Complex Process of Developing Intelligence in Robots

Estimated reading time: 4 minutes

Though language learning comes naturally to a child, encoding this complex process into a computer system is difficult; it lacks the physical and emotional connections to sounds and objects that are vital to the process of interpreting, conceptualizing, and understanding language.

During language learning, children will make mistakes, learn from experiences, change their behavior, and continually listen to and practice the language. Eventually their brains will make the correct associations between sounds, objects, actions, or ideas—and the language is learned.

Today’s computer systems lack neurons and empathy, two ingredients vital for human language learning. But Onyeama Osuagwu, a Ph.D. student in electrical and computer engineering, is working to build a system that can think and learn, and hopefully one day comprehend, through his research in three areas: how systems compute, how systems become intelligent, and brain-machine interfaces with these systems.

“I’m creating a methodology that’s trying to bring together interactions from sensory, visual, and auditory input, and hopefully piece them together,” said Osuagwu, who is co-advised by Steve Levinson, professor of electrical and computer engineering and full-time faculty member in Beckman’s Artificial Intelligence Group, and Lynford Goddard, associate professor in electrical and computer engineering and affiliate faculty member in Beckman’s Bioimaging Science and Technology Group.

As computing becomes more important in our digital world, from phones and computers to cars to space shuttles, developing an intelligent system, especially one that can learn from and communicate with humans, continues to drive scientific research forward. Not only will developing an intelligent system tell us more about how humans learn, but it will give way to a more robust robotic system that can interact with humans in a natural and organic way.

To test his artificial intelligence frameworks, Osuagwu uses Bert, the iCub humanoid robot, housed in the Language Acquisition and Robotics Lab at the Beckman Institute. Bert is an advanced humanoid robot with traits inherent in most people: joints that move, binocular vision, hearing, a sense of balance, as well as awareness of the position and movement of its body.

Bert, as the embodiment of an artificial intelligence system, is an important component in Osuagwu’s research.

“In Levinson’s group, we focus on mind, brain, and language. I’m teasing these things apart, to get to the language aspect,” Osuagwu said. “I’m trying to build parts of mind and brain in order to help the robot obtain language abilities on its own.”

“For us to learn, we develop a sense of understanding of the human condition—an empathy towards the feelings of others—so a computer system needs to have that as well. If the system doesn’t have a way of putting itself in the interactions, it can’t know the meaning of it,” said Onyeama Osuagwu.

Part of language learning is developing the ability to understand semantics—the meaning behind the words. Physical embodiment plays a role in this, as well, Osuagwu argues.

“We’re working off the assumption that, especially in electronics, one of the things we miss in trying to encode language learning in a computer is that a computer doesn’t have a physical form. It doesn’t have a direct mode of interaction or the means of gaining an understanding of the words,” Osuagwu said.

“For us to learn, we develop a sense of understanding of the human condition—an empathy towards the feelings of others—so a computer system needs to have that as well. If the system doesn’t have a way of putting itself in the interactions, it can’t know the meaning of it.”

For example, it’s difficult for a system to talk about the qualities of a pen if it’s not able to share and understand the feelings of the weight of a pen, how it writes on paper, or what it looks like.

Being human is really complex, Osuagwu says, so in order develop human intelligence in computing systems, he’s building a system that uses a “less is more approach.”

“We could try to mimic every single neuron, but that isn’t really replicating the human experience of learning, which is what we’re trying to do. When babies are born, they’re a blank slate—we don’t know anything about them. They become who they become over time through experience. So it’s kind of ridiculous to set up a robot and think that you know exactly what it’s going to do because that’s not how human beings work,” said Osuagwu. “So I’m giving Bert the basic tools, and letting the system develop through instinctual processes and learned experience.”

Bert has recently been featured in several videos, including this clip from General Electric (GE) and Robots 3D from National Geographic, now playing in theaters in museums, science centers, and cultural institutions worldwide. Visit the theater map to find a theater near you.