Toward emotionally intelligent artificial intelligence

    UAlberta computing scientists work toward conversation-capable digital companion for seniors.

    By Andrew Lyle on July 25, 2019

    If you’ve ever asked Siri a question, odds are you’ve been able to quickly tell you’re not speaking with a real person. Machine responses tend to be dull, generic, and devoid of emotion—but new research by University of Alberta computing scientists is aiming to change that, taking the key first steps towards machines that can carry a conversation.

    “Chat bots like Siri or Alexa are primarily used to look up information or do a task for you—answering questions in the shortest time possible,” said Osmar Zaïane, professor in the Department of Computing Science, co-author of the study and director at the Alberta Machine Intelligence Institute (AMII). “We envision a device that’s emotionally intelligent. Where an elderly person can say ‘I’m tired,’ or ‘it’s beautiful outside,’ or tell a story about their day—and receive a response that carries on the conversation and keeps them engaged.”

    Alberta's strength in AI

    In addition to his role as a professor in the Department of Computing Science, Zaïane is director at the Alberta Machine Intelligence Institute (AMII). Drawing from world-leading academic research at UAlberta and other institutions, AMII helps Alberta workers reskill and upskill for high-demand careers in artificial intelligence, and guides Alberta-based businesses as they implement artificial intelligenceI across operations and build their in-house capabilities and teams.

    And the results are promising. The team’s model was able to express responses that matched requested emotions in most cases—though Zaïane noted that some emotions, like surprise and love, were easier to express than others, and that this is just one of several steps in turning the vision of a digital companion into a reality.

    “In this study, we coached the program by telling it which emotion to express in its response. Our next study will focus on having the program independently decide on what emotion to express, depending on the persona it’s talking to.”

    An emotional response

    As Canada’s population continues to age, there are ever greater numbers of elderly Canadians wanting to continue living at home. But loneliness poses a very real concern for health and quality of life.

    “Loneliness leads to boredom and depression, which causes an overall deterioration in health,” explained Zaïane. “Studies show that companionship—a cat, a dog, other people—helps tremendously. The advantage for caregivers of a digital companion like this is it can also collect information on the emotional state of the person, noting if they are frequently feeling sad, for example.”

    But as one might imagine, developing an artificial intelligence capable of understanding when humans are expressing emotion and responding in an appropriate way is no small challenge. 

    “When an elderly person tells you something that’s sad, it’s important to respond with empathy,” said Zaïane. “That requires that the device first understand the emotion that is expressed. We can do that by converting the speech to text and looking at the words that are used. In this study, we looked at the next step: having the program express emotions—like surprise, sadness, happiness—in its response.”

    The paper, “Generating responses expressing emotion in an open-domain dialogue system,” was published in the proceedings of the 5th International Conference on Internet Science (doi: 10.1007/978-3-030-17705-8-9).