Just after high school and before my first year in college, I taught mathematics in a rural village in Kenya. After a few months, the mayor of the town invited me to his home. I sat in an anteroom that had a swept dirt floor and freshly whitewashed walls. He left me to sit alone with a small battery-operated radio. The radio played loud static. There were no radio channels available. After 5 minutes or so, he returned to ask me what I thought of his radio. Being polite, I said, “Very nice, Mr. Mayor!”
Why did he do this? He had be aware that it was only noise. To this day, I think he was showing me that he was looking towards the future. The technology was an aspirational symbol. “I am taking this town into the future!”
We see an analog of this when school districts spend large funds to ensure students have computers, even though there is limited infrastructure to support the technology and there is no compelling programming. It is aspirational – a way to tell parents and children that they are looking to the future.
In the last few years, people have started to look to the future of Artificial Intelligence in education. I would like to clarify what that future could be, so the educational enterprise can set appropriate (and functional) aspirations.
By way of background, artificial Intelligence (AI) automates tasks that humans perform . AI applications range from reading x-rays to driving a car to understanding what you say to your phone. In education, this leads people to think about automating teaching. Indeed, the earliest AI systems simulated human tutors. For example, the computer assigns a child an algebra problem. The computer tracks whether the child follows the correct steps. When a child makes a mistake, the program intervenes to help. The key intelligence is the ability to track student performance over time to recognize what sub-skills the child has mastered or missed. It can then move the student backwards or forwards in the content. It is similar to what many tutors do, where they diagnose the gaps in student understanding, and back up to cover the missing material. The expression “adaptive educational technology” refers to the software’s abilities to move a learner forward or backward in the content (or, the test questions, in the context of adaptive testing). Notably, unlike the best tutors, adaptive technologies do not currently change the method of instruction to match individual student needs. There is room for improvement.
Intelligent tutors have shown positive effects, especially for domains that involve the application of rule-based procedures, such as mathematics, programming, and logic. They are a smart idea. Even so, I do not think imitating and eventually replacing the teacher is the right aspirational goal for educational technology. Rather than focusing on the efficiencies gained from automation, we should think about how AI can support fundamentally new ways of learning made possible by the computer. It is only in the past 20 years that students could learn by exploring computer simulations of natural phenomena, such as glacier formation, bird flocking, and gravity. How can AI support this, and other, new forms of learning? This is an important question to explore, because many current applications of AI dictate what students should do, often by giving them step-by-step instructions within a very fixed instructional model. Ironically, these applications make students more like computers, so they too follow step-by-step instructions. This is an old model of instruction largely suited to a different time in history. What we really want for the future is students with rich knowledge bases who can make smart choices about what and how to learn.
Done right, AI can be exceptionally powerful in helping students to become independent learners, when there is no teacher available to tell them exactly what they should be thinking and doing. How might it do this? There are already signs. AI can provide rich feedback. For example, work at Stanford is making it possible for children with autism to gain feedback about the emotions of other people. Children wear smart glasses that can detect emotions from the faces of other people. The glasses can also show messages to the children, for example, by labeling a facial expression as, “Happy.” This helps the children learn to recognize emotions. AI systems can also observe whether students are using good design-thinking strategies and exploring alternatives, which is often hard for a teacher to monitor in a classroom of 25+ children. AI systems can even help teachers keep track of who is speaking the most in class, which alerts teachers to the possibility that they are talking too much of the time or that some groups of students are being systematically shut out.
More broadly, a great promise of AI in education is that it will provide a steady stream of evidence and feedback to students, teachers, and even policy makers. With new sources of evidence, educators and learners will finally be in a position for continual improvement. Education has always been a data impoverished field. Once-a-year standardized tests are coarse metrics, and it is incredibly demanding on teachers to provide regular and rich feedback. A kindergarten teacher once asked me to finish the sentence, “Practice makes ______.” I fell for it and said, “perfect.” She said, “No, practice makes permanent. Practice with feedback makes perfect.” AI systems are increasing in their abilities to provide meaningful feedback. This ranges from feedback to improve writing, to improve critical thinking, to improve learning from open-ended tasks. AI systems can also provide feedback to teachers, and not just about their students but also about their teaching. Systems are being developed that can interpret transcripts and even raw video footage of teaching to help teachers notice missed opportunities or suggest new ways of asking questions. AI systems can help policy makers and administrators, for example, by discovering early warning signs of potential failure.
My aspirational version of Artificial Intelligence in education does not replace the teacher with robots. Instead, future AI will work with teachers to get information they and their students need to learn and improve. This human-centered approach to AI puts learning in the hands of people, and it puts AI in the role of helping people do what they still do best – learn when they have useful feedback.
Daniel L. Schwartz, PhD, author of The ABCs of How We Learn (W. W. Norton, 2016) is the Dean of the Stanford University Graduate School of Education and holds the Nomellini-Olivier Chair in Educational Technology. He is an award-winning learning scientist, who also spent eight years teaching secondary school in Los Angeles and Kaltag, Alaska. His special niche is the ability to produce novel and effective learning activities that also test basic hypotheses about how people learn.