I'm actually pretty positive about the impact AI is going to have on higher education in the long term. At least the actual education part. AI suddenly reverses the long-standing situation in which it was easier and more appealing to grade students based on effort more than understanding.
Unfortunately, a big part of education has come to be about evaluating the extent to which a student is willing to show up and put in the effort. Having taught in higher education I was constantly appalled at the extent to which many instructors seemed more focused on evaluating whether students put in the work than how much they understood.
AI suddenly makes it very hard to evaluate how much effort a student has put into a course. It is, however, still quite easy -- using in class exams/essays -- to evaluate their level of understanding.
I get that this is going to be a very difficult transition and it may threaten the role of the college degree as a general pass to higher pay (perhaps for the good) but I really hope it leads to a deeper examination of what we actually want education to do.
Schools need to accept that many of the techniques we have historically used for student evaluation simply no longer work. That doesn't mean people can't learn things anymore - they're just not going to learn material the way schools have been trying to teach it to them.
I have to assume that almost everyone reading this has LEARNED THINGS from using AI, even if for some of us it's the just kind of trivia you used to look up on Wikipedia - and for many it has been much more. It's easy imagine how a student having a guided and observed discussion about a given subject with a knowledgeable AI could be used as a real pedagogical tool. Students could get familiar with background knowledge and concepts to use for a really rigorous class discussion, maybe a better one than you would have gotten hoping everyone had done (and absorbed) the reading.
Sometimes I think we are overstating how much kids are learning "nothing." I assume teachers still employ tests with the "hey, come sit in this room and answer these questions on a piece of paper" method and I assume that kids are still passing them. It could be that the fake essays aren't evidence that kids aren't learning anything anymore, they show that the essays in particular weren't actually helping much anyway.
It's unfortunate, but even at college level I think a lot of students still struggle with the basic task of writing an essay to an extent that it can crowd out their actual intellectual engagement with the thing they are supposed to be learning about. The "how many pages does it have to be?" mindset works against the grain of learning new ideas. That may have gone from "problem we don't know how to fix" to "problem we will never fix." But because most people do not write essays in their lives outside of school, ever - not unlike dodgeball - we can figure out a way to make that work. An AI tutor could theoretically help them learn about how to READ an essay, which is a more important skill that about half of adults struggle with.
It's really a dilemma. I teach college and now students do all their writing in class, even essays are split over multiple days with multiple stages. I personally love AI, and I do learn from it. But I know what I am asking for (prompting) and I have asked Claude to be a sparring interlocutor -- and I never have it write for me. In that role, AI is invaluable in teaching me complicated ideas/papers, etc., that I work through in a dialogue with Claude.
But I know the students use AI as a shortcut, either by having it write or think for them. Most students at my school hate AI and know it's cheating. But if one, two, or three people use AI on papers, it ruins the class for me, hence my move to in-class writing.
So there is an important role for AI, but you have to be smart enough to know how to use it.
On the other hand, I hate how every populist, left and right, is against data centers, even smart professors at my school. These supposedly smart people have swallowed the anti-AI and anti-data center narrative without any critical reflection, without looking into sources and data that confounds their view. I think it's incredibly sloppy that physicists, chemists, and people in hard sciences simply accept Karen Hao's arguments and phoney data. I have shown them Andy Massley's work on the subject, but it reconfirms their confirmation bias. It's quite disappointing. But as I know from attending faculty meetings, the collective IQ of the faculty drops when we're in a large crowd with emotional reasoning running high.
I strongly agree with your last point. The public discussion of AI and data centers in particular has been a depressing demonstration of how more Americans than we realized form opinions and hold positions based on dubious information, and exhibit a strong cynicism-driven resistance to getting better information. Purely emotion-driven analysis and political action has mostly been a problem on the right this century, but we’re starting to see that that development was more of a matter of supply than demand.
There is also a failure in epistemological ethics: people who oppose AI for high-minded ideological reasons are accepting as allies people whose objections are both different from their own and empirically weak, and either happily riding their support to a larger coalition of majority opposition (like corporate Republicans exploiting pro-life voters), or convincing themselves the real motivation of the masses must actually match their own. Like nah, man, they might genuinely be mad about an imaginary water problem.
(Apologies for sentence length. I don’t have the energy to edit this right now. Hah. I think it scans.)
There's a more basic thing I've been thinking wrt AI. We need to be thinking in terms of many scenarios.
For example, some people are thinking in terms of AI and job loss. This is sensible, because one possible path forward is that AI causes large-scale job loss and we need to deal with it.
