In STEM AutoTutor, an artificially intelligent computer program for teaching middle school math and science, virtual characters help students think through their goals and the different approaches they might try to solve a problem as well as the procedures they need to know.

Schools may be critiqued as “factories,” but robots aren’t going to replace human teachers any time soon. Still, that doesn’t mean that artificially intelligent systems won’t transform education just as they are changing a variety of fields and practices, from the way oncologists diagnose cancer to how lawyers analyze cases.

Intelligent-tutoring systems like ALEKS (for Assessment and LEarning in Knowledge Spaces), Cognitive Tutor, and a new program in development by IBM’s Watson initiative are starting to expand in K-12 education, and experts argue that teachers need new training not only to use intelligent systems in the classroom but also to prepare students for careers in increasingly technology-integrated fields.

“Any skill that a computer can teach is going to be done by a computer in the workplace, and that’s something people don’t think about enough,” said Christopher Dede, an education and technology professor at the Harvard Graduate School of Education. For that reason, he said, teachers can use computer programs not simply to replace pieces of their instruction, but to model for students how to work with technology professionally. “It changes the skills people need to be employed. AI changes teaching, yes, but more important than that, AI changes the goals and purposes of teaching,” Dede said.

SEE ALSO

SCHOOL & DISTRICT MANAGEMENTIN THEIR OWN WORDSAwake at 2 A.M., Agonizing Over Life-and-Death Decisions: A Superintendent's Story

Artificially intelligent tutoring systems, or ITS, are computer programs that model students’ psychological states as well as their prior knowledge to personalize instruction for them. As students interact with them, the programs collect data about how the students approach each problem, when they are likely to get frustrated, and so on. The system evolves in response to the people who use it, to improve the lessons and assessments it presents.

“In the tutorial, you have a conversation, and the tutor-machine knows an awful lot about your background in the course and can build on that in a way you can’t in a regular classroom,” said J.D. Fletcher, a researcher with the Institute of Defense Analyses and a primary developer of the U.S. Navy’s Digital Tutor ITS, which is used to train Navy staff for technical jobs in the force, such as troubleshooting systems on a ship. “Some of your kids will take one day what it takes others four days to learn. In a traditional classroom, the fast students are left twiddling their thumbs. ... If you have [an ITS] engaging in a conversation with you, the tutor can just keep piling on the questions to you that are progressively more difficult.”

Such tutoring systems have had mixed effectiveness over the years, but more recent programs have shown significant promise. A 2014 meta-analysis of several different ITS found they were as effective in helping students learn as a person leading one-on-one or small-group instruction and more effective than full-sized teacher-led classes, workbooks or textbooks, or traditional computer-based instruction.

A separate evaluation of the Navy’s intelligent- tutoring system found those who used it outperformed those using standard technical training—not just on other tests, but also on practical troubleshooting exercises. Navy staff who had been trained using the tutoring program also attempted more challenging problems and tasks than students who had been trained in other ways.

“Whether the [ITS] is like a human or not doesn’t matter if it works better in some ways,” said Kenneth Koedinger, a professor of human-computer interaction and psychology at Carnegie Mellon University, who helped develop another artificial-intelligence teaching program, Cognitive Tutor. “In a system that big, you can replicate a strategy in a reliable way and try it against a separate strategy and see what works better, very quickly. You can’t do that with a classroom teacher.”

Yet across the board, researchers developing the programs argue that teachers are critical to making the systems work effectively. “These intelligent-tutoring systems, people always worry they are going to replace teachers,” said Art Graesser of the University of Memphis, who developed the AutoTutor and ALEKS systems. “I would argue they don’t, but they take over a lot of tasks teachers don’t like to do: to grade papers, to cover the same skills over and over. ... In ideal systems, teachers will be creating the material, working with students on broader life goals.”

Yet Graesser, Koedinger, and others all agree that teachers need more specific professional development in how to integrate intelligent systems into their classrooms.

“Teachers can say, ‘Oh, the tutor teaches X, I teach Y.’ That does not work,” Dede said. “It’s actually a very rich kind of sharing of responsibility between the teacher and the machine. The people who build the intelligent-tutoring systems often don’t understand this very well and don’t provide support to teachers to implement them.”

Chalapathy Neti, the vice president of IBM’s Watson initiative, agreed. The Watson intelligent system has already been used to help accountants at H&R Block unravel tax law and to help oncologists at the Mayo Clinic diagnose cancers, but the system is just being launched this year in higher education and preschool. Neti said the group is piloting cautiously, while keeping teachers in the development process.

“The time a doctor has with a patient is very episodic and sporadic, but a teacher is with the student every day. We need to lay a foundation for the learner,” Neti said. “We don’t think of ‘AI’ as artificial intelligence, we think of it as ‘augmented intelligence,’ and we are thinking of how we improve this partnership” between teachers and computers.