Who Wins in the Battle Between Content and the Classroom?
How AI-Generated Video Will Reshape the Future of Education
“You wasted $381,500 on an education you coulda got for $0.00 and a few annoying Grammarly ads.”1
TL;DR
AI will make it easier than ever to generate high-quality educational content at little or no cost. If content is all you offer, it will be hard to compete. Universities have long survived by bundling content with credentials, community, and access to opportunity. That model still holds value, but each part of the bundle is facing new pressure. The challenge now is to rethink what makes the educational experience worth paying for and to invest in the parts that truly matter for learning and growth.
The AI Video Revolution
One of the most fascinating capabilities of artificial intelligence, which is advancing rapidly, is its ability to generate images and, more recently, videos. I've seen many impressive videos created by AI, and there's been a clear leap in the ability to produce videos from prompts over the past two years.
I recall being impressed a few years ago by a homemade video titled Harry Potter by Balenciaga. The person who made it clearly had creative skills, imagination, and, it appears, a decent set of technical skills to put this together.
However, as with all technological advances, this already impressive effort probably represented the most limited AI video we would ever see.
Today, models produce much higher-quality video and audio. More importantly, however, they are also far more accessible to the average person. We now have an entire TV channel featuring very creative (and quite impressive) AI videos generated with Google Veo.
I have no doubt that AI video generation technology will have a significant impact on many industries.
In this post, I will explore how AI-powered video generation could impact a sector I care deeply about: education.
Why Education Deserves Special Attention
Right now, I see a lot of discussion about AI video generation focusing on raw model power. There are entire genres of YouTube posts, LinkedIn demonstrations, and Substack posts that compare the subtle virtues of different video models (VEO3 vs. SORA vs....).
While I believe that raw model power is an essential factor in understanding how this technology will affect organizations and the economy (and society), I think that reasoning about the impact of AI starting from “disembodied capabilities” is much less valuable than thinking through how a technology will fit into the organizational and economic systems of specific industries. (In case you are interested in a deeper dive into how to think about organizational design in the context of AI, check out my working paper with Sampsa Samila and Alex Oettl [substack post here and SSRN Link here])
Without the specific details of sector-specific analyses, it is difficult to grasp, even in general terms, the impact this technology will have.
While many industries are likely to be impacted by AI's video generation capabilities, education—encompassing primary, secondary, and higher education—may face particularly severe consequences.
The Rise of Video as the New Lecture
By now, most students and teachers are accustomed to videos being a key part of educational content. Whether assigned by a high school math teacher to review concepts like L'Hôpital's Rule or used in MBA programs and college courses where professors record lessons to explain accounting, business strategy, or finance concepts, video has become a vital element of the learning experience.
Video serves as both a complement and a substitute for traditional in-person lectures or explanations, which were typically delivered live. Video-based educational content has already had a significant impact on education, and artificial videos are expected to have a similar effect.
Easily accessible educational video content has been available in various forms for over twenty years.
What caused it to spread rapidly was the ability to stream videos worldwide at a low cost and with ease. Streaming allowed anyone to distribute videos, resulting in a surge of content from all parts of the world and from anyone.
Over time, the ability to distribute content led to the endogenous emergence of innovations that further lowered the cost of creation. Specialized tools like Loom, as well as editing software such as Canva, allowed people to produce higher-quality videos. For those earning likes, clicks, and ad revenue, investing in better tools and production processes became worthwhile because the economic opportunity was present due to the shift in “Distribution” from 0 to 1.
Nevertheless, creating high-quality video content still required a lot of effort. You have to write a script, record it, edit it, and add visual features. While some tasks, such as adding subtitles, have become easier over time, they still require effort and skill.
The video revolution brought about by AI, combined with the ease of distribution through platforms like YouTube, in particular, means that content creation, especially video content creation, is now near costless (AI can produce content such as scripts and stories, as well as video).
Today, you don't even need a recording device, high-end equipment, or any editing skills to produce videos that once required someone with deep skill and expertise.
Making a B-School Explainer with VEO and HeyGen
I wanted to see if we could do something more “everyday” with AI video generation: create a believable video that explains an academic concept in a way that would be helpful to a student.
To do this, I signed up for a free trial of Google AI so I could test VEO2 and VEO3.
My goal was to have the AI create an explainer video for Porter's Five Forces, a basic business strategy concept I discussed in a different post.
