Carnegie Mellon University researchers developed a new type of mixed-reality platform that can help children learn basic scientific concepts while experimenting in the physical world with the help of an AI agent.
Makerspaces and STEAM (science, technology, engineering, art and math) learning spaces are becoming common fixtures in schools, libraries and museums. Students of all ages can use them to learn coding, understand physics, build robots, learn sewing and more through self-led experiences with physical objects or virtual instruction on a tablet or device. But, when children interact with these tools, how much are they actually learning?
Nesra Yannier, a systems scientist at CMU’s Human-Computer Interaction Institute where she also received her Ph.D., and her colleagues Ken Koedinger and Scott Hudson, examined the effectiveness of learning spaces and developed a special “intelligent science station” to help improve children’s ability to learn underlying scientific principles. Their research, published in the International Journal of Artificial Intelligence in Education, showed that achieving active learning of science and engineering requires more than hands-on activity.
Meet NoRILLA (Novel Research-based Intelligent Lifelong Learning Apparatus), a mixed-reality system that includes a shaking table, a friendly cartoon gorilla and an intelligent tutoring system (that can be applied to many different content areas). This intelligent science station allows children to explore the physical world while getting personalized guidance from the virtual world.
Yannier wanted to understand if exploring and constructing with physical objects, like blocks, was enough to teach children scientific concepts. She wondered if having AI inquiry guidance on top of physical experimentation could improve children’s learning and lead to deeper understanding.
“Traditionally, we think that practice makes perfect, that people can learn by just tinkering and exploring — for example if you just keep practicing building you will become a better builder,” Yannier said. “But our research showed that without interactive guided-discovery based on proven learning techniques and methods, children don’t actually learn the underlying scientific principles and the skills needed to transfer to real-world construction and problem-solving.”
The experiment that Yannier and colleagues conducted had three conditions. In the Explore-Construct condition, children are challenged to build and construct things, like in a traditional maker space. For example, they may see a prompt on the screen asking them to build a tower that will not fall over when the table shakes. The children build the tower with blocks and press a button to start the shaking. If their tower falls, the children may try again, or they may move on to another challenge.
In the Guided Discovery condition, the intelligent science station prompts the children to go through the inquiry cycle with personalized interactive feedback from the gorilla character. As they do experiments, the system is able to observe and interpret children’s actions to accurately monitor and evaluate their predictions, experiments and explanations.
In the Combined condition, children alternate between the Guided Discovery and Explore-Construct conditions.
Yannier’s team found that the Guided Discovery and Combined conditions performed significantly better than the Explore-Construct condition. Children who played with the Intelligent Science Stations with AI inquiry guidance were better able to explain the scientific concepts they learned and apply them to real-world construction. In fact, the Combined condition, where they received reactive pedagogical interaction and guidance from the AI agent, led to 10 times more learning compared to the Explore-Construct condition, which is similar to what many maker spaces, museum exhibits and physical products are like currently.
“The most surprising result was that even though children were doing a lot of building in the Explore-Construct condition, their buildings did not improve at all. They practiced the most building in this condition, but the transfer to real-world building was the worst,” Yannier said. “When we added the guided inquiry with AI interactivity, we saw huge improvements.”
While learning from the intelligent science station, children are also having fun. Yannier said some of her favorite comments from children included:
“I wish all my science classes were fun like this!”
“I never thought something we do at school could be so much fun!”
And even, “This is the best day of my life!”
“We’ve been hearing from teachers that the kids are really persistent. Even if they fail they want to come back and try again,” Yannier said.
Koedinger, a professor of human-computer interaction and psychology, said that NoRILLA works because it utilizes proven learning methods and techniques.
“The learning science behind this is based on decades of CMU research in cognitive psychology,” he said. “This system is a real breakthrough in bringing intelligent tutoring systems from computer screen interaction into the real world where we can watch and help kids learn science while they are actually doing it.”
The intelligent science station provides additional benefits. It can be used in situations where teachers do not have training in science, guiding them along with their students. Additionally, the technology is not just valuable for children. The researchers said it could also be applied in manufacturing apprenticeships. For example, it could be adapted to teach factory workers how to operate a 3D printer or other equipment.
The NoRILLA system is currently being used in schools, museums and early childhood programs to teach STEAM concepts including physics, math, geometry, scientific curiosity and critical thinking. Yannier and her team are working on creating technology to expand intelligent science stations for different scientific apparatus, such as ramps, balance scales, falling objects and more. They hope to disseminate intelligent science stations to many more learning settings to benefit children from different backgrounds.