Improving Designs to Make Intelligent Agents Smarter
By Stacy Kish
Artificial intelligence (AI) is having a moment; however the uncertainty surrounding any disruptive technology can be both exciting and threatening. Sara Moussawi, associate teaching professor in CMU’s Information Systems Program, has focused her research on filling in a small piece of this larger AI puzzle.
For the past decade, humans have befriended and welcomed AI into their homes. First, we welcomed personal intelligent agents (PIA) like Apple’s Siri on the iPhone 4S in 2011 and three years later with Alexa in Amazon’s Echo. Chatbots, a type of AI conversational intelligent agent (CIA), have also become more common as people interact with customer service at companies.
Moussawi is interested in understanding how people perceive and use intelligence agents, specifically PIAs and CIAs. These devices use natural language to engage with users and help them complete tasks.
In the industry, PIAs and CIAs can be grouped according to how they are used. These systems are designed to act on the user’s behalf to filter information, automate simple and complex tasks and collaborate with the user to solve problems. These agents learn and adapt to their environment and a user’s preferences. This form of AI also exhibits human-like characteristics that ease the integration of the technology into the user’s daily life.
While Siri and Alexa have been at the leading edge of the AI revolution, the use and integration of PIAs and CIAs are likely to expand as algorithms improve, responding more effectively to complex tasks and sensory feedback as well as increasing awareness of the task-user context.
With more than a decade in the marketplace, Moussawi was curious about what keeps people engaged in using their devices.
“I am interested in understanding how [people] use these systems so we can improve them from a design perspective to produce a better user experience,” she said.
Her work builds on the unified model of information technology continuance of use. Using data obtained from large (cross-sectional) studies of end users, she examines the AI system characteristics, like perceived intelligence and human-like qualities. She also examines the level to which people view these technologies as an extension of themselves and whether that plays a role in increasing their engagement with and perceived usefulness of these agents.
In her past work, Moussawi has found that people are more inclined to adopt PIAs if they find them useful and engaging. She found that while the usefulness of the technology persists, the effect of engagement seems to wear off. She found users continued to use the technology when the AI’s performance exceeded the person’s original expectations. It also doesn’t hurt if the agent possesses human-like qualities for adoption and continued use. Moussawi’s work in this area builds on the research of CMU’s Sara Kiesler, which examines how to control and customize the technology to give the impression that the PIA is an extension of the person and their capabilities.
This stream of data is critical because through technological advances, devices evolve. As they do, Moussawi believes designers should focus on system features that enhance autonomy, proactiveness and natural language interactions to ensure the engagement aspect of the human-intelligence agent connection remains strong.
Moussawi believes that future designs should take the user’s personal needs and context into consideration. She advocates for developing systems that better react and engage with the user’s unique behavior. For example, Moussawi is currently examining how to enhance overall creativity through a user-AI collaboration. In particular, her team is exploring creativity dynamics to better understand how it shapes human-human, human-AI and all-AI teams. She hopes that a better understanding of creativity dynamics will help improve human-AI teams working on creative tasks, like designing art or new consumer products.
“I see exploring [creativity] as an opportunity, and I am really excited to investigate how user-AI collaboration dynamics can shape creativity,” Moussawi said. “Better understanding how to make this possible will allow for a wide integration of these technologies into creative work contexts.”