I think there is a very serious flaw with your argument, which starts off with a confusion between information and knowledge and extends to the most crucial but absent point: that the goal of education should be to teach students TO THINK.
If you already know how to think, AI can be a great tool for extending your knowledge, if used with care.
But college students in survey courses don't know how to think and analyze information — and they will never learn that from AI (or, for that matter, from bad professors). That is the value of getting a real education.
In my experience, learning how to think happens in several ways. Two of these include analyzing difficult questions in a group context (a classroom, which includes real discussion) and creating thoughtful arguments and presenting them in writing, otherwise known as papers or essays.
Students will never learn to think just by reading things on their own without any outside intellectual friction and dialogue, and they will never learn to present a thoughtful argument if "their" writing is outsourced to AI.
If we follow your suggestions, students will be robbed of the ability to genuinely learn and think. They will also be robbed of the chance of learning how to write.
You seem to be saying:
"Given the chance, many students will avoid the real work needed to learn how to think, engage in critical inquiry, and to write — and so what? In the world of AI, those skills are now passé, so we should just let them slide through the educational system, which is nothing more than a processing exercise anyways."
Agree. There's a layer of judgment and base knowledge at play when one initiates the context in a conversation with a LLM. It's very easy to end up on knowledge local minima without those 2 skills, and judging requires enough base knowledge to assess the reliability of statements, even more when those statements reach a certain level of complexity
Are you so sure that your students are learning to think? If you are, how are you? Are you quite convinced that you have something to do with how someone "thinks"? Can thinking really be taught? Or is it that case that people come equipped with ability to think or they do not?
Do you know the difference between information and knowledge? What is it?
Bracing as usual, but it reminds me of a psychology professor I had as an undergrad. Her lectures just repeated the textbook, but she was a great speaker and never once did I think to myself, I could have just read this myself and taken a test. She was the human expert in front of me, the one I impressed by visiting office hours and talking about relevant insights from sociology, the one who held me accountable on tests and who would have written me a recommendation had I needed one. The general accessibility of general knowledge didn't begin with GPT. We can ask "what do we need professors for?" but to some extent we could have asked that 30 years ago and I'm not sure the old answers are entirely out of date. I don't mean to imply that AI isn't enormously transformative, and Hollis is a bold and lucid thinker about that, but in thinking about its implications we shouldn't talk as if, pre-GPT, all knowledge was a carefully guarded secret whispered to students behind closed doors after they took an oath of secrecy.
Terrific post, but I think it's worth considering two advantages *some* professors have over AI that you don't mention.
First, a great teacher can get students interested in material they would have otherwise overlooked.
Second, actual back-and-forth with a formidable human being is a different experience than grilling an obliging LLM (as valuable as that is), and it's extremely valuable preparation for everything in life that comes after college.
Sadly for legislators and administrators, neither of those can be industrialized, so they aren't reliably scalable
The syllubus is a source of truth, and AI in this case (internet sourced LLM) is a soure of content, factual, fabricated, opinion. The biggest value of a syllibus perhaps is the discipline of drafting and publishing it. This might be akin to a software developer comment interleaved in source code.
Good post,with a GREAT image. All of this is spot on target.I do wonder, though, about the feasibility of last mile education in anything beyond small groups. Last mile sounds like a realization of the Oxbridge tutorial system-which combines motivated,highly qualified students,top notch faculty (both professors and tutors/fellows),and infinite $$ resources.But does it scale to public education? How? Should it?
Its a coincidence that I was reminded of this post when I published a post earlier this week called 'the last meter economy'. We appear to be working the same insight from different angles.
Your message here: AI delivers the general knowledge; the only defensible faculty role is working at the edge — specific, local, curated. The relationship between professor and student is what survives.
My key point: AI handles the scalable work; new jobs appear where automation stalls — physical presence, trust, judgment, exception-handling. The irreducibly human persists at the edges.
Same insight, different domains. The last mile or meter is where value consolidates when the middle gets eaten.
If last mile knowledge like the smell of Murat's bath is important, I wonder if domain experts would be better served by posting it online. Then, within a few years, that knowledge would be accessible to AIs and by proxy to all students using AI.
How would it serve them better? I think you mean, wouldn't this just go online and eventually turn up in AI results? I think Hollis would say, yes, that would eventually happen, but new knowledge is constantly being produced and what we can offer students (or part of what we can offer them) is a front-row seat to that.
I think there is a very serious flaw with your argument, which starts off with a confusion between information and knowledge and extends to the most crucial but absent point: that the goal of education should be to teach students TO THINK.
If you already know how to think, AI can be a great tool for extending your knowledge, if used with care.
