New Blog Post: Is AI Enhancing or Replacing Education?

Why do we care so much about using Generative AI in education? Many of us are strong supporters, many are strong opponents, and virtually no one is indifferent. Why didn’t we care that much about, say, blockchain in education? Or cybersecurity in education? Most of our concerns about GenAI in education boil down to one word: cheating, which none of us can tolerate.

Wait, but we have seen this before. The 1966 teacher protests against the use of calculators that are now being ridiculed all around social media. As a matter of fact, the teachers were right. They protested the use of calculators in elementary schools - where calculators are effectively banned even by now. In high schools, however, calculators got regulated. They are permitted in some tests but not in others. After all, the calculator has become an educational tool. As my high schooler used to say, “I spend more time on my TI-84-Plus than on my phone.”

So, will GenAI become an educational tool? Most certainly. A very powerful one. That can do all your homework. In all subjects. Without you even being there. But will it, really? 

First of all, GenAI isn’t great with numbers. In earlier GenAI versions, numbers were treated like words, so “7” and “8” were, to some extent, synonyms. Latest versions are dramatically better though - try GPT-o1 on math problems. Second, GenAI has been hallucinating and will always hallucinate, from time to time. When it doesn’t know something, it doesn’t know that it doesn’t know - so it will try to come up with something, often nonsense. Third, GenAI may be politically incorrect, which makes its use legally problematic in K-12. In higher education, however, students are adults and thus can sign the terms of service. Fourth, GenAI is so reactive: you ask - it answers. To be useful in education, GenAI must proactively guide the student through the knowledge acquisition process (to be fair, nothing stops the GenAI technology from being proactive).

Based on the above, it feels like today’s GenAI is a lousy educational tool. So, should we, ummm, just ignore it? Pretend it never happened? Or ban GenAI in education entirely, to stay on the safe side? We sure can, but what are we gonna do with teachers? They are already used to using GenAI to grade homework. For teachers, that’s not considered cheating, right? And what about students? Who’s gonna write their essays now?

If we ignore GenAI in education, there will be no education. Teachers will - hush hush - ask GenAI to come up with a syllabus for their new course. That syllabus will appear within seconds, better than anything they could come up with on their own. The complete course material will be ready in minutes, perfectly aligned with that perfect syllabus. The teacher will go to class and essentially act as a speaker system for automatically generated content. Homework will be created by the teacher’s GenAI, solved by the students’ GenAI, and graded by the teacher’s GenAI. Ditto midterms and final tests. On the surface, everything will look the same, the way it used to work for the past couple of centuries - but it will not be education.

But how has it actually worked so far?! Putting K-12 aside, let the government worry about it. Let’s focus on universities. From a business perspective, higher education is the only marketplace in the world, in which - paradoxically - neither the service provider (a Professor) nor the consumer (a Student) is interested in the service, which is paid for by a third party, typically a Parent). The astronomically priced transaction occurs between the University and the Parent, who purchases “good education” for their child. The University then employs the Professor to provide the “good education” to the Student who - for all practical purposes - is employed by the Parent. The “good education" stands between the Professor and their research lab, just as it stands between the Student and their social life. Neither of them wants to be in class.

Up until now, a considerable amount of work has been put by the Professor in creating and maintaining the class material, which is a part of their contract with the University. Meanwhile, the Student has worked hard to keep their grades high, fulfilling their “contract” with the Parent. Alas, not anymore. With GenAI, both contractual obligations will be met with minimal effort,  thereby resolving the paradox of the higher education marketplace. Welcome to transactional education, where the economic model finally makes sense!

Whoever suffers from transactional education is obviously the student who ends up learning nothing. Admittedly, some students do come to university to study. Not all of them end up graduating though, because studying is damn hard. So, can GenAI make studying less hard? Yes, it can.

Higher education is one-size-fits-all: one professor – many students. If a student misses something at the start of a lecture, they are lost until it’s over. GenAI, on the other hand, can be a truly personalized teacher, as Khan Academy is already showing. When GenAI teaches, students have a real chance not to get lost.