VEO2
I actually started with VEO2, not realizing it didn't have sound. I'll paste it just for fun, because it's both funny and interesting (text is still a problem in these videos).
Here’s my prompt: “Generate a video of a professor (Indian) on a blackboard explaining Porter's five forces to students in an elite MBA program.”
VEO3
I then used this same prompt in VEO3.
Frankly, it isn’t that impressive.
I think what I'm asking this AI to do is quite complex and multidimensional. The overall video elements are impressive, and the voice is very good.
However, the clip is extremely short (I realize you are asking the model to do a lot and so have to string together small clips into a larger sequence, but this seems like a hassle to me).
There's also a significant misalignment between what the person is saying and what’s on the board. Additionally, what’s on the board is misspelled and incorrect.
This does not mean that AI-generated videos for education are doomed.
Educational content delivery tends to be a complex effort (you need the content, script, Blackboard writing, speaking, and all of these elements must align to give a high-quality experience to the person watching).
Even with its impressive capabilities, VEO3, as a monolithic general model, is challenged by the superadditivity of delivering educational content.
Moving to Modular Production
ChatGPT + HeyGen
So, I decided to try creating a video that either eliminated or modularized some of the components required to produce an educational video, moving from superadditive production to a more linear approach.
I found a cool technology called HeyGen, which I've subscribed to, and it appears to be much simpler than VEO. It essentially features a series of avatars where the AI only moves a few body parts (such as the mouth and hands), while the rest of the avatar remains still. All you need to do is provide a script, which you can create yourself or get from another source.
What I did here is simple. I used a prompt in ChatGPT 4o to generate a script explaining Porter's Five Forces. Then, I pasted it into HeyGen, and it created the video after I selected the persona and voice. Here’s the prompt that generates the script.
Write a 5-minute script explaining the business concept of Porter's Five Forces. The video should explain the concept to MBA students, provide examples from recent news, and also discuss the limitations of the concept. Please include only the part of the script that needs to be read and nothing else.
If anything, this is a significantly lower-tech version of the video than VEO3. However, splitting "script + video” into two dedicated steps, using ChatGPT for writing and HeyGen for rendering, allows each model to perform at its best.
The video is watchable and informative.
Interestingly, an integrated system like VEO3, which attempts to handle both tasks simultaneously, appears to struggle in matching the quality of a specialized disaggregated approach with substantially weaker visual rendering.
ChatGPT + HeyGen
Here’s a version where I adjusted a few parameters in ChatGPT and selected a different avatar in HeyGen, resulting in a pretty decent Hindi language video.
Write a 5-minute script explaining the business concept of Porter's Five Forces. The video should explain the concept to MBA students, provide examples from recent news, and also discuss the limitations of the concept. Please include only the part of the script that needs to be read nothing else. The script should be in Hindi and the examples need to be localized to the language's context.
Again, not bad. Very watchable.
Just-in-Time Learning?The New Frontier of Educational Content
My quick question to myself is whether I can use some of these videos, with some refinements, in a class.
I believe the answer is yes.
The ChatGPT + HeyGen videos are actually quite good and could be improved further through iteration on the prompt, the video avatar, and perhaps incorporating slides in the background (which I believe HeyGen allows).
All of this suggests that the marginal cost of producing a video has decreased even further. Since the cost of distribution is already very low, and now the cost of creation is also declining, prices will decline rapidly as competition inevitably intensifies.
Moreover, this process can be automated into code that generates an entire course programmatically using AI. Here’s a condensed pseudocode of what this might look like.
concepts = OpenAI_API.generate("List key n-gram concepts for [Intro Business Strategy]")
FOR concept IN concepts:
script = OpenAI_API.generate("5-min script explaining [" + concept + "] to MBA students")
FILE.save("scripts/" + concept + ".txt", script)
video_url = HeyGen_API.create_video(script, avatar_settings)
youtube_link = YouTube_API.upload(video_url, OpenAI_API.generate("Title for [" + concept + "]"), OpenAI_API.generate("Description for [" + concept + "]"))
save_to_site(concept, youtube_link)
launch_website()But a much more exciting possibility is the ability to produce just-in-time explainer videos. For example, you could say to the app, “explain Institutional Theory to me,” and receive a high-quality video within minutes (and as compute and models improve, in fractions of seconds). [business idea: a platform that hosts educational videos generated by students that can be shared with anyone.]
