How much of the value of a university education is in meeting interesting people to think and be less lonely with? Not dealing with this aspect is part of the problem that made MOOC less successful than envisioned.
I'd like to say it is very important, but students' attendance at classes is often limited nowadays, with courses often being put online after the lecture. There could also be purely social institutions arising to fill the gap (e.g. networking clubs for AI-trained students).
Two thoughts: At the bachelor level +85% of students attend classes at my (fairly large) university. That's not particularly unusual, I think. The students are also not particularly interested in online lessons - they know they wouldn't really watch them.
That seems very unusual. At many universities here in AU attendance rates are way lower. But I do think that the Grok University model, as intriguing as it is, misses out on various peer effects that we know are highly relevant. That might well be crucial for its viability.
I work at Aarhus University so that message was really confusing, at first :). To be fair, the 85% is not a university wide number. But not an unusual number at the bachelor degree. And I reckon very few put their courses online after the lecture. Don't know of any at my faculty.
Incidental to your thesis, it seems to me that this would open an opportunity for colleges that would otherwise be shuttered to change their business models to that of providing the college experience. This would mean dorms and recreation and clubs. The classrooms would be repurposed or closed. The education would all be Grok.
Consider a public university. This would allow some of the infrastructure investment to be recouped. Taxpayers would like that.
This is an excellent piece. Given that I am seriously considering pivoting into academia after spending the past few years in an alt-ac position, the scenario you've painted is forcing me to grapple with things that could plausibly play out in a future 'Grok University'.
There are two things that I think are understated though:
1) Signalling expertise in fields like medicine, nursing and physiotherapy via a 'Grok University' certificate would not suffice. I think the same can be said about research in the 'hard sciences'. I have a family member who does graduate-level research in neuroscience, and I know for certain that many of the things that makes her a skilled neuroscientist was picked up tacitly through her lab's apprenticeship model.
2) A 'Grok University' costing $1000/year is still out of reach for many people in the Global South. I worry this model would exacerbate many of the inequalities that are currently playing out in the world today. Local universities serve as a pipeline to the local hospitals, research institutions, schools, offices, and other sectors of the local economy. A 'Grok University' credential might be benefit individuals in the Global South who are able to make it into the industries where this hypothetical Elon Musk holds a lot of influence, but at the end of the day, the rate of development in these localities would likely be negatively impacted
During my undergrad in my genetics class I remember working on the effect that Alcoholics Anonymous / church groups had on hyperdopaminergic and hypodopaminergic -induced addictions.
Literature shown these particular groups have virtues and rewarding effects that yield some form of dopaminergic levels regulation - diminishing the need for drug intakes, risk-seeking, etc.
I wouldn't discount the ability of group environments to alleviate the difficulty of what takes usually great cognitive efforts. Like learning.
This is a great article, and I think your analysis is accurate. Thank you.
One possible outcome is that we end up with a two-tier system, where elite schools still provide a boutique hands-on catered experience while less prestigious institutions go all-in on AI-based training for credentialling for practical career paths (e.g. business, finance, accounting, etc.).
The two barriers you name are the whole thing, and credentialing is the harder one because it is a belief problem and not a technology problem. A degree is worth something because employers think students work hard for it, and students work hard because employers think that. A better tutor does not touch that loop anywhere. The physical testing centers are the only part of the scenario that actually attacks it, which is why that detail stayed with me.
Fantastic article. I think it's actually very conservative in its assumptions.
*Of course* an AI tutor which adapts to you, never loses its cool and uses all the tricks in the book to encourage learning will be vastly more effective than a distant lecturer in the amphitheater's pit. Have you ever had ChatGPT explain anything to you? (yes, AI will kill us all, but *Damn* it's good)
Musk is very much the guy who could pull it off. In addition to his resources and ability to prime the pump on credentialization by showcasing SpaceX / Tesla / Musk, Inc. employment prospects, he can leverage his unique and global brand (how many followers on X? Couple hundred *millions*?)
I think it will happen, and at AI-diffusion speed. Some universities will hang on, mostly on the strength of affluent parents' nostalgia ("I want you to have a college experience like I did"). The rest will revert to being sports franchises.
The real threshold for AI education is not knowledge, but two capabilities that are much harder to build.
The first is an assessment system.
This is not assessment in the sense of grading an exam, but continuously building an understanding of a specific student. Getting a question wrong is only an outcome. The system has to determine how that outcome was produced. Only when the assessment is sufficiently accurate can personalization have a reliable basis.
