Is machine learning in education the new textbook?

Education technology company Knewton strives to deliver personalized learning experience for college students using machine learning technology. Can Knewton's product really replace textbooks as the company aspires to?

Imagine taking a microeconomics class when the course material in front of you at any time is instantaneously tailored to your knowledge base and gaps. You no longer will be bored out of your mind because the class is too easy, nor will you be so discouraged because the class is too hard. Moreover, you will only need to pay a quarter of what you would pay for an expensive textbook. Such tailor-made learning experience is what Knewton strives to deliver with machine learning.

Knewton is an adaptive learning education technology company, using machine learning to predict college students’ learning gaps and recommending personalized pedagogies to fill in these gaps. Knewton’s core product Alta, the company deemed as a textbook replacement, is entirely built upon machine learning technology. Although teachers have adopted individualized pedagogies for students’ different learning paths for a long time, machine learning enables such personalized learning to scale quickly through tools such as Alta.

There are three key features within a machine learning algorithm: “feature extraction, which determines what data to use in the model; regularization, which determines how the data are weighted within the model; and cross-validation, which tests the accuracy of the model”[1]. Alta’s development and improvement encompasses all three key features:

First, Knewton team has initially worked with large publishers, such as Pearson, to collect and label learning data from their students, such as mouth clicks on concepts, wrong answers to specific questions etc. Through labeling millions of learning data, Knewton extracts features of students’ learning profile. Second, Knewton develops its “Knowledge Graph” in any given course, mapping out the relationships between the learning objectives and various competency features captured from learning data[2]. During the regularization process, the algorithm determines which competency features are more likely to predict how much the learning objectives have been achieved. Based on progress on learning objectives, Alta predicts a student’s specific knowledge gaps and decides on which material a student should see next to address these gaps. Third, Alta cross-validates its predictions with past students’ competencies and predicted learning gaps based on the same data features and concludes whether its prediction is consistent with out-of-sample data.

As Knewton’s entire product is built upon machine learning, the key issues that the company is facing are all associated with features of machine learning technology. In the short term, Knewton needs to address student data privacy protection issue[3] and needs to prove casual relationships between learning outcomes and Alta usage beyond mere predictions. First, Alta is predicated on collecting millions of learning data from students. Such data is sensitive and personal and requires rigorous protection. Knewton management has disguised all the student personal information, such as name, gender, and race. Second, machine learning is used to predict knowledge gaps based on students’ current knowledge, but the current algorithm alone cannot generate any causal relations about why students have these knowledge gaps[4].  Moreover, there is no further evidence to prove that filling students’ knowledge gap using Alta can help students master their materials. To address this issue, the management has commissioned Johns Hopkins University’s Center for Research and Reform to study causal relations between students’ current knowledge base and knowledge gaps as well as between filling their knowledge gaps and mastering materials[5].

In the long run, Knewton plans to convince more stakeholders not only in higher education, but also in corporate training and K-12 schools to use Alta as a replacement for traditional course materials[6]. Knewton believes that high-quality course materials are no longer exclusive to large textbook publishers. Knewton can source free or low-cost materials and develop the corresponding Knowledge Graph and adaptive learning algorithm to meet broader training and development needs. To attract broader potential clients, Knewton offers various service models, such as learning management system integration, to encourage more schools and employers to experience the product.

Currently Knewton functions best in domains with more fixed rules, such as accounting and finance. However, it has limited capacity for less standardized courses, such as management and creative writing[7].  To better meet diverse customer needs in the short term, Knewton should refine its algorithm to accommodate subjects with more nuanced features and patterns.  Moreover, the current product focuses exclusively on academic subjects while adaptive learning technology can also be applied to vocational trainings. In the long term, if Knewton wants to position itself to provide a wide range of personalized learnings, the team can consider developing algorithms for scalable and personalized vocational training, such as coding and user-interface design.

While adaptive learning based on machine learning has huge potential to improve students’ learning outcome and improve educators’ effectiveness, many questions remain: will Alta eventually eliminate the need for textbooks? How will traditional textbook publishers, such as Pearson, react to such product innovation?