Some people are thinking in terms of AI risk because of giving bad actors more power to do bad things. This is sensible, it's a real thing that can happen and we should care about.
Some people are thinking in terms of deepfakes and loss of human spaces because of AI-written text, bots impersonating humans, etc. This is also something that can happen and is worth dealing with.
Some are thinking in terms of hostile AI posing a threat to human existence, or simply coming to dominate the Earth and sweep us aside when we're in the way. That's also sensible to think about.
And some people are thinking about narrower short-term stuff like how to make sure high school/college kids learn to read and write and do math and learn stuff, instead of just making their robot do their homework. That's also worthwhile.
As a parallel, consider the cold war. There must have been a lot of people looking at the invention of nuclear technology and thinking "my God, this is liable to be the end of us." And it was sensible to think about that at many levels, from environmental and waste disposal regulations for nuclear materials, to how to build and regulate nuclear power plants, to nonproliferation (not letting everyone have their own nukes), to MAD and great power conflicts, to missile defenses, to medical protocols for dealing with radiation exposure, and so on. If everyone just dropped everything and obsessed over nothing but the impending destruction of civilization in the Northern hemisphere in nuclear fire, a lot of important stuff for human flourishing wouldn't have happened, and maybe the world being more broken would have led us to a worse outcome wrt not having a nuclear exchange with the USSR.
Huckabee has reasons to be frustrated with Israeli behavior:
"Anti-Christian behavior has tested the support of Israel’s staunchest allies in the U.S. When Mike Huckabee took over as U.S. ambassador to Israel last year, he was shocked to discover that evangelical Christian organizations were struggling to obtain visas. A strong supporter of Israel, Huckabee ran a tour company that brought thousands of Christian pilgrims to Israel and the West Bank and never had problems.
Huckabee went straight to Israel’s minister of interior, but the problem persisted, so he turned to Prime Minister Benjamin Netanyahu, who finally helped release the visas.
The main obstacle, said Huckabee, appeared to be an official at the Ministry of Interior who held anti-Christian prejudices. The U.S. Embassy is still working to streamline the visa process to avoid future problems.
“The best friends that the state of Israel has are evangelical Christians from America,” Huckabee said in an interview. “And if their organizations could not get their people to come serve on a clergy visa, that was the most unfortunate message they could possibly send.”"
I will not compare them to their terrorist organization foes, but how much more enlightened a society are they these days than the Iraqis or Gulf Arabs, let alone the Egyptians?
I think the difference is that Israel still has a sizable minority who hold liberal values. But they can't win political power anymore - the liberal parties have consistently been held to 25%ish of the vote for a decade at this point. Iraq and Gulf Arabs have almost nobody there who holds such values.
It's a difference but in a majoritarian democracy it is almost like none exists.
I think we're very pleased with our choice! Pretty quick/easy read, which is nice for something we're both reading on the side. We read Piranesi for our last joint book, which was another breezy read (finished it in one sitting).
Using AI to write essays and assignments for you on topics you're supposed to have learned is like rolling up to a gym with a forklift and then bragging about how much you can lift.
My (perhaps old-fashioned) view is that there's no substitute for *knowing* things. If you're an immunology student and you're supposed to write about the latest state of the art advances in chimeric antigen receptors, and you ask ChatGPT to do it for you, what have you learned? And how will you know if ChatGPT is hallucinating? If I ask ChatGPT to tell me about chimeric antigen receptors, and it starts with "chimeric antigen receptors acquire antigen specificity via V(D)J recombination," I know enough to respond "ChatGPT, you're drunk, go home," but a first year immunology student won't.
This is why we have to switch to in-class, handwritten assignments for assessing student progress. "But writing an essay under time limits is haaaard!" Yes it is, which is why I would grade the *knowledge* component over polish and style and grammar/spelling. If you're in my immunology class, and I give you a piece of paper and tell you to write everything you know about chimeric antigen receptors, you may not be able to write a beautiful essay in half an hour, but you ought to be able to give me some bullet points and a rough sketch of what a chimeric antigen receptor looks like, what it's used for, what are the advantages and limitations of chimeric antigen receptors, how are researchers working to improve chimeric antigen receptors, etc. You either know this stuff or you don't.
The thing is, ChatGPT is a wonderful tool for learning things, finding references, etc. But also an easy tool to make do your homework, so that you can f--k off instead of studying.
But as George Carlin pointed out, if it’s the end of the world, your underwear won’t be clean for long. First you’re going to say it, then you’re going to do it.
