Online Learning in Emerging Markets – Machine Learning as an answer to skill-gaps?

How a Mumbai-based online education company is leveraging machine learning to create more meaningful learning experiences.

“Big Data is becoming a key part of the educational landscape, too. The same kind of learning machines that share our lives with us on social media, on our smartphones and on the web take on a special importance as they begin to occupy the educational field.”[1]

– Ben Williamson in Big Data in Education

Artificial intelligence is changing industries across the globe – from retail, to healthcare and manufacturing.[2] The education industry had long been extremely manual, thereby making the collection and use of data extremely difficult.[3] The advent of online learning has changed these dynamics and leading education-focused Venture Capital investment firms such as LearnCapital and OwlVentures focus their investment decisions primarily around the use of Machine Learning (“ML”).[4] The impact areas of machine learning in education are multifold and can be clustered into three key areas [5]:

All of these areas create significant learning efficacy as well as profit improvement opportunities for organizations.[6]

UpGrad, an online higher education company based out of Mumbai, India, with +250,000 learners across Emerging Markets currently leverages ML in two distinct ways:

  • Improvement of grader performance (Operations) – whenever a student submits an assignment, his/her assignment will be evaluated by an independent grader from UpGrad’s pool of +300 industry experts, who give feedback to students on any assignment submitted. An ML algorithm has been trained to identify and filter poor-quality feedback such as “Good job” or “Not good”, therefore, encouraging graders to provide more detailed feedback to students. This, in return, increases learning efficacy and customer satisfaction among students who appreciate detailed feedback.
  • Improvement of Lead-Scoring Mechanisms (Learner Acquisition) – in order to improve marketing performance, the company leverages ML algorithms to score prospective studentson the likelihood of enrolling into a course through data points such as website behavior, search behavior and attendance in events. This allows UpGrad to optimize marketing spends across different channels.

In the near future, as part of the job-placement product initiative, the company plans to implement further ML algorithms to improve career-support by matching open job descriptions with student skillsets. The algorithm is supposed to learn from the companies’ rejection or acceptance of a suggested candidate profile.

In the long-term, the company plans to use ML to provide students with content suggestions based on their skill-profile and performance in the online program. In this case, the algorithm will predict the likelihood of any student to pass or fail a certain assignment and learn from the actual performance. This data point will then be used to suggest relevant additional readings or assignments prior to the exam.[7]

Looking at the current and future plans of the company to leverage ML, I would recommend expanding the usage, especially in the Learning Efficacy category. Companies such as VIPKid in China have partnered with Microsoft to implement an ML platform that uses the learner’s webcam to assess his/her facial reaction to a new concept and sends prompts to the instructor on how the learner responded in order to improve the learning experience.[8] Similar technologies could be adopted by UpGrad to send personalized content recommendations. In addition to that, I recommend using feedback of companies who hired UpGrad’s learners to redefine the curriculum as well as the assignments used to test certain skillsets. The algorithm could use data collected from partner companies on the performance of candidates and ex-students in certain skill-areas and use these to improve learning efficacy further, thereby improving the ROI for both companies who hired from UpGrad as well as learners.

One key question that will influence technological advancements in education is the degree to which human interaction in learning can actually be replaced by machines – how much human touch is needed to teach skills? Can a machine replace the human relationship element, or will education always require the human touch to offer its full potential?

(798 words)


[1]Ben Williamson, “Big Data in Education – The digital future of learning, policy and practice”, (London, UK: Sage Publications, 2017), p. 10

[2]M.I. Jordan, T. M. Mitchell, “Machine Learning: Trends, perspectives, and prospects”, Science (Jul 2015), Vol. 349, Issue 6245, 255 – 260

[3]R. T. Kompen, P. Edirisingha, X. Canaleta, M. Alsina, “Personal learning environments based on web 2.0 services in higher education”, Telematics and Informatics (2018)

[4]Author’s interview with Tom Costin, Managing Partner at Owl Ventures, New York/ NY, November 12, 2018

[5]Maud Chassigol, Aleksandr Khoroshavin, Alexandra Klimova, Anna Bilyatdivona, “Artificial Intelligence trends in education: a narrative overview”, Procedia Computer Science 136 (2018), 16-24

[6]Tom Mitchell, “What can machine learning do? Workforce implications”, Science (Dec 2017), Vol. 385, pp. 1530 – 1534

[7]Author’s interview with Ravijot Chugh, Head of Product at UpGrad. New York, NY/ Mumbai, India, November 11, 2018

[8]Henry Kronk, “VIPKid is making moves,” eLearning Inside, August 30, 2018,, accessed November 2018


Machine Learning and Music: What About the Record Labels?


Square Inc.: Demystifying the Black Box of Small Business Lending

Student comments on Online Learning in Emerging Markets – Machine Learning as an answer to skill-gaps?

  1. It’s so interesting to learn about the impact of machine learning in education. Thank you for this great article !

    It’s incredible to think about how far we can go thanks to machine learning and I completely share your recommendation for Upgrade regarding expanding the usage of AI in the learning efficacy segment. I am convinced ML can greatly improve operational efficiency and quality of learning. I however do not believe that it could fully replace a human being. In fact, I am convinced that kids need an inspirational and kind figure to give them the willingness to learn and that is something a machine will never have.