You can also imagine that as computing continues to improve and become more affordable, these videos will become interactive, allowing you to engage in a conversation with the avatar.
In a way, we already have this with the current chatbots, just with text and audio. For instance, on longer solo drives, I sometimes “converse” with ChatGPT to learn about a new topic (my daughters have renamed my number on their iPad “Nerd Alert”).
As I consider this, a key challenge will be optimizing the backend to prevent the repeated creation of redundant videos and to highlight the best version among many, particularly when the content is largely identical. This won’t be all that different from what happened with streaming and internet content distribution (e.g., with the founding of companies like Akamai, which made internet content distribution even more efficient).
The MOOC Moment Revisited
What are the implications for the education industry from the substantially decreasing cost of high-quality video content generation and distribution?
When I was an assistant professor at Stanford University, the school organized a faculty session where a group of faculty from the engineering school discussed massive open online courses, or whatever they were called at the time. There was a large-scale online course on machine learning at that time, offered at Stanford, that went gangbusters, and several faculty members were starting to take these courses and spin them out into separate companies (e.g., Coursera, Udacity, etc.).
At the time, I was focused on figuring out how to teach my class of 35 students, so I didn't fully understand the bigger implications of what was happening.
But these faculty members were working to make education accessible worldwide, nearly instant, and available at a fraction of the cost of attending a place like Stanford. It was an ambitious and inspiring mission, and MIT had already demonstrated demand for it through its OpenCourseWare project.
A few months later, we had a faculty meeting where our dean asked how institutions like ours should (or should not) respond to online education. I remember that meeting very clearly. At the time, the prevailing sentiment was that online education did not align with our mission. Still, the door was left open with a wait-and-see attitude.
Around the same time, a nonprofit called Khan Academy was also making waves in the primary and secondary education worlds. Its approach was somewhat different from the emerging massive open online courses, as its content was hosted on YouTube, rather than a walled-off platform. Khan Academy adopted a straightforward method: Salman Khan, the instructor, recorded lessons using a digital tablet and posted them online. Anyone worldwide with internet access could watch them for free.
The promise of educational technology, especially video-based content available both freely or at a low cost, anywhere, was inspiring.
Does More Educational Video = More Learning?
So, what’s the impact of this rapid growth in online educational content?
My initial thought is to examine the descriptives. This acts as a kind of gut check. (With the obvious caveat that correlation does not equal causation)
Have student educational outcomes improved over time due to the availability of free and easily accessible educational videos?
The descriptives don't seem promising.
Analyzing PISA scores reveals that, despite a substantial rise in online educational content at all levels, primary, secondary, and tertiary, performance on standardized tests remains weak. Scores are declining in nearly all OECD countries.
Furthermore, scores were already falling even before the COVID pandemic occurred.
EdTech might not have caused the decline, but it also doesn’t seem to have helped.
A substantial body of high-quality evidence also examines the impact of educational technology across various contexts. The results here are also pretty mixed.
Some of the strongest evidence of positive impact comes from targeted, personalized interventions. For example, Banerjee et al. (2007) and Muralidharan et al. (2019) show that adaptive, computer-based tutoring systems in India produced significant short-term improvements in student learning, particularly for those who struggled the most. Math scores increased by nearly half a standard deviation in some cases, which is pretty large for an effect size! However, these effects faded: Banerjee et al. found that a year after the intervention, the effect sizes had dropped by nearly 80%.
However, when examining broader, large-scale technology rollouts, such as providing laptops to every child or expanding broadband access in schools, the evidence becomes much less promising.
Studies from Peru (Cristia et al., 2017), Uruguay (Yanguas, 2020), and the U.S. (Goolsbee and Guryan, 2006) found no significant improvements in test scores, even though students clearly used the technology more. These programs often lacked meaningful instructional content or support for teachers, which seems to have limited their effectiveness. Another study by Belo et al. (2013) found that broadband access was actually detrimental to educational achievement, mainly because students spent time watching YouTube videos during school hours.
In some cases, technology may have actually worsened situations. A notable study in Romania by Malamud and Pop-Eleches (2011) found that giving low-income families vouchers to buy home computers led to lower grades for students, likely because the computers were used more for games than for homework.