A student may get the same question wrong because they did not understand the concept, because their reasoning went down the wrong path, because they lack prerequisite knowledge, or simply because of carelessness or a calculation error. They may even have calculated the correct answer but written it down incorrectly. The outcome is the same, but the student’s actual state is completely different, and what needs to be taught next is therefore completely different.
This requires AI to continuously build and update a model of the student: what they know, what they do not know, and why; their personality, interests, subject preferences, learning pace, and which teaching methods work better for them.
The second is an active planning system.
This is even harder. Assessment only answers, “Where is the student now?” Planning has to answer: “Where should we take them, and what should the next step be?”
This requires AI not merely to provide answers in response to questions, but to have a goal to reach and a clear direction, and then actively plan a path based on the student’s current state: what to learn now, what to learn next, how much to increase the difficulty, which method to use, where to stop and fill in missing knowledge, and where to move directly forward.
And this path cannot remain fixed once it has been planned. Every time the student completes a learning task, new information is produced. The system has to reassess the student’s state, determine whether the original plan still applies, and then adjust the next step.
So genuinely personalized education requires more than an AI with vast knowledge that can answer questions. It requires a continuously operating system:
Assess the state → build a model → determine the goal → plan the path → execute → obtain feedback → reassess → adjust the path.
Knowledge is only a resource that this system can draw on. The real difficulty is whether it can accurately understand a specific person and, around a goal, continuously organize subsequent actions in a directed way.
Interesting scenario build out. The relevance of signaling and the use of credentials to demonstrate meta qualities (as opposed to using the knowledge gained) seems archaic too.
Credentials in a technical field (eg say forensics) can be directly relevant to the job. Not so much in other cases. Perhaps, AI could also help save time and money by figuring out a better way for employers to assess skills.
One problem for universities is that the credibility of their credentialisation has decreased over time (grade inflation, organised cheating). A standardised examination could be a very compelling alternative in some fields.
Believing AI universities will replace Traditional high ranking universities (like the Ivys or Tsinghua/Seoul) by providing higher quality education at cheap prices is like believing people go to Ivys to get educated or that people compete to get in because they value higher learning. It's like believing people value Louis Vuitton/Prada for their textile quality or Lamborghini/Ferrari for their aerodynamic impressive engineering.
Believing AI universities will replace Local mid-ranked universities is like believing an old grandma will stop going to church once she finds out Sunday sermons are streamed online with high quality video and audio; or that teen girls buy apple phones out of appreciation of its specs
Believing AI universities will replace low ranking colleges and universities for struggling students is like believing that cronic homelessness can be solve with cheap and abundant housing and nothing more.
“AI will revolutionize Education !!!!” will join the ranks of “This is the year of the Linux desktop !!!! ” as yet another example of tech guys not understanding people. It will require a massive amount of social proof and even then ….
“Dad, Mom, so I know I studied hard to get into UPenn and I got the grades for it and all that but I think I would rather go for grok university. I could learn at my own pace right ? 😅” == “Honey, what the point of a diamond ring and that fancy dress, you know I love you right? 😙 ”
I'd say that's a bit pessimistic. The market for something like Grok University initially could probably be people unable to afford living near their local mid-ranked universities or UATX types opposed to wokeness on a principle. That's a non-negligible share of the population.
Another alternative to consider is that non-elite universities fade into irrelevancy as AI turns entry-level white collar jobs enormously scarce. "Study whatever and land an email job" path already is hardly viable, and technical fields software slowly start skewing senior. After all, the mere concept of professors being replaceable by AI means that training for this particular profession is a potentially uncertain path.
This was a great read! I think there are two major hurdles this doesn't address though:
1. I think employers look at education as a proxy for intelligence and conscientiousness, which are reasonably good predictors of employee quality. The tests for individual capability with the material would probably cover intelligence to a good degree, but especially if individualized learning makes continuing easy and rewarding, it may not be a good proxy for conscientiousness. This would likely mean students would need a stronger external resume also, which could be off-putting.
2. Traditional education is one of the few ways to significantly increase IQ (good meta-analysis is Ritchie 2018). Its not clear something like this would offer the same benefit. As a result, education increases quality of life, longevity, etc. This would need to be able to match that, and AIs don't have a great reputation around changes in cognitive ability. I think it could probably eventually be tuned for this, but it would take time and might lower completion rates by making learning more challenging.