Relevant to today's article, just saw that Bukele is touting a new AI Tutor program which lead to "Results in reading, mathematics and science were above the national average and comparable to the average results in Germany and Sweden."
The link was skim on the details so I hope they get released soon, but personalized teaching has long been the white whale of EdTech and now finally seems within reach. Could be pretty cool!
I'm skeptical that a barbarian whose solution to crime was to put everyone with a tattoo in prison has done anything remotely useful with language models.
I'm sure there may well be plenty of other legitimate critiques of Bukele, but when it comes to crime he was wildly successful executing a straightforward strategy. Virtually the entire region is attempting to imitate him in some form or another.
Weirdly, I think the crime rate was going down pretty rapidly before he came to power. Though surely locking up every gang member he could get his hands on made life much better for the people who weren't gang members or their close friends/relatives.
The murder rate had been declining, but violent deaths in El Salvador were more like violent deaths in Ukraine and Gaza than the phenomenon of violent deaths in a place like USA. In other words it had more to do with decisions made by gang leadership on how much to fight each other and incentivize violence than any change in social conditions.
It could be that less recordable kinds of crime, like extortion, gang-recruitment, kidnapping-for-ransom, etc... were also going down, but I've never really heard that. My sense is the gangs weren't slowly weakening much prior to Bukele's crackdown, even if the murder rate was far below the peak of 2015 or so.
I said I'm skeptical of his educational achievements; incarcerating a large proportion of your population to decrease crime isn't exactly a galaxy brained idea. I myself am in favor of incarcerating a larger proportion of the American population... but not in gulags.
The sooner Bukkake gets Maduro'd by the Ossoff administration the better; man's gotta pay for the SECOT nonsense (we're lucky no Americans got sent thereO.
Have you spent any time in El Salvador and actually visited their prisons and jails? Did you care about the gangs and how the people were treated before Bukele became a cause celebre of sorts? I ask because Salvadoran prison/jail facilities were and still are in many places terrible and disgusting. People crammed into cages where they poop in baggies and throw that into a barrel. There are some nasty places and CECOT is not too bad in comparison. At least they are fed. Many places in El Sal the family needs to bring food for prisoners to keep them healthy. Is that good -- no -- but the way to fix it is increase state capacity and the underlying economy.
Right now El Salvador has about 1.7% of its population in prison, compared to the U.S. at .7%. (Roughly). So 2.5X. That's not a "large proportion of the population" in my mind. It's definitely high, but in return, you have a vastly safer country that can try to build an economy, and you have thousands of people a year not being murdered, and thousands not trying to flee to the United States.
IF El Sal can eradicate the gang virus (which was turbo charged by U.S. deportees) and build capacity, hopefully it can prevent a resurgence once people are inevitably let out. And figuring out how to try and rehabilitate some of the people, and find the ones who are wrongly imprisoned, is a noble goal. But sometime for society to survive extreme measures are warranted. And I think Bukele got it right -- even if he originally wanted to make deals with the gangs.
Is "build more housing to lower prices" a galaxy-brained idea? Was Matt's "1 billion American's" a galaxy brained idea?
Good government is about effectively implementing simple ideas. If Bukele had randomly tripled the incarceration rate, or did it in some other less effective way it wouldn't have accomplished anything. But he did it in a way that was very effective at destroying the violent gangs running half his country.
I think it helped a lot that the gangs that were terrorizing everyone conveniently marked themselves with specific tattoos, and enforced the correctness of the labeling by murdering any non-members with those tattoos. I have to guess that future Salvadoran gangs will look for less unambiguous ways to identify themselves and will make this kind of crackdown harder.
It did help, but he basically won that war. Maybe some day there will be new street gangs in El Salvador, but they are unlikely to be anywhere near as terrifying, powerful or violent as the gangs Bukele defeated.
More to the point, it's a very different problem for which the useful tools and insight are probably unrelated to those needed to solve the "gangs have taken over the country" problem.
Did teachers make you want to learn in primary/secondary school? or was it fear of your parents, fear of falling behind your peers, wanting to have options as you grew up, etc
I don't see how that basic relationship changes.
I'm not familiar with this program but I imagine they're using standard teachers and AI tutors to augment that teaching. Having a tutor always on call as you struggle to understand something seems very useful to me.
The models are moving so fast and the technology is new enough that I don't think we can say that definitively. Especially in the US where knowledge workers are sharply anti AI.