So, my main takeaway from the research appears to be that EdTech is neither inherently good nor bad. Its value depends entirely on how it is used, how it is implemented, and the instructional model supporting it. There are both success stories and many, many cautionary tales.2
Free Content, Expensive Degrees
Beyond the impact of educational technology on student learning, which already has a vast and complex body of literature that many are still trying to reconcile, there is a less-studied area, I believe, that warrants exploration. This relates to the impact of the explosion of cheap educational content on the broader educational industry.
In attempting to understand the impact of zero-priced educational videos on the educational industry, I’m trying to reconcile several key facts.
The first point is that a vast amount of educational content is now available online, much of it free of charge.
Simply visit YouTube and search for any subject you’re interested in. You can find a video on it.
The second fact is that MOOCs, especially those that require payment or subscription, never really took off as some people had expected. Many of these platforms have undergone multiple CEO changes, and despite numerous attempts, they’ve struggled to establish a sustainable business model. Some, like Byjus, turned out to be frauds.
Take Coursera, for example. When it went public, it was valued significantly higher than it is now. Since then, it has lost approximately 80 percent of its value.
The same trend applies to lower-cost competitors like Udemy, which has also decreased by about 75 percent of its value since going public.
All you need is standard economic logic to predict these outcomes.
When there's a very close competitor, such as free online videos, it's challenging to persuade people to pay for your product.
Coursera, in particular, was selling a bundle that included the video content and a certificate. Assuming the value was the maximum of the two components, and given that the content was free on YouTube, the Coursera credential wasn’t worth much, which is why the stock dropped.
Why has In-Person Education Survived?
In-Person Enrollment has Remained Flat or is Increasing
The interesting puzzle is this: despite all these challenges, if you examine enrollment figures, especially in degrees that people are not required to pursue and must pay for, such as college or graduate school, the numbers have remained fairly steady. In some cases, they have even gone up.
For instance, based on data from the National Center for Educational Statistics, enrollment in four-year and two-year colleges has at worst remained flat.
In the case of post-baccalaureate degrees, there has been a steady upward trend over time, particularly among females.
However, to understand these trends, you need to examine them more closely to see what is really happening beneath the surface.
What is clear is that although the average sticker price of college has increased over time, the net price, which is the cost after discounts, financial aid, and other forms of support, has actually decreased. This is despite the fact that the cost of running a college has increased dramatically (e.g., services, scholarships, staff, etc.).
The Classroom is Not What it Used to Be
The third point is undeniable if you've recently taught in a classroom. The learning environment has “evolved.”
Students frequently fail to complete their assignments and often lack focus during lessons. Many are physically present but mentally distracted (some never show up, even once).
These issues are not isolated to large, non-selective schools; similar trends are observed across various institutions. [here, here, here, here, …]
The classroom experience appears to have undergone significant changes for many individuals, including both students and instructors. I’ve read at least a dozen op-eds from professors across the country, from various institutions, all saying the same thing: students do not participate, do not complete the readings, and do not engage in class.
I believe there are many reasons for this.
Here's one of mine: Why pay attention in class when most of the material is easily accessible outside of class for free? The class then becomes a hurdle, something to get past so you can earn your degree. The goal is to do just enough to pass. Grades are often inflated, so the strategy is to meet only the minimum requirements, get the credential, and move on to the job market (but why would anyone want to hire you?).3
At the end of each semester, I often receive multiple emails from students who, to be honest, seem completely unaware of their own disengagement. They request participation credit despite never attending class. A few years ago, I received an email from a student who emailed me to tell me how much they enjoyed listening to my “classes while [they] cook, clean, and handle other basic chores around the house."
(Research suggests that there may be a negative effect of lecture capture on attendance for some students.)
The Higher Ed Bundle Under Pressure
So, what exactly did free content do?
I think it's helpful to consider this from the perspective of a basic theory of substitutes.
Coursera, Udemy, and Udacity (which was eventually bought by Accenture) were mostly platforms for content. A credential from Udemy or Udacity doesn't hold much weight. The same is true for Coursera. Employers still prefer to hire candidates with degrees from traditional universities.
Free content available on platforms like YouTube directly competes with paid online content. When a close substitute is available for free, both the demand for and the price of the paid version typically decrease.
That is essentially what happened here.