But college students in survey courses don't know how to think and analyze information — and they will never learn that from AI (or, for that matter, from bad professors). That is the value of getting a real education.
In my experience, learning how to think happens in several ways. Two of these include analyzing difficult questions in a group context (a classroom, which includes real discussion) and creating thoughtful arguments and presenting them in writing, otherwise known as papers or essays.
Students will never learn to think just by reading things on their own without any outside intellectual friction and dialogue, and they will never learn to present a thoughtful argument if "their" writing is outsourced to AI.
If we follow your suggestions, students will be robbed of the ability to genuinely learn and think. They will also be robbed of the chance of learning how to write.
You seem to be saying:
"Given the chance, many students will avoid the real work needed to learn how to think, engage in critical inquiry, and to write — and so what? In the world of AI, those skills are now passé, so we should just let them slide through the educational system, which is nothing more than a processing exercise anyways."
Agree. There's a layer of judgment and base knowledge at play when one initiates the context in a conversation with a LLM. It's very easy to end up on knowledge local minima without those 2 skills, and judging requires enough base knowledge to assess the reliability of statements, even more when those statements reach a certain level of complexity
Are you so sure that your students are learning to think? If you are, how are you? Are you quite convinced that you have something to do with how someone "thinks"? Can thinking really be taught? Or is it that case that people come equipped with ability to think or they do not?
Do you know the difference between information and knowledge? What is it?
Bracing as usual, but it reminds me of a psychology professor I had as an undergrad. Her lectures just repeated the textbook, but she was a great speaker and never once did I think to myself, I could have just read this myself and taken a test. She was the human expert in front of me, the one I impressed by visiting office hours and talking about relevant insights from sociology, the one who held me accountable on tests and who would have written me a recommendation had I needed one. The general accessibility of general knowledge didn't begin with GPT. We can ask "what do we need professors for?" but to some extent we could have asked that 30 years ago and I'm not sure the old answers are entirely out of date. I don't mean to imply that AI isn't enormously transformative, and Hollis is a bold and lucid thinker about that, but in thinking about its implications we shouldn't talk as if, pre-GPT, all knowledge was a carefully guarded secret whispered to students behind closed doors after they took an oath of secrecy.
Terrific post, but I think it's worth considering two advantages *some* professors have over AI that you don't mention.
First, a great teacher can get students interested in material they would have otherwise overlooked.
Second, actual back-and-forth with a formidable human being is a different experience than grilling an obliging LLM (as valuable as that is), and it's extremely valuable preparation for everything in life that comes after college.
Sadly for legislators and administrators, neither of those can be industrialized, so they aren't reliably scalable
The syllubus is a source of truth, and AI in this case (internet sourced LLM) is a soure of content, factual, fabricated, opinion. The biggest value of a syllibus perhaps is the discipline of drafting and publishing it. This might be akin to a software developer comment interleaved in source code.
I think this is right.
Good post,with a GREAT image. All of this is spot on target.I do wonder, though, about the feasibility of last mile education in anything beyond small groups. Last mile sounds like a realization of the Oxbridge tutorial system-which combines motivated,highly qualified students,top notch faculty (both professors and tutors/fellows),and infinite $$ resources.But does it scale to public education? How? Should it?
Beautiful post.*
*Based on what ChatGPT told me you wrote.**
**Just kidding. I read the whole thing.
Its a coincidence that I was reminded of this post when I published a post earlier this week called 'the last meter economy'. We appear to be working the same insight from different angles.
Your message here: AI delivers the general knowledge; the only defensible faculty role is working at the edge — specific, local, curated. The relationship between professor and student is what survives.
My key point: AI handles the scalable work; new jobs appear where automation stalls — physical presence, trust, judgment, exception-handling. The irreducibly human persists at the edges.
Same insight, different domains. The last mile or meter is where value consolidates when the middle gets eaten.
If you're curious: rajeshachanta.substack.com/p/the-last-meter-economy
So, so good. And a great reminder that norms follow values. You can’t change the trajectory of this with edicts and standards.
smell of Marat's bath.. reminded me of the Robin Williams speech: https://www.youtube.com/watch?v=8GY3sO47YYo
Amazing post, thank you!
If last mile knowledge like the smell of Murat's bath is important, I wonder if domain experts would be better served by posting it online. Then, within a few years, that knowledge would be accessible to AIs and by proxy to all students using AI.
How would it serve them better? I think you mean, wouldn't this just go online and eventually turn up in AI results? I think Hollis would say, yes, that would eventually happen, but new knowledge is constantly being produced and what we can offer students (or part of what we can offer them) is a front-row seat to that.