Employer uptake outside the founding company’s network would be a useful signpost for this scenario. A guaranteed interview can get the first cohort through the door, but independent employers repeatedly hiring those graduates would tell us much more about whether the credential has acquired lasting value.
I suspect AI will destroy education, mostly because education has destroyed itself, with elite institutions tending towards becoming hedge funds with meal cards.
But it is a bad thing, not a neutral one.
For one, it ignores the research that learning is situational. Lave and Wenger, communities of practice, the apprenticeship literature, all point to learning as embodied, occurring in situ, and requiring interaction. It is certainly possible to learn the rules of statistics from a book, but without interaction, time and application in a social milieu it is meaningfully degraded.
Moreover, the “credentialling” model will fail for the same reason No Child Left Behind has failed so miserably. Test-based outcomes simply turn into test-driven inputs.
This is equivalent to the modern production of produce, whose optimization for color, size, low blemishes, and resistance to damage during transport has deleted flavor and nutrients, notably the most valuable ones like omega-3 fatty acids and trace minerals. Omega-3 is especially tragic because it was deliberately bred out of modern seeds to improve shelf-stability.
So we will watch credentialing and curricular development will replicate the same credentialing and rating crisis we saw in the mortgage collapse, creating objectives that are cheap to reach and losing distinctions that are essential until it is too late.
In fact, it will be worse. During the loan crisis, at least the ratings agencies cosplayed at independence, here Musk will be the degraded product and quality control.
And Musk will chortle his way to the bank while we move to “nutrition-free” education.
I think you are right that this will happen, but it will be the ultimate enshittification of society itself.
Did you listen to Derek Thompson's recent podcast with dean of Harvard Business School? It implicitly speaks to your scenario, since the dean argued that future education needs to focus much more on the university as a community, focusing on teaching students skills and values that an online education cannot convey: For example how to work together in groups, how to present whatever an AI has built etc. I found the outlook promising, and would certainly tap into challenges that one has to obtain in real, physical communities.
How much of the value of a university education is in meeting interesting people to think and be less lonely with? Not dealing with this aspect is part of the problem that made MOOC less successful than envisioned.
I'd like to say it is very important, but students' attendance at classes is often limited nowadays, with courses often being put online after the lecture. There could also be purely social institutions arising to fill the gap (e.g. networking clubs for AI-trained students).
Two thoughts: At the bachelor level +85% of students attend classes at my (fairly large) university. That's not particularly unusual, I think. The students are also not particularly interested in online lessons - they know they wouldn't really watch them.
That seems very unusual. At many universities here in AU attendance rates are way lower. But I do think that the Grok University model, as intriguing as it is, misses out on various peer effects that we know are highly relevant. That might well be crucial for its viability.
I work at Aarhus University so that message was really confusing, at first :). To be fair, the 85% is not a university wide number. But not an unusual number at the bachelor degree. And I reckon very few put their courses online after the lecture. Don't know of any at my faculty.
Incidental to your thesis, it seems to me that this would open an opportunity for colleges that would otherwise be shuttered to change their business models to that of providing the college experience. This would mean dorms and recreation and clubs. The classrooms would be repurposed or closed. The education would all be Grok.
Consider a public university. This would allow some of the infrastructure investment to be recouped. Taxpayers would like that.
This is an excellent piece. Given that I am seriously considering pivoting into academia after spending the past few years in an alt-ac position, the scenario you've painted is forcing me to grapple with things that could plausibly play out in a future 'Grok University'.
There are two things that I think are understated though:
1) Signalling expertise in fields like medicine, nursing and physiotherapy via a 'Grok University' certificate would not suffice. I think the same can be said about research in the 'hard sciences'. I have a family member who does graduate-level research in neuroscience, and I know for certain that many of the things that makes her a skilled neuroscientist was picked up tacitly through her lab's apprenticeship model.
2) A 'Grok University' costing $1000/year is still out of reach for many people in the Global South. I worry this model would exacerbate many of the inequalities that are currently playing out in the world today. Local universities serve as a pipeline to the local hospitals, research institutions, schools, offices, and other sectors of the local economy. A 'Grok University' credential might be benefit individuals in the Global South who are able to make it into the industries where this hypothetical Elon Musk holds a lot of influence, but at the end of the day, the rate of development in these localities would likely be negatively impacted
During my undergrad in my genetics class I remember working on the effect that Alcoholics Anonymous / church groups had on hyperdopaminergic and hypodopaminergic -induced addictions.