 

Word count: 800

 

 

[1] Yeomans. “What every manager should know about machine learning” . Harvard Business Review Digital Articles (July 7, 2015).

[2] The Knewton Platform. https://www.knewton.com/the-power-of-altas-adaptive-technology/ (2018)

[3] Cavanagh. “Knewton Launches Partnerships With Ed. Companies in China, Japan”. https://marketbrief.edweek.org/marketplace-k-12/knewton_launches_partnerships_with_ed_companies_in_china_japan/. EDWEEK Market Brief. (August 15, 2015)

[4] Johnson. “What Can Machine Learning Really Predict in Education.” https://www.edsurge.com/news/2018-09-26-what-can-machine-learning-really-predict-in-education. EdSurge (September 26,2018)

[5] Wan. “Knewton’s New Business Attracts New $25M in Funding. But Some Things Don’t Change.” https://www.edsurge.com/news/2018-08-21-knewton-s-new-business-attracts-new-25m-in-funding-but-some-things-don-t-change. EdSurge (August 21, 2018)

[6] Young. “Hitting Reset, Knewton Tries New Strategy: Competing With Textbook Publishers.” https://www.edsurge.com/news/2017-11-30-hitting-reset-knewton-tries-new-strategy-competing-with-textbook-publishers. EdSurge (November 30, 2018)

[7] Conklin. “Knewton Review.” https://journals.aom.org/doi/10.5465/amle.2016.0206. Academy of Management Learning & EducationVol. 15, No. 3. (2016)

Previous:

DMs for Dollars

Next:

Machine learning will revolutionize the future of insurance

Student comments on Is machine learning in education the new textbook?

  1. Throughout my school years, it is always painful for me to read the thick and boring textbooks, so I personally find this article about Knewton very intriguing. This business model will help students to study more efficiently and thus save much time. Although currently Knewton functions best in domains with more fixed rules only, it still has a huge market potential. I think it has a strong competitive edge in the market of preparation for standard tests, e.g. CPA for accounting, CFA for finance, TOEFL for English language learning, GMAT. These tests are standardized globally and have huge customer bases, and Knewton can help test takers to better prepare for the tests in an efficient and effective way.

    To answer your last question, as a revolutionist against the traditional textbooks, Knewton certainly poses competition and pressure on conventional players such as Pearson, but it also offers opportunities. Pearson alike companies already have interactive learning platforms for students to read digital version of the textbooks and complete assignments online. If they are able to incorporate Knewton’s adaptive learning education technology into the existing interactive platforms, these companies will actually be able to improve the user stickiness.

  2. This is an extremely interesting note! In addition to your points, I think that it will be interesting to see Knewton’s impact on student motivation levels. By this I mean that its adaptive teaching method might incentive students to work harder to get the right solutions, in order to avoid longer teaching sessions and additional homework materials to fill-in their knowledge gaps.

  3. Thinking back to the Aspiring Minds case, I wonder how much incremental value Knewton is adding with machine learning. In Aspiring Minds, some of us raised the question of whether the tests were more predictive than regular standardized testing. Similarly, I wonder the value of Knewton vs. a student being able to select a class that is closest match to their level vs. massive open online courses (which are usually free).

    In addition, how do we know that Knewton is correctly identifying learning gaps for students? It seems like they are just running a multiple choice pre-test, but how does it know whether a student got a concept right through luck vs. actually understanding the material. Unlike Aspiring Mind which is evaluative, there is much higher risk with Knewton since students could end up completely misunderstanding basic concepts because they happened to guess right on the pre-test.

  4. Interesting read – one of the major problems in education is that educators are not really answering the question “why students have these knowledge gaps”. I think this is an important question that any new tool like Knewton should try to address, but only by answering this question, can a tool like this really tailor its teaching style specifically for the student. That being said, I think there is a lot of promise with apps like Knewton, specially since textbooks are antiquated and unnecessarily expensive.

Leave a comment