If it's true that with AI tutors El Salvador achieved German levels of performance, that's encouraging right? And I doubt the World Bank and PISA are putting out puff pieces. (But like I said i want to see more formal results).
Seems kind of defeatist to assume that we already figured out education 100 years ago and there's no more improving it. I think we can what we can learn from the last 15-20 years in edtech is that introducing more screens is just more distractions actually hurts learning. But that doesn't mean there's not other solutions that are actually more effective
I think a common pattern is that the pilot program (done in well-resourced schools with dedicated handpicked teachers and often students with the problem kids filtered out) shows impressive results, but later attempts to apply the same techniques in normal schools with normal teachers and students don't.
For education, my understanding is that there is no technique that is remotely as powerful as selecting your students, so everyone kinda tries to do that whenever possible. Somehow get your students to be the children of wealthy accomplished parents, or children of parents who jumped through an elaborate and demanding set of hoops to get their kids into the program, or make the children jump through the hoops, or have a school reputation that convinces the badly behaved/unmotivated/dumb kids not even to apply, or just do the magnet school thing and give the kids (effectively) an IQ test and take only very smart kids. All that will give you big gains in test scores, graduation rates, AP classes, etc.
Other techniques may be useful, but they have like 5% of the impact of selecting your students, and they're much harder to implement.
My low-confidence impression is that this is one reason ed research seems like it usually doesn't yield much--there's this huge effect that confounds the tiny effects you actually want to measure, and that effect is hard to entirely avoid Worse, we only hear about the educational interventions that looked like they were really successful, but those are the ones that are most likely to have had some influence of the huge effect we don't want to measure (we already know a class full of kids with an average IQ of 120 will do well) that swamped the small effects of better educational techniques or slightly smaller class sizes or something.
I'm sort of torn on the issue of AI in K-12. Fundamentally I think addressing it at the level of law seems to be a bad idea.
In general I think unfettered access to a computer is a bad idea in grade school. I've fully been stripped of my techno-enthusiasm by the Chromebook experience. In my class my filter settings aren't you can go to any website except these it's you can only go to these handpicked websites. Plus physical proximity. There's like all kinds of ways that Ai can go bad.
But to say that there's never ever a productive use for AI seems ridiculous. As an autistic student with academic special interests I would have been all over brainstorming and research uses of AI. I admit this is an edge case but there's a lot of edge cases. I teach a gifted class who a small number of those students might benefit from a sprinkling of AI in their searches compared to google. It's not something I'd undertake using lightly. But it seems completely improbable there would be literally no use for some tasks that some children might do in an entire school system as big as New York or LA. I'd want edge cases to be able to be optimally reached as best as possible. At the extreme upper end I could see uses for it in a lot of enrichment projects. Gifted covers this incredible range from kids who are just kind of smart to a relatively profound group of people who honestly do need personalized attention that just won't ever scale to really reach their potential.
On the other other hand I've seen teachers who hand the kids Chromebooks and say go ahead and listen to music on you tube while you do your math problems and then are shocked when those math problems aren't done.
Bad cases make bad law. And a lot of the problems with the modern world is thinking that the existence of edge cases is a good way to set defaults.
Every single school district in the universe has IEPs that can cater for edge cases.
Just like IEPs allow all kinds of changes to testing that normally aren't allowed.
Someone formally diagnosed as autistic could get an IEP allowing use of AI. If, somehow, there was actual empirical evidence beyond just vibes, that was something beneficial to them.
Dunno why we'd skip past demanding empirical evidence of efficacy.
I don't think we should allow edge cases to set defaults. I think the default should be using AI with your students is a bad idea. I think law is a bad way to set norms.
I teach a class of exceptional students at the high end. Part of the tradeoff of that deal is finding ways to create opportunities for them to be challenged and I think careful small scale experiments make a lot of sense here without having to wait for permission from a bureaucracy. The whole reason they send their children to a charter is to avoid this kind of bureaucratic structure. I'm not sure how these laws apply to choice public schools and charters in New York and LA.
It reminds me a lot of the discussion around Khan academy. I'm pretty sure if Khan academy had been around when I was a kid and I'd had access to it, I'd have started learning calculus in the eighth grade or something. Because I was a weird kid who found that stuff really interesting. But it probably wouldn't have had nearly so big an impact on the median kid, who'd maybe look up something when they were confused in a math class or something, but not much more than that.
I agree it's something that's for unusual cases but it just so happens I teach a class of mostly gifted students with a small number of high achievers who are just shy of gifted. The median IQ in my class is like 140. I'm not even saying I have a ready made use for AI but it doesn't seem totally implausible that I would want to have a narrow targeted use.