Literature shown these particular groups have virtues and rewarding effects that yield some form of dopaminergic levels regulation - diminishing the need for drug intakes, risk-seeking, etc.
I wouldn't discount the ability of group environments to alleviate the difficulty of what takes usually great cognitive efforts. Like learning.
Great scenario nonetheless
Very interesting observation.
This is a great article, and I think your analysis is accurate. Thank you.
One possible outcome is that we end up with a two-tier system, where elite schools still provide a boutique hands-on catered experience while less prestigious institutions go all-in on AI-based training for credentialling for practical career paths (e.g. business, finance, accounting, etc.).
The two barriers you name are the whole thing, and credentialing is the harder one because it is a belief problem and not a technology problem. A degree is worth something because employers think students work hard for it, and students work hard because employers think that. A better tutor does not touch that loop anywhere. The physical testing centers are the only part of the scenario that actually attacks it, which is why that detail stayed with me.
Fantastic article. I think it's actually very conservative in its assumptions.
*Of course* an AI tutor which adapts to you, never loses its cool and uses all the tricks in the book to encourage learning will be vastly more effective than a distant lecturer in the amphitheater's pit. Have you ever had ChatGPT explain anything to you? (yes, AI will kill us all, but *Damn* it's good)
Musk is very much the guy who could pull it off. In addition to his resources and ability to prime the pump on credentialization by showcasing SpaceX / Tesla / Musk, Inc. employment prospects, he can leverage his unique and global brand (how many followers on X? Couple hundred *millions*?)
I think it will happen, and at AI-diffusion speed. Some universities will hang on, mostly on the strength of affluent parents' nostalgia ("I want you to have a college experience like I did"). The rest will revert to being sports franchises.
The real threshold for AI education is not knowledge, but two capabilities that are much harder to build.
The first is an assessment system.
This is not assessment in the sense of grading an exam, but continuously building an understanding of a specific student. Getting a question wrong is only an outcome. The system has to determine how that outcome was produced. Only when the assessment is sufficiently accurate can personalization have a reliable basis.
A student may get the same question wrong because they did not understand the concept, because their reasoning went down the wrong path, because they lack prerequisite knowledge, or simply because of carelessness or a calculation error. They may even have calculated the correct answer but written it down incorrectly. The outcome is the same, but the student’s actual state is completely different, and what needs to be taught next is therefore completely different.
This requires AI to continuously build and update a model of the student: what they know, what they do not know, and why; their personality, interests, subject preferences, learning pace, and which teaching methods work better for them.
The second is an active planning system.
This is even harder. Assessment only answers, “Where is the student now?” Planning has to answer: “Where should we take them, and what should the next step be?”
This requires AI not merely to provide answers in response to questions, but to have a goal to reach and a clear direction, and then actively plan a path based on the student’s current state: what to learn now, what to learn next, how much to increase the difficulty, which method to use, where to stop and fill in missing knowledge, and where to move directly forward.
And this path cannot remain fixed once it has been planned. Every time the student completes a learning task, new information is produced. The system has to reassess the student’s state, determine whether the original plan still applies, and then adjust the next step.
So genuinely personalized education requires more than an AI with vast knowledge that can answer questions. It requires a continuously operating system:
Assess the state → build a model → determine the goal → plan the path → execute → obtain feedback → reassess → adjust the path.
Knowledge is only a resource that this system can draw on. The real difficulty is whether it can accurately understand a specific person and, around a goal, continuously organize subsequent actions in a directed way.
Not sure if you saw this, very close to your story: https://www.nytimes.com/2026/08/22/business/dealbook/harvard-ai-faculty.html?smid=nytcore-ios-share&referringSource=articleShare
Yes, I had seen that. Seemed to me a somewhat superficial product to make more money on the side. I think the key will be credentialisation.
Interesting scenario build out. The relevance of signaling and the use of credentials to demonstrate meta qualities (as opposed to using the knowledge gained) seems archaic too.
Credentials in a technical field (eg say forensics) can be directly relevant to the job. Not so much in other cases. Perhaps, AI could also help save time and money by figuring out a better way for employers to assess skills.
One problem for universities is that the credibility of their credentialisation has decreased over time (grade inflation, organised cheating). A standardised examination could be a very compelling alternative in some fields.
Definitely.
