Equipped with machine learning, Kiwi can adapt its teaching plan according to the needs of each child.
Many children with autism suffer from learning disabilities and developmental delays, including problems with social interaction, communication and behavior.
This makes learning new skills a major challenge, which is not usually handled well in traditional schools.
Upon learning of the obstacles faced by children with autism, researchers developed Kiwi, a small personalized learning robot.
As part of the study, the University of Southern California team installed the “socially assistive robot” in the home of 17 children with autism, to see if they could measure the child’s interest in a particular task.
For a month, the children who participated in the study were challenged with math games on a tablet, while Kiwi provided feedback and instructions, such as giving tips after a wrong answer or congratulating them after a correct answer.
As the lessons progressed, the algorithms adjusted the feedback of the robot and the difficulty of the games for each child, according to their individual needs. To adapt them, the device used machine learning.
The incredible results
At the end of the month, scientists analyzed the involvement of the participants and noticed that the robot had 90% accuracy in detecting the child’s interest. This was confirmed with a series of data like head position, task performance, eye contact with Kiwi and voice tone.
At the end of the study, all children improved their mathematical skills, and 92% of them had positive advances in social capacity. The results were published in Frontiers in Robotics and AI and Science Robotics.
It can become more efficient than human teaching
Robots assisting children with autism is something that has been studied. According to the USC researchers, this is more effective if the robot can react to each child’s behavior differently, and that is a difficult task for most robotic systems.
“If you think of a real learning environment, the teacher is going to learn things about the child, and the child will learn things from them,” said Shomik Jain, the study’s lead author.
“It’s a bidirectional process and that doesn’t happen with current robotic systems. This study aims to make robots smarter by understanding the child’s behavior and responding to it in real-time.”
For instance, this teaching method could be more efficient than human care, as it does not depend on places and times, but critics of the methodology fear that these benefits may also become a risk.
“Artificial intelligence methods cannot and should not be used as cheaper substitutes for treatment with human doctors,” said Alena Buyx, professor of ethics in medicine and health technologies at the Technical University of Munich, in a statement.
“Human therapists are crucial, but they are not always available or accessible to families,” explained Kartik Mahajan, co-author of the study. “That’s where social worker robots like that come in.”