I have developed some units I'm really proud of using internet resources and I think this is kind of insanely proscriptive. Hey I'm allowing you to go to these specific websites for this specific task really opens up opportunities for students to interact with materials.
Just an obvious for instance our state provides a list of whole books which they think are excellent for teaching many of the standards. They do not however provide the books and a google classroom with a PDF of the books. Or I have a unit about the development of American music and readings on Jazz and Blues followed by videos of examples of Louis Armstrong make it stick better than me talking about it ad infinitum.
There's a lot of useful things and the desire to exert uniform control seems pretty harmful especially to the gifted population I've been teaching the last two years.
There's an easy solution for that. Provide the books or provide a pdf of the books on the local database that doesn't have access to the broader internet.
Moreover, i'm a big fan of one standardized curriculum. Like that way, you can do standardized tests on not just reading in a mass. But on social studies, history, science
Etc
But kids cannot have unfettered access to the internet
And it's been proven time and time again. That whatever types of restrictions you put on them, the kids get around it
This goes along with my other idea that we are going to have to get rid of homework. And instead links in the school day so that all schoolwork is done, any proctored environment
I think I agree with you on standardized curriculum but that’s not where we’re at all. And for advanced learners we’re given even less. Figure something out as long as you get growth.
Google Classroom is a really good bulletin board system. I really doubt the district would create one nearly as capable.
I really think you’re underestimating the power of walking around the room. Wholly technical solutions do fail but the thing that doesn’t is mixing that with old fashioned class management tricks of walking around and paying attention to what kids are paying attention to. 3rd graders aren’t exactly subtle when they’re goofing off.
Nice article but I don’t think I agree with this -
“While a student working in an engineering lab at M.I.T. probably should use A.I. to expand the possibilities of their studies and contributions to the world”
Yes that might be fine after they have learnt the fundamentals but not before. I really doubt most students even at top ranked schools will be able to use it that judiciously.
AI-enhanced search is really good at finding research in unusual places - that's the case I could see for a grad student at MIT. If a researcher in Ghana posted something on his department homepage, AI is more likely to find it than any other tool I know.
TIL: "big chunks of LI (including malls!) are still on septic! Part of the reason is lack of a sewer system makes it easier to stop any new construction"
Suffolk's 2025 fair-housing analysis actually labels antiquated wastewater infrastructure an “impediment” to higher-density and affordable housing, because it makes that development either impossible or prohibitively expensive. It also says that “expanding access to sewers is a growth-inducing action” and that sewer construction creates the opportunity for increased development density.
East Hampton's official waterfront plan states that development should continue to be planned so as to avoid the future need for sewers and sewage treatment.
We don’t gotta hand it to anyone, but it genuinely is impressive the extent to which NIMBYs spent the last 50 years innovating how to stop development.
I sometimes wonder how this period is going to echo through history. I’m sure various civilizations have gone through NIMBY phases, but generally speaking the broad sweep of history has been relatively YIMBYtopian — people were usually always happy to expand and develop.
But this crisis was one that most of the West induced on itself for a very specific reason: the environmental backlash against industrial pollution and climate change. Its causality is thus probably pretty unique vs. most other NIMBY phases. And its sheer scope and scale is also pretty unique!
The correction we’re currently witnessing is just in its infancy. It’d be interesting to see what people think about the NIMBY mistake (if they even see it as one!) and how they view the correction to it in a century or two. Will it have been successful? Will it have required some larger, unexpected intervention that we haven’t seen yet? Will it take some sort of turn that we YIMBYs haven’t yet been able to conceive of?
I think about this all of the time. Having ceded so much authority, even veto power, to “the community,” how can even state-level government claw it back for the broader public interest? There are legal strategies that could be pursued but I’m not so sure about the political will that would be required.
Thanks for the excellent article, Halina. My wife teaches at an R1 public university and my daughter just started her first year at a small liberal arts college. In my experience, almost everyone at both types of institutions is hungry for answers to the questions you've raised.
I think the most likely outcome will emerge organically. It will be sloppy, confusing, and often contradictory. Institutions that don't take capabilities or changing incentives seriously are going to struggle, and probably disappear.
Agree, thanks for the great work. It underplayed what I suspect will be the load-bearing* issue: educational institutions are to a great extent *solutions* to the problem of lack of consensus among stakeholders. “The faculty want parking, the alumni want football, the students want sex,” etc. Legibility breaks the mechanism for achieving working consensus — it forces stakeholders to acknowledge tradeoffs among each other.
