The Benefits and the Limitations of Machine Learning in Education

By Will McGuinness

Hardly a day goes by where we don’t hear about the latest development in Artificial Intelligence and Machine Learning. In the 1990s, IBM trained Deep Blue to take on Kasparov in chess. In the mid-2010s Google developed a program, called AlphaGo, to play Go and challenge the world’s best. AlphaGo trained on the knowledge and insights of experts by studying thousands of professional games, amassing lifetimes of experience and wisdom in just a few short months. In May 2017, AlphaGo defeated Ke Jie, the #1 ranked Go player in the world. It wasn’t even close. AlphaGo won 3-0. Just half a year later, Google announced AlphaGo Zero, which blew the old AlphaGo out of the water, defeating it nearly 90% of the time. The main difference? The new version, AlphaGo Zero, didn’t learn from the masters. It learned from itself, taking only the basic rules of the game as its textbook, then playing millions of games against itself.

As machine learning has advanced in chess and Go, it would be reasonable to think we can rely on it for great advances in education as well. But while machine learning brings great promise for the future of education, relying only on computers—even the best—would be a big mistake.

Reasoning Mind, the company I work for, has spent the last 18 years developing online math programs for pre-K through 8th-grade students, and while we have long recognized that computers can have great benefits in the class, we frequently must remind ourselves that there are limits to what they can do. As our co-founder Julia Khachatryan stated, “computers are only a tool to deliver the experience of learning from the world’s best math teachers.” We take the instruction modeling approach, which attempts to model the activities and decisions of some of the world’s most effective teachers in its artificial intelligence—like how they remedy gaps in student knowledge, how they decide the difficulty level of problems to give students, and how they scaffold explanations. We’ve found, however (as you may expect), that a successful implementation also requires teachers to take an active role in student learning.

Based on our experience, we’ve put together the following summary of some of the benefits and limitations to look out for as artificial intelligence enters the classroom.

The benefits:

  • Artificial intelligence in the classroom allows teaching to be differentiated and personalized. Computers can deliver customized lectures to each student, taking a significant workload off instructors and freeing them up to work one-on-one with students or fix more complicated problems in students’ misconceptions. Teachers do not need to teach to the middle of the class, and all students’ needs are met. When the quality of the curriculum is high and when the approach is implemented with fidelity, it has translated into results.
  • Online programs can grade work instantly, giving teachers access to formative data immediately. Educators no longer have to wait until a quiz or test to find out that a student is struggling and can correct misconceptions before they solidify.
  • Computers can also be leveraged to improve student engagement. For example, researchers measured 89% time on task in Reasoning Mind’s Foundations program when implemented with fidelity. Over the course of a school year, this can add up to an additional 40 hours of math instruction.

Of course, while artificial intelligence has already surpassed human skill in chess and Go, teaching is a far more complex activity. While games have a rigid structure with a strict set of rules, as educators we know that teaching students is pretty much the opposite situation. The main limitations behind the usage of machine learning in the classroom tend to revolve around this difference:

  • As Steigler and Hibert explain in The Teaching Gap, learning is an inherently cultural process. Computers can help streamline and improve this process, but they cannot replace the cultural element of learning, which can only come from another human.
  • In a blended learning classroom, the computer and the teacher both have an essential place, doing what each can do best. According to Reasoning Mind’s CEO, Gregg Fleisher, “Teachers know and understand each student in a way that computers simply can’t. The program provides a strong curriculum and differentiation, while the teacher ensures that every student receives the personal instruction and support needed to progress.” It’s hard to imagine even the most advanced computer program replacing this human understanding, even if the computer provides detailed information on what the student knows and doesn’t know.
  • Learning is more than downloading knowledge or passing an exam. As identified in Getting Smart’s Ask about AI, developing a sense of purpose is critical to self-directed learning. While computers can provide suggestions about what students like, developing this purpose and instilling it in others is an exclusively human activity.
  • Another important aspect of education is teaching about morality and bias. Morals and biases exist in a human world, and can only be understood in that context. While computers may seem to not have morals and be unbiased, as machines begin to decide who should get loans or who should go to jail, AI can easily pick up our human biases if unchecked.

We should be acclimating ourselves to the ideas of AI and Machine Learning to education. It will doubtless bring important, needed improvements in education, but we would be smart to not overlook the wisdom humans have developed and passed down over generations in education. The classroom is a complex environment and will always contain an essentially human element that no computer can replace.

For more, see:

Will McGuinness is a Product Support Manager at Reasoning Mind. You can connect with him on Twitter: @WillM_RM.

Stay in-the-know with all things EdTech and innovations in learning by signing up to receive the weekly Smart Update.

Guest Author

Getting Smart loves its varied and ranging staff of guest contributors. From edleaders, educators and students to business leaders, tech experts and researchers we are committed to finding diverse voices that highlight the cutting edge of learning.

Discover the latest in learning innovations

Sign up for our weekly newsletter.



I think where Artificial Intelligence and Machine Learning can help is the assessment of comprehensive writing and speaking questions. If you see grading MCA type questions are very easy, but when it comes to task-based or questions it's really hard. 
Here is a site that is doing great in this aspect. is one of kind application that can actually analyse and grade comprehensive writing and speaking questions.
Even, this is a great analytical tool for both teachers and parents.


Machine learning is already being used in applications like Google classroom, these technologies are especially useful in the lockdown countries have because of the COVID-19 pandemic

Leave a Comment

Your email address will not be published. All fields are required.