Believing AI universities will replace Traditional high ranking universities (like the Ivys or Tsinghua/Seoul) by providing higher quality education at cheap prices is like believing people go to Ivys to get educated or that people compete to get in because they value higher learning. It's like believing people value Louis Vuitton/Prada for their textile quality or Lamborghini/Ferrari for their aerodynamic impressive engineering.
Believing AI universities will replace Local mid-ranked universities is like believing an old grandma will stop going to church once she finds out Sunday sermons are streamed online with high quality video and audio; or that teen girls buy apple phones out of appreciation of its specs
Believing AI universities will replace low ranking colleges and universities for struggling students is like believing that cronic homelessness can be solve with cheap and abundant housing and nothing more.
“AI will revolutionize Education !!!!” will join the ranks of “This is the year of the Linux desktop !!!! ” as yet another example of tech guys not understanding people. It will require a massive amount of social proof and even then ….
“Dad, Mom, so I know I studied hard to get into UPenn and I got the grades for it and all that but I think I would rather go for grok university. I could learn at my own pace right ? 😅” == “Honey, what the point of a diamond ring and that fancy dress, you know I love you right? 😙 ”
I'd say that's a bit pessimistic. The market for something like Grok University initially could probably be people unable to afford living near their local mid-ranked universities or UATX types opposed to wokeness on a principle. That's a non-negligible share of the population.
Another alternative to consider is that non-elite universities fade into irrelevancy as AI turns entry-level white collar jobs enormously scarce. "Study whatever and land an email job" path already is hardly viable, and technical fields software slowly start skewing senior. After all, the mere concept of professors being replaceable by AI means that training for this particular profession is a potentially uncertain path.
This was a great read! I think there are two major hurdles this doesn't address though:
1. I think employers look at education as a proxy for intelligence and conscientiousness, which are reasonably good predictors of employee quality. The tests for individual capability with the material would probably cover intelligence to a good degree, but especially if individualized learning makes continuing easy and rewarding, it may not be a good proxy for conscientiousness. This would likely mean students would need a stronger external resume also, which could be off-putting.
2. Traditional education is one of the few ways to significantly increase IQ (good meta-analysis is Ritchie 2018). Its not clear something like this would offer the same benefit. As a result, education increases quality of life, longevity, etc. This would need to be able to match that, and AIs don't have a great reputation around changes in cognitive ability. I think it could probably eventually be tuned for this, but it would take time and might lower completion rates by making learning more challenging.
Employer uptake outside the founding company’s network would be a useful signpost for this scenario. A guaranteed interview can get the first cohort through the door, but independent employers repeatedly hiring those graduates would tell us much more about whether the credential has acquired lasting value.
Wow, this is depressing.
I suspect AI will destroy education, mostly because education has destroyed itself, with elite institutions tending towards becoming hedge funds with meal cards.
But it is a bad thing, not a neutral one.
For one, it ignores the research that learning is situational. Lave and Wenger, communities of practice, the apprenticeship literature, all point to learning as embodied, occurring in situ, and requiring interaction. It is certainly possible to learn the rules of statistics from a book, but without interaction, time and application in a social milieu it is meaningfully degraded.
Moreover, the “credentialling” model will fail for the same reason No Child Left Behind has failed so miserably. Test-based outcomes simply turn into test-driven inputs.
This is equivalent to the modern production of produce, whose optimization for color, size, low blemishes, and resistance to damage during transport has deleted flavor and nutrients, notably the most valuable ones like omega-3 fatty acids and trace minerals. Omega-3 is especially tragic because it was deliberately bred out of modern seeds to improve shelf-stability.
So we will watch credentialing and curricular development will replicate the same credentialing and rating crisis we saw in the mortgage collapse, creating objectives that are cheap to reach and losing distinctions that are essential until it is too late.
In fact, it will be worse. During the loan crisis, at least the ratings agencies cosplayed at independence, here Musk will be the degraded product and quality control.
And Musk will chortle his way to the bank while we move to “nutrition-free” education.
I think you are right that this will happen, but it will be the ultimate enshittification of society itself.
Fascinating - enjoyed it.
Did you listen to Derek Thompson's recent podcast with dean of Harvard Business School? It implicitly speaks to your scenario, since the dean argued that future education needs to focus much more on the university as a community, focusing on teaching students skills and values that an online education cannot convey: For example how to work together in groups, how to present whatever an AI has built etc. I found the outlook promising, and would certainly tap into challenges that one has to obtain in real, physical communities.