The structure of most academic institutions -- a weak executive, consensus-focused committees, significant autonomy for academic departments -- is poorly suited to deal with the rapidly changing world of AI.
That structure has some upside for individual professors and departments, who can't be blocked or hamstrung by their institution. But a large share of those individuals have to make good choices, or the institution as a whole will lose credibility or become irrelevant.
None of which is to say that I know exactly what a good answer looks like, though I can tell some bad ones a mile away. No one can afford to stay ignorant of these changes, so the ostrich strategy is institutional suicide.
"Last month, the Massachusetts Institute of Technology published its committee report on “A.I. use in teaching, learning, and research training,” which is arguably the most exhaustive institutional self-examination on the subject to this point. M.I.T., a leader in both A.I. development and use, has produced a master key that, far beyond acting as a single elite institution’s one-size-fits-all prescription for all educators, is a thorough accounting of every value an educational institution could be organized around and the ways A.I.’s integration can advance or undermine the stated goal."
I'm actually pretty positive about the impact AI is going to have on higher education in the long term. At least the actual education part. AI suddenly reverses the long-standing situation in which it was easier and more appealing to grade students based on effort more than understanding.
Unfortunately, a big part of education has come to be about evaluating the extent to which a student is willing to show up and put in the effort. Having taught in higher education I was constantly appalled at the extent to which many instructors seemed more focused on evaluating whether students put in the work than how much they understood.
AI suddenly makes it very hard to evaluate how much effort a student has put into a course. It is, however, still quite easy -- using in class exams/essays -- to evaluate their level of understanding.
I get that this is going to be a very difficult transition and it may threaten the role of the college degree as a general pass to higher pay (perhaps for the good) but I really hope it leads to a deeper examination of what we actually want education to do.
So college football is apparently college futbol, where the refs have a special watch that nobody else can see.
Schools need to accept that many of the techniques we have historically used for student evaluation simply no longer work. That doesn't mean people can't learn things anymore - they're just not going to learn material the way schools have been trying to teach it to them.
I have to assume that almost everyone reading this has LEARNED THINGS from using AI, even if for some of us it's the just kind of trivia you used to look up on Wikipedia - and for many it has been much more. It's easy imagine how a student having a guided and observed discussion about a given subject with a knowledgeable AI could be used as a real pedagogical tool. Students could get familiar with background knowledge and concepts to use for a really rigorous class discussion, maybe a better one than you would have gotten hoping everyone had done (and absorbed) the reading.
Sometimes I think we are overstating how much kids are learning "nothing." I assume teachers still employ tests with the "hey, come sit in this room and answer these questions on a piece of paper" method and I assume that kids are still passing them. It could be that the fake essays aren't evidence that kids aren't learning anything anymore, they show that the essays in particular weren't actually helping much anyway.
It's unfortunate, but even at college level I think a lot of students still struggle with the basic task of writing an essay to an extent that it can crowd out their actual intellectual engagement with the thing they are supposed to be learning about. The "how many pages does it have to be?" mindset works against the grain of learning new ideas. That may have gone from "problem we don't know how to fix" to "problem we will never fix." But because most people do not write essays in their lives outside of school, ever - not unlike dodgeball - we can figure out a way to make that work. An AI tutor could theoretically help them learn about how to READ an essay, which is a more important skill that about half of adults struggle with.
It's really a dilemma. I teach college and now students do all their writing in class, even essays are split over multiple days with multiple stages. I personally love AI, and I do learn from it. But I know what I am asking for (prompting) and I have asked Claude to be a sparring interlocutor -- and I never have it write for me. In that role, AI is invaluable in teaching me complicated ideas/papers, etc., that I work through in a dialogue with Claude.
But I know the students use AI as a shortcut, either by having it write or think for them. Most students at my school hate AI and know it's cheating. But if one, two, or three people use AI on papers, it ruins the class for me, hence my move to in-class writing.
So there is an important role for AI, but you have to be smart enough to know how to use it.
On the other hand, I hate how every populist, left and right, is against data centers, even smart professors at my school. These supposedly smart people have swallowed the anti-AI and anti-data center narrative without any critical reflection, without looking into sources and data that confounds their view. I think it's incredibly sloppy that physicists, chemists, and people in hard sciences simply accept Karen Hao's arguments and phoney data. I have shown them Andy Massley's work on the subject, but it reconfirms their confirmation bias. It's quite disappointing. But as I know from attending faculty meetings, the collective IQ of the faculty drops when we're in a large crowd with emotional reasoning running high.