  2. Wow, what an awesome idea. It’s interesting to think about the way that machine learning may impact the skillsets of individuals beyond the task at hand. Take, for instance, the feedback you are providing to graders who include cursory comments like “good job” at the end of an assignment. You are forcing them to be more critical, more analytical, in their feedback. Though the machine is doing this with the explicit goal of improving the UpGrad product, it could also result in encouraging the graders, generally, to be better critical thinkers. Traditionally, formal feedback has been dispersed and uncommon. Machine learning is providing constant feedback and may lead to faster individual improvement.

  3. This is a great article about Machine learning in the education space. I particularly liked how clear and structured this paper is.
    The graphic at the very beginning made it very clear about how ML can be applied in education space and the author elaborated on education operations and learner acquisition. It is clear that he has a great understanding of the industry as well as the specific company and it was interesting to learn about the methods they obtained clients.

    I also like his suggestions regarding Learning Efficacy. I liked the comparison and ‘lessons learned’ from ‘VIP Kid’ in China since that company has realized tremendous success in the past few years. Seeing the children of my friends using it and improving in language skills was amazing and I am looking forward to seeing how UpGrad can make a difference in higher education.

  4. To the open question of to what extent human interaction in learning can be replaced by machines — I think the answer is similar to what it is in other industries: we should identify the activities that machines will be able to perform better than humans (e.g., analyzing large data sets and spotting trends) and identify the activities that humans will have a distinct advantage in over machines (e.g., empathy, creativity).

    In an education context, I think this means that the role of a teacher in a classroom (or in a portal in the case of online education) will evolve significantly to be less about delivering the academic material and more about managing the classroom and the holistic learning experiences of the students. Teachers are an example of a profession that likely won’t be “displaced” by advancements in artificial intelligence, but rather, will simply be disrupted in terms of what activities are the highest value add for them to perform rather than the machines.

  5. I completely agree with your recommendation about learning efficacy. Analyzing student’s reactions in class can provide invaluable insight to the instructor on how well he/she is teaching, hence enabling improvement in teaching effectiveness.
    To your question on the need for a human touch in education, I believe that machines can help to teach hard skills and concepts. However, I believe that teaching soft skills and developing student’s emotional intelligence will be increasingly important in the future, and in this space, human touch is fundamental. Therefore, I don’t think that machines will be able to fully replace instructors/teachers in education.

  6. From the article, it seems that currently ML can only provide incremental values to education (e.g., assistance with grading, detection of students’ attention, data analytics etc.). It seems like ML will not fundamentally change the industry of education as teachers and administrators will still be a an essential part of the industry, and the heavy operations side of education will not be changed in the short term.

  7. You mentioned that the machine learning algorithm is identifying poor quality feedback such as “Good job.” What exactly is the machine learning to do? Is it gathering data and learning what constitutes a good versus poor quality feedback? What ultimately defines a good quality feedback? It would be interesting to tie the machine’s learnings to the actual performance of the students after they have received their feedback. In this manner, good feedback would be categorized as such only if the students’ performance improved. While we would be assuming that all students proactively change their behaviors to incorporate their feedback, the ultimate goal should be “effective” feedback as opposed to “good” feedback.

  8. Great article, and amazing work by UpGrad!

    Regarding your question, “Can a machine replace the human relationship element, or will education always require the human touch to offer its full potential?”; I believe this is not entirely possible, as there is a notable difference between training and education. Training relates to providing or enhancing a skillset. Education relates to the advancement of the human person, in its entirety. A computer won’t be able to reach far in the this field. However, I do think that there is a lot that ML can do to provide access to training to many more people; and to do it more effectively.

  9. One of my biggest questions when it comes to education is how these types of programs address the motivational component of learning. I think it’s incredible that these machines can propose assignments for students in anticipation of their weak points, but often students avoid the subjects that they find more difficult, so how do these type of programs create incentives for adherence? In other words, how do they provide the motivation to actually follow through on doing the assignments ahead of the exam?

  10. This is truly an insightful article. Using machine learning to grade written exams will remove inconsistencies. Throughout my schooling, I have been a strong believer that many teachers are inconsistent with their own grading, yet alone, one teacher vs. another. This makes being a student more difficult since you have to adjust your writing style from teacher to teacher and sometimes are unsure why a teacher is inconsistent with their grading.

  11. Education is currently evolving to be a highly-technological and sophisticated sector. Innovative business models in the EdTech space are emerging by allowing machine learning systems to function as a tool for educators to improve the classroom experience. In the near-term, I don’t believe the human interaction can be replaced by machines only. I think this technology needs to work in conjunction with teachers and professors. Even though the performance of the machine in the classroom can be superior to the human, I think both the machine and the human can work together and the effect in the learning experience will be synergistic. Students in the classroom face emotions that will be hard for the machine to feel.

Leave a comment