The case for traditional education is much more subtle. Traditional universities weren’t selling content like the online vendors. They sold a package: You also receive a credential with labor market value, as well as support services, faculty, peer groups, and resources provided by the institution.
However, a significant portion of the content is now freely accessible. You can find similar lectures, explanations, and problem sets online with little effort.
This changes the value proposition for in-person educational institutions.
Many students no longer needed to attend classes primarily for the content (and recorded lectures made freely available by the professor don’t help either!).
If they are attending, it is for something else. In the simplest case, they are there only to get the credential. And even the value of that credential is becoming less certain, so students are increasingly offered scholarships and other incentives to enroll.
For many students, especially international ones, the value of a U.S. college degree is also less about the education itself and more about what it opens up: access to the U.S. job market.
Who Pays, and for What? Rethinking the Value Proposition of Education
I believe free online content must now be taken as a given, and it will become increasingly plentiful and notably improved thanks to AI.
In many instances, this content surpasses what most professors can produce independently. The videos I made earlier are significantly better than what I could have created myself.
Pandora’s box is now open.
We won’t be able to eliminate the free content students can access, and competing solely on content seems like a fool’s errand. Just look at what free content did to Coursera and Udemy. Brutal.
Moreover, lectures are an experience good, so why would someone pay hundreds of thousands of dollars upfront for content quality they can't easily assess beforehand, especially with so much free content available online to watch at home in your pajamas?
So what happens next?
To me, the main issue is that universities will need to figure out what they can bundle with their content to justify and maintain current prices. With the rise in scholarships and in-kind spending on services, actual “margins” in higher education are shrinking.
Whether we can sustain current price levels depends on the value of the other components in the bundle. Are they worth what students are paying?
What I am about to say does not directly concern artificial intelligence, but rather whether the surge of AI-generated content in education can be managed effectively, considering the various other components in the bundle and the potential threats to the value of these other elements from policy shocks or competitors.
The first major factor is the big elephant in the room: the growing uncertainty about access to American higher education for foreign students. Public policy is shifting in ways that may make it harder for students to come or to feel confident they will be able to stay and work after graduation. If students cannot come, are afraid to come, or lose access to the U.S. labor market after arriving, then the value of the bundle they are buying decreases.
Even for students already in the United States, the value of the credential itself may decline. Some states, such as Pennsylvania, have already begun to eliminate college degree requirements for many public sector jobs. And there are a host of new startups that are working with companies to provide skill-based credentialing. If these skill-based assessments or alternative credentials gain wider acceptance, there will be further downward pressure on the prices students are willing to pay for traditional degrees.
This naturally raises the question: what additional elements can be bundled with content that people are genuinely willing to pay for?
At an abstract level, there are two options on the table: (a) take the bundle as a given and reduce prices or (b) figure out how to increase willingness-to-pay by including higher-value items in the bundle.
I’m pretty sure that educational leaders do not want (a); so we’ll have to figure out (b).
This is already a very long post, so I will leave it for another day to think through what we can bundle with our content to survive the competition from AI.
Ultimately, I remain romantic about education. Knowledge brings economic, social, and personal power.
Watching a video, especially one generated by AI, is not “learning” or “education.”
The real challenge for us will be to inspire students to push themselves harder than they have ever done, to engage fully with ideas, and to aspire not just to get jobs, but to do great things.
How do we deliver on this promise?
Original Quote: "You wasted $150,000 on an education you coulda got for $1.50 in late fees at the public library." ― Matt Damon, Good Will Hunting.
Thanks for reading the footnote. My unfiltered take is that EdTech operates like any other business, mainly selling these technologies to parents and school systems without addressing the core issue: that studying, learning, and excelling in school are inherently challenging. Regardless of how advanced the technology is, without a supportive human system that encourages, incentivizes, and structures learning, no amount of ed-tech will solve the problem. Additionally, EdTech is similar to other educational solutions we’ve been sold for years, which have also been harmful to learning, such as the way reading is taught in primary schools. I recently listened to the Sold a Story podcast with my kids, including one who had a similar experience to many of the children featured, and I highly recommend it for any parent.
And there's another aspect to consider. Although content is often free and easily accessible, learning depends on more than just the material. Educational success fundamentally results from the interaction between content and student effort. Even the most excellent materials won't lead to learning if students do not actively engage. This prompts another question: in a world filled with abundant, high-quality, free content, how can we effectively support and motivate students in that process?