I strongly agree with your last point. The public discussion of AI and data centers in particular has been a depressing demonstration of how more Americans than we realized form opinions and hold positions based on dubious information, and exhibit a strong cynicism-driven resistance to getting better information. Purely emotion-driven analysis and political action has mostly been a problem on the right this century, but we’re starting to see that that development was more of a matter of supply than demand.
There is also a failure in epistemological ethics: people who oppose AI for high-minded ideological reasons are accepting as allies people whose objections are both different from their own and empirically weak, and either happily riding their support to a larger coalition of majority opposition (like corporate Republicans exploiting pro-life voters), or convincing themselves the real motivation of the masses must actually match their own. Like nah, man, they might genuinely be mad about an imaginary water problem.
(Apologies for sentence length. I don’t have the energy to edit this right now. Hah. I think it scans.)
There's a more basic thing I've been thinking wrt AI. We need to be thinking in terms of many scenarios.
For example, some people are thinking in terms of AI and job loss. This is sensible, because one possible path forward is that AI causes large-scale job loss and we need to deal with it.
Some people are thinking in terms of AI risk because of giving bad actors more power to do bad things. This is sensible, it's a real thing that can happen and we should care about.
Some people are thinking in terms of deepfakes and loss of human spaces because of AI-written text, bots impersonating humans, etc. This is also something that can happen and is worth dealing with.
Some are thinking in terms of hostile AI posing a threat to human existence, or simply coming to dominate the Earth and sweep us aside when we're in the way. That's also sensible to think about.
And some people are thinking about narrower short-term stuff like how to make sure high school/college kids learn to read and write and do math and learn stuff, instead of just making their robot do their homework. That's also worthwhile.
As a parallel, consider the cold war. There must have been a lot of people looking at the invention of nuclear technology and thinking "my God, this is liable to be the end of us." And it was sensible to think about that at many levels, from environmental and waste disposal regulations for nuclear materials, to how to build and regulate nuclear power plants, to nonproliferation (not letting everyone have their own nukes), to MAD and great power conflicts, to missile defenses, to medical protocols for dealing with radiation exposure, and so on. If everyone just dropped everything and obsessed over nothing but the impending destruction of civilization in the Northern hemisphere in nuclear fire, a lot of important stuff for human flourishing wouldn't have happened, and maybe the world being more broken would have led us to a worse outcome wrt not having a nuclear exchange with the USSR.
"and those that are designed to prepare students for tangible jobs in an evolving workforce"
You cant think critically without deep knowledge on a topic.
You cant write well without years of practice.
Without many years of productive struggle including in college you will be worthless
How truly terrible do things have to be getting for Mike Huckabee to be claiming someone is committing terrorism AGAINST Muslims.
“Huckabee Condemns ‘Terrorist’ Attacks on Palestinians in West Bank Town Hall”
https://www.nytimes.com/2026/09/05/world/middleeast/huckabee-west-bank-settler-attacks.html
Huckabee has reasons to be frustrated with Israeli behavior:
"Anti-Christian behavior has tested the support of Israel’s staunchest allies in the U.S. When Mike Huckabee took over as U.S. ambassador to Israel last year, he was shocked to discover that evangelical Christian organizations were struggling to obtain visas. A strong supporter of Israel, Huckabee ran a tour company that brought thousands of Christian pilgrims to Israel and the West Bank and never had problems.
Huckabee went straight to Israel’s minister of interior, but the problem persisted, so he turned to Prime Minister Benjamin Netanyahu, who finally helped release the visas.
The main obstacle, said Huckabee, appeared to be an official at the Ministry of Interior who held anti-Christian prejudices. The U.S. Embassy is still working to streamline the visa process to avoid future problems.
“The best friends that the state of Israel has are evangelical Christians from America,” Huckabee said in an interview. “And if their organizations could not get their people to come serve on a clergy visa, that was the most unfortunate message they could possibly send.”"
https://www.wsj.com/world/middle-east/spitting-attacks-show-rise-of-anti-christian-hate-in-israel-5ea40ac7
The more I hear about it the more that sounds like some sub-Saharan African country. "Western Democracy" my hind foot.
I will not compare them to their terrorist organization foes, but how much more enlightened a society are they these days than the Iraqis or Gulf Arabs, let alone the Egyptians?
What is the equivalent to the occupation of the West Bank (to say nothing of Gaza) being conducted by Iraq or the Gulf Arabs or Egypt?
