Riiid: Disrupting education with AI, one test at a time

A startup from South Korea wants to change the way students approach testing

by Max Smolaks 15 January 2020

What if I told you that a simple smartphone app could considerably improve a student’s standardized testing scores?

That’s the promise made by South Korean startup Riiid, which develops ‘santA.Inside,’ an AI-based tutoring engine for the growing global test-prep market. The company claims its product can far outpace human tutors at a fraction of the cost.

The first application of the engine focused on the Test of English for International Communication (TOEIC) – a popular test in Korea, Japan and Thailand, designed to measure achievement in using English in a business setting.

YD Choi, Riiid

According to Riiid, after just 20 hours of training, the product called Santa TOEIC improved users’ scores by 124 points – with the total score measured on a scale from 10 to 990 points.

Next, the company is tackling something much more serious – the SAT, a standardized test widely used for college admissions in the US.

To find out more, we talked to YD Choi, AI tech lead at Riiid, who presented his work at the AI Summit New York.

AIB: Riiid is an AI education company. What are some of the most promising applications of AI in the education sector?

YD: AI is widely used in various domains including healthcare, speech recognition, automobile, face recognition, and so on. However, cutting-edge machine learning and deep learning technologies are not sufficiently applied to the field of education. Riiid pushes this direction of AI in education (AIEd) further by applying state-of-the-art models to challenging tasks such as score prediction, dropout prediction and optimum learning path generation.

Our technology is proven to help students preparing for standardized exams like TOEIC and SAT. For example, it is estimated that users who used SantaTOEIC for ~20 hours have the score gain of 130 points on average. We also share our models and datasets to the AIEd research community, facilitating the academic development of the subject further.

AIB: You develop AI-based tutoring software. What does it do, exactly? What are the benefits of an AI tutor?

YD: Our AI tutor helps students to prepare for standardized exams. Currently, we launched SantaTOEIC and SantaSAT for TOEIC and SAT respectively. When a student signs up our service, our tutor gives 6~10 diagnostic questions and estimate his/her exam score based on our state-of-the-art neural network models. After that, the tutor recommends questions and lectures that maximize the probability of attaining the user’s target test score.

Since the tutor automatically assesses students and suggests contents without any human intervention, the cost of teaching students decreases, and teaching becomes easily scalable. Also, students do not need to spend a lot of money and time to prepare for exams anymore, as the cost is affordable and an efficient learning path is designed for each individual.

AIB: What machine learning technologies are used at Riiid?

YD: Both NLP and RL-based models are used in our products. Our dataset consists of students’ learning histories, including question-solving and lecture-watching logs. NLP-based models are appropriate for handling such time-series data by considering such user behaviors as tokens.

Our Transformer-based model shows the state-of-the-art performance on Knowledge Tracing, the most fundamental and popular task in AIEd. Also, we use similar models to predict whether a student will drop out of a session or to simulate virtual students for training RL models.

We use RL-based algorithms for content recommendation models. The ultimate goal of AI tutor is to provide educational content that fits the needs and abilities of each student. We achieve this by maximizing the probability of the student achieving his/her target exam score. Our algorithm is based on RL with a suitable reward function that incorporates the goal achievement probability and other factors involved in the learning activity of a student.

AIB: Do you believe most of the traditional, classroom-based courses will be replaced by online learning, or something else?

YD: Although online learning and AI-based learning showed efficacy and replaced some parts of classroom-based learning, their current ability does not cover the whole range of education that includes human communication and non-quantifiable qualities. With current technology, we expect AI tutors to replace classroom-based learning by human teachers in case of education with measurable goals, such as test preparation or teaching objective knowledge.

AIB: Riiid has registered a total of 41 registered patents and patent applications. How important is R&D for the company?

YD: At Riiid, a team of 40 (out of 80 employees total) expert professionals are dedicated to R&D. Those are from prominent Korean and overseas universities (UC Berkeley, Yale, Seoul National University, KAIST, POSTECH, Korea University, etc), all of whom hold master’s or doctoral degrees in mathematics or computer science.

Striving to study and research cutting-edge AI technologies (Transformer, Attention, Bert, and others) with an aim of further enhancing and optimizing our solution, R&D team continues to publish research papers at the top global AI conferences even to ensure firm’s technological novelty. As you mentioned, we protect our technology assets by patent registration while forming high entry barrier against potential competitors. As an AiEd startup, R&D is our core and pride. It is the area that most of the firm’s investment is made into, and something that truly differentiates Riiid from other traditional education companies.