Students lose interest in studying when they can’t answer the big question why. Why am I spending these 20 minutes being taught this theory? How is this theory gonna help me in my career? Most students know why they study - they don’t know why they’re wasting their time now, at this very moment. And when they don’t know why, they will always find something better to do. Professors, for their part, are notoriously bad at explaining why they cover this specific material (maybe they aren’t so sure themselves?). That’s where GenAI comes to the rescue! It can always answer the question why. It has the potential to keep students focused.

But if that’s the case, why do students need a university at all? Because taking an online class alone in your bedroom is depressing. Humans are social animals, and GenAI doesn’t help here. The word “university” comes from Latin for “community” - being a part of the community is what young people look for in higher education.  Hence, the future of higher education lies in combining the opportunities offered by GenAI with the communal spirit of a brick-and-mortar school. In short, using GenAI in the physical classroom.

Lecture-style education is dead. Lectures were unbearably boring years ago, when we attended college. They are criminally boring now that an alternative exists. Some schools have already been practicing alternative education approaches. For example, West Point has been using the Thayer Method, in which “taking boards” (solving problems in class, on a blackboard, in groups of two or three students) plays the main role. Recently, the Thayer method was piloted at Georgia Gwinnett College, with considerable success. A newly founded, A+ ranked Minerva University is a brick-and-mortar school that teaches exclusively via online seminars, capped at 19 students. But having GenAI as a main teacher in a physical classroom is still unheard of. Let’s see if this can actually work.

Imagine a small classroom with 10–20 desks arranged in a circle (possibly two concentric circles?), each desk facing outward. Every desk is equipped with a computer running GenAI, so that all monitors face inward. Each student has their own desk, an office chair that can swivel, and a noise-canceling headset. One desk is replaced by a small whiteboard, with the class instructor seated next to it on an office chair.

Each class is split into short, alternating sessions of independent study and discussion. The instructor is in charge of the class syllabus but not entirely of the progress pace. At the start of the class, the instructor presents a topic, which students then study independently at their own speed using GenAI and wearing headsets if needed. If a student encounters difficulty, the instructor is there to help. After each independent study session, everyone turns around, and the instructor facilitates a discussion in which students explain what they have learned. The goal of the discussion is to sync up collective knowledge acquisition that may have taken different paths. Active student participation is mandatory and assessed. Both independent study and discussion can involve practical problem-solving. Working in groups of two or three is encouraged but not required. All discussions are recorded, and all independent study sessions are logged.

On the instructor side, the workload is fairly minimal: follow the syllabus, mediate discussions, and always know why each specific topic is being studied. In fact, the instructor might not even be a professor but rather a graduate student or a junior domain specialist. During the discussion, the instructor’s task is to catch GenAI hallucinations and correct them right away. After each class, GenAI can summarize all recordings and track progress. Based on these summaries, it can grade each student’s participation and propose (personalized?) practical problems for the next class. Homework doesn’t seem necessary unless the course workload can’t fit in class. The final test would likely be an oral one (a short one-on-one conversation between the instructor and each student). This might be time-consuming and exhausting for the instructor, but it spares them the preparation for written tests - and makes student cheating virtually impossible.

On the students’ side, they are in the driver’s seat. Each student acquires knowledge actively, which should keep them focused. If a student drifts off during an independent study session, GenAI may alert the instructor (besides, all monitors are visible from the instructor’s spot), and the instructor can steer the student back on topic. At all times, students should know why they are studying a specific topic. If they don’t, the instructor always has the answer. Students move at their own pace: some might cover only a little material, while others might go faster. If a student isn’t sure how to proceed, the instructor should offer a suggestion. During independent study sessions, students are free to use any type of educational material that suits their learning style: reading online textbooks, watching videos, conversing with GenAI etc. They may work alone or in small groups. In discussions, they learn from each other’s experiences and share their own, which helps solidify their knowledge.

There are many reasons why this model might fail. For instance, if there’s a wide deviation in students’ intellectual abilities, it could fail. If the instructor isn’t properly trained, it could fail, too. Obviously, both professors and students might object. Many courses have run the same way for years - why would a professor want to make a U-turn now? Would a student prefer to work extra hard in class, or keep drifting in the back row? Ultimately, GenAI-driven transactional education is easy and seems to work for everyone except the parents, who deplete their savings for their children’s “good education.” So we - parents - should be watching very, very closely.

First published in the CXO Tech Magazine

Questions? ron.bekkerman@njit.com