The Center for Data Innovation spoke with Kent Fuka, president and CEO of Querium, an Austin-based startup that uses artificial intelligence (AI) to power its StepWise STEM education intelligent tutor. Fuka spoke about the benefits of personalized learning and AI-based tutors.
Gillian Diebold: Why is personalized learning important in education?
Kent Fuka: Different students have different learning styles. Students learn at their own pace. Classroom instruction often addresses the average students, but leaves some students behind and fails to challenge advanced students. Self-paced systems help address the speed with which students grasp new material. One-on-one instruction can address learning differences for particular students but is impractical for deployment at scale. We conceived of StepWise after spending years working with districts and colleges that had struggled with student success in math. We saw AI as having the potential to act as a force multiplier for teachers, giving students most of the advantages of one-on-one human tutoring, but available to all students 24/7, 365 days a year. StepWise AI can help students at the instance of their confusion while working through a math problem.
Diebold: Can you explain the technology behind the StepWise tutoring system?
Fuka: StepWise AI combines a patented rules-based expert system with a powerful symbolic algebra system. The AI rules in StepWise are developed through our work with expert classroom teachers on the best methods for explaining particular math concepts, and on the best method to solve particular types of math problems. We currently support problems for over 500 math learning objectives, ranging from pre-algebra through calculus. The feedback that students receive is based on the wisdom of expert math teachers rather than what a symbolic math system might decide is the best sequence of steps to solve a particular problem.
Students can combine multiple math operations within a single step that they enter, just as when they would “show their work” on scratch paper. As a student enters a step in their solution to a math problem, StepWise quickly performs two checks. First, StepWise checks to make sure that the student’s next step follows from their previous work according to the rules of algebra or calculus. Second, StepWise also checks to make sure that the student’s step still leaves them on a path to a correct solution to the problem they are working on. If the student’s entered step passes both of these checks, then StepWise accepts the step as correct and gives the student positive encouragement to keep going. If the student’s step is incorrect, StepWise lets the student know that their step is not correct, but does not immediately tell them what they did wrong.
StepWise implements something called “productive struggle.” Students are free to ask StepWise for hints if they get stuck, and StepWise will give them a sequence of progressively more specific hints to get them back on track. Students also have the ability to request to see the complete sequence of steps for the current problem. StepWise also can generate a large number of similar math problems so that students can practice problem variants multiple times until they are comfortable that they have mastered a given type of problem. So, StepWise can be used for drill and practice, for homework, for quizzes, and for exams.
Diebold: What are the benefits of an AI tutor over other types of virtual tutors?
Fuka: AI has infinite patience with students and always gives consistent feedback to students 24/7, since AI never gets tired or frustrated. Virtual human tutors must be available at all hours, must be individually vetted, and cost more than AI solutions.
AI-based systems like StepWise are cost-effective for delivery to huge numbers of students. StepWise does not require advanced devices for every student. It works on Android and iOS devices, on Chromebooks, and on leading desktop web browsers. The AI in StepWise is hosted on servers at cloud data centers, so StepWise can be improved constantly without requiring users to download and install new software releases.
Diebold: What data do you use to train your AI, and how do you ensure the dataset is representative?
Fuka: Over the past five years we have had thousands of students across the US work on problems in StepWise. Our system keeps track of all of the work that students do in StepWise. We are able to constantly analyze that data to seek ways to improve the feedback that StepWise provides to students. For example, if we find that many students ask for a hint in a particular situation, StepWise AI can then be modified to give students advisory help in this situation without being asked by the student. StepWise can also be trained to detect situations where students are wandering into the “weeds” to help give them a gentle push to get back on track. Our student data comes from students in middle school, high school, and college, and from students across the country. We don’t use pure machine learning to train StepWise in the feedback it provides. Instead, we started with the work of expert classroom teachers from a variety of classroom situations. Our AI training comes from learning from the experiences of students who already use StepWise.
Diebold: What are some of the primary concerns you have encountered regarding AI in education?
Fuka: StepWise is non-threatening and does not attempt to be a teacher avatar. StepWise attempts to give the same feedback that an expert teacher would give, while also providing coaching encouragement, but it does not present a human face to students. We have not encountered resistance from teachers in the use of StepWise. Once teachers understand that StepWise was developed in collaboration with master classroom teachers, they readily accept StepWise to help them help their students succeed.
Teachers know that there aren’t enough hours in their day to provide personalized help to all of their students and are happy that they can gain insights into areas where students struggle the most in math. Administrators are happy that StepWise won’t break their budget and that it can work on the types of devices that districts already have. We’ve found that educators are very willing to accept AI-based solutions to critical learning challenges as long as the efficacy of those solutions can be demonstrated. We have data across middle school through college that suggests that students do significantly better in their math courses and on standardized tests with the help of StepWise.