Moving beyond intelligent assistants?

    Computing scientist explores AI evolution.

    By Jennifer Pascoe on February 23, 2018

    An example–using two varieties of rabbits (orange and purple triangles) and wolves (blue triangles)–of what artificially intelligent agents will do of their own accord. In this demonstration, orange rabbits like the grass, whereas purple rabbits dislike the grass but like the orange rabbits. Wolves prey on rabbits. All agents must eat to survive or their markers will disappear.

    How is artificial intelligence affecting our lives? How long will AI agents remain our intellectual assistants? What’s next in their evolution? Given the recent surge in attention to AI and the increasing integration of AI in our daily lives, these questions are on many people’s minds, including Vadim Bulitko’s.

    “We can’t answer these questions by just pretending the concerns do not exist,” said Bulitko, professor in the University of Alberta’s Department of Computing Science. “There appears to be no easy way to answer. But if machines are going to become more and more autonomous, the only way to deal with these questions are to be at the edge of AI with active research.”

    One of the areas Bulitko explores in his research is the intersection of ethics, emotions, and machine evolution. He will be presenting a talk on his latest work, “Deep Learning, Artificial Evolution and Novel AI Behaviours,” at the upcoming AI, Games and Creativity event presented by LASER (Leonardo Art Science Evening Rendezvous) later this month.

    Mac Walters (left) and Vadim Bulitko (right) will both be presenting at LASER (Leonardo Art Science Evening Rendezvous).

    “I’ve worked on computational models of emotions like fear, distress, hope. In humans, these emotions are ‘programmed’ through our developmental, social, and cultural psychology. And we can program machines with computational models imitating these human emotions. But what if machines were to have their own emotions? Would humans even understand them?”

    Bulitko’s interdisciplinary work sees him collaborate with researchers from a variety of disciplines, including psychology and philosophy. His current collaborators include Marcia Spetch, professor in the Department of Psychology, John Simpson, digital humanities researcher with Compute Canada, and studio artist Marilene Oliver, professor in the Department of Art and Design, and coordinator of the local LASER events.

    Bulitko also actively encourages student collaborators from across campus, many of whom develop an interest in this research after taking his popular “Computers and Games” course.

    One of his most recent publications was a co-collaboration with Simpson, as well as undergraduate students Shelby Carleton and Delia Cormier in the Faculty of Arts, who subsequently completed internships last summer with the world-famous Edmonton-based video game company BioWare.

    The group collaborated on exploring the evolution of novel surprising behaviours of non-playable characters in video games. As scripted AI gets reused across multiple characters, there is a risk that players will become uninterested. The group’s research therefore has potential applications in the video game industry, as manually engineering interesting behaviors for non-playable characters can be prohibitively expensive.

    “When you play a video game, you’re controlling a character, but every other character is non-playable. These non-playable characters are controlled by AI. What emotions do they exhibit? And can these characters become increasingly complex with minimal human engineering? What’s the next step in this evolution?”