gets meeting participants closer to their goal: a scheduled meeting. Interaction designers create and maintain the personality of their AI agents and make the experience of interacting with them as natural and efficient as possible.
To keep x.ai’s AI personal assistant working as expected, Diane must determine the sort of response a given situation warrants and then compose the core elements of that response in Amy and Andrew’s “voice,” while accounting for all the possible consequences for those interacting with them—the customer and their guests as well as anyone helping to coordinate the meeting. Diane needs to keep in mind the user journey and the next action the user or their guest needs to take to arrive at a scheduled meeting as quickly and painlessly as possible. That means spending a lot of time thinking about things like dialog structure, phrasing, grammar, and, word choice.
“One of the more helpful exercises for me,” says Diane, “has been to ask myself, ‘Okay, if I was a personal assistant, trying to be the best personal assistant I can be, how would write this [response]?’”
Diane draws a comparison to traditional UX designers. Whereas a user experience designer for websites and apps might think about things like buttons and colors, Diane thinks about how simply written sentences impact user outcomes. For example, Amy and Andrew don’t ask open ended questions. When Amy and Andrew need information, they ask pointed questions to get that information as quickly and completely as possible without any ambiguity that could lead to extra back and forth. That is, Amy might ask, “Are you available at this specific time and date?” rather than “When are you available to meet?”
That’s not the same for all AI, though. A general purpose assistant like Siri might ask, “How can I help you today?” It’s the job of the AI Interaction Designer to think through each interaction and figure out how the AI should respond to effectively do what it is designed to do.
A day in the life
AI Interaction Designers own the personality of their AI agents. “We actually have a style guide for their personality,” says Diane. “Types of phrases they use a lot, things they will never say, or general personality traits.” The guide helps Diane ensure that Amy and Andrew always sound like the same assistant.
By the time Diane joined x.ai, Amy and Andrew had already been in development for a couple of years, so much of that foundational work was already done. Much of the personality was defined by her predecessor, whose background was in folklore and theater.
“By the time I joined, much of Amy and Andrew’s personalities had already been defined,” says Diane. Instead of creating a personality for the assistants from whole cloth, much of the challenge for Diane was getting what she calls the “weird, abstract knowledge of who Amy and Andrew are” transferred to her.
There are still plenty of challenges to keep Diane busy. At a very basic level, a vertical AI assistant must move through three stages to determine how best to respond to a request. First is Natural Language Understanding. The AI looks at what was said and asks, “Okay, what was said to me? What does this mean?” Next is Reasoning—now that the AI understands what was said, what’s the appropriate response? Finally, there’s Natural Language Generation, in which the AI compiles a response from a database of text snippets, authored by Diane and her predecessor.
For an AI Interaction Designer like Diane, most of her work falls into the second two steps. On a day-to-day basis, she’s crafting responses to help Amy and Andrew react to new situations. (What Amy and Andrew actually say in their emails to meeting guests isn’t precisely plotted for every scenario—the AI is deciding what to write based on many different variables—but the actual words come from the AI Interaction Designer.)
“I think about the traits we attribute to personal assistants,” she explains. “We expect them to be really professional and friendly, polite, very detail oriented, firm when they need to be. These are the things you’d want in a good human assistant.” Diane takes those expected traits and translates those into responses from Amy and Andrew.
Meeting scheduling is a lot more complex than people might imagine. Even though Amy and Andrew’s personality and scope of language are pretty well defined at this point, new circumstances or nuances within scheduling scenarios regularly arise. “The number of scenarios [for meeting scheduling] is exponential,” Diane says, “but within those scenarios it’s more like a closed universe of, ‘okay, Amy only talks about time and locations.’”
That’s the challenge of a conversational UI. Users can provide an infinite number of responses with an infinite number of phrasings, and they expect Amy and Andrew to know exactly how to respond. Unlike a traditional website or app where you may have at most 10 options at any given time, the open-ended nature of conversation means AI Interaction Designers have to account for an almost unlimited number of paths the user might take. Their goal is to make sure the AI keeps people on track and completes the desired task.
Diane’s job is also defined by the medium of Amy and Andrew’s responses, in this case, email. AI Interaction Design for other mediums (say, chat or voice) present a different set of challenges and considerations.
Becoming an AI Interaction Designer
The role of AI Interaction Designer requires an interesting mix of skills: part psychology (understanding the user), part linguistics (parsing language to its core elements), part computer science (translating plain English to code), and a healthy dose of creative writing. Diane, whose background is in cognitive science with a focus on computation, describes her job as a mix of the humanities and computer science. It offers an outlet for creative expression, but is rooted in logical thinking.
“I think having some understanding of psychology or linguistics along with some computer science,” would be helpful for this kind of role, says Diane.
While the job of AI Interaction Designer is a multidisciplinary one, the skills that AI Interaction Designers draw upon are many of the same ones that any designer utilizes. You need to be able to empathize with users and understand their motivations, and then think logically about how to guide them through a process.
Like many “jobs of the future,” this role is one that’s still being defined. “Even for me, the job has sort of evolved,” says Diane. “I’m finding myself doing different things every single day.”