I believe the Gulf States provide at least some assistance to the Saudis in Yemen.
I think the difference is that Israel still has a sizable minority who hold liberal values. But they can't win political power anymore - the liberal parties have consistently been held to 25%ish of the vote for a decade at this point. Iraq and Gulf Arabs have almost nobody there who holds such values.
It's a difference but in a majoritarian democracy it is almost like none exists.
My wife and I are reading Circe for our personal book club! Lanterns rules too.
What do you think so far?
I think we're very pleased with our choice! Pretty quick/easy read, which is nice for something we're both reading on the side. We read Piranesi for our last joint book, which was another breezy read (finished it in one sitting).
Using AI to write essays and assignments for you on topics you're supposed to have learned is like rolling up to a gym with a forklift and then bragging about how much you can lift.
My (perhaps old-fashioned) view is that there's no substitute for *knowing* things. If you're an immunology student and you're supposed to write about the latest state of the art advances in chimeric antigen receptors, and you ask ChatGPT to do it for you, what have you learned? And how will you know if ChatGPT is hallucinating? If I ask ChatGPT to tell me about chimeric antigen receptors, and it starts with "chimeric antigen receptors acquire antigen specificity via V(D)J recombination," I know enough to respond "ChatGPT, you're drunk, go home," but a first year immunology student won't.
This is why we have to switch to in-class, handwritten assignments for assessing student progress. "But writing an essay under time limits is haaaard!" Yes it is, which is why I would grade the *knowledge* component over polish and style and grammar/spelling. If you're in my immunology class, and I give you a piece of paper and tell you to write everything you know about chimeric antigen receptors, you may not be able to write a beautiful essay in half an hour, but you ought to be able to give me some bullet points and a rough sketch of what a chimeric antigen receptor looks like, what it's used for, what are the advantages and limitations of chimeric antigen receptors, how are researchers working to improve chimeric antigen receptors, etc. You either know this stuff or you don't.
The thing is, ChatGPT is a wonderful tool for learning things, finding references, etc. But also an easy tool to make do your homework, so that you can f--k off instead of studying.
Strong agreement.
I can tell you when I interview employees. I always ask a number of in person accounting questions
If you can't answer them on the spot, you don't get the job
And i've passed on a number of people, many of them. Recent graduates with degrees in the subject
So they just spent 4 or 5 years, thousands of dollars and wasted all of it
11/10, no, 17/10
Sharty, dear, I know you hate AI, but I just fixed my washing machine by uploading some pictures and asking for directions.
Analogistically (typo?), opening up your car door to fix it on YouTube is awesome.
It may be the end of the world, but my clothes will be clean.
I mean, if it's the end of the world, surely you want to face it in clean underwear!
Isn't that a line (or close to a line) from a Bloom County strip?
But as George Carlin pointed out, if it’s the end of the world, your underwear won’t be clean for long. First you’re going to say it, then you’re going to do it.
Though the good news is that probably there won't be anyone around to laugh at you....
Relevant to today's article, just saw that Bukele is touting a new AI Tutor program which lead to "Results in reading, mathematics and science were above the national average and comparable to the average results in Germany and Sweden."
https://x.com/nayibbukele/status/2096255497410118075?s=20
The link was skim on the details so I hope they get released soon, but personalized teaching has long been the white whale of EdTech and now finally seems within reach. Could be pretty cool!
I'm skeptical that a barbarian whose solution to crime was to put everyone with a tattoo in prison has done anything remotely useful with language models.
If it was such a dumb and barbaric strategy why couldn't more civilized governments figure it out or do something more effective?
How do I know they haven't? People like him tend to be bent as a spoon. He's accountable to no one.
I'm sure there may well be plenty of other legitimate critiques of Bukele, but when it comes to crime he was wildly successful executing a straightforward strategy. Virtually the entire region is attempting to imitate him in some form or another.
Weirdly, I think the crime rate was going down pretty rapidly before he came to power. Though surely locking up every gang member he could get his hands on made life much better for the people who weren't gang members or their close friends/relatives.
The murder rate had been declining, but violent deaths in El Salvador were more like violent deaths in Ukraine and Gaza than the phenomenon of violent deaths in a place like USA. In other words it had more to do with decisions made by gang leadership on how much to fight each other and incentivize violence than any change in social conditions.
It could be that less recordable kinds of crime, like extortion, gang-recruitment, kidnapping-for-ransom, etc... were also going down, but I've never really heard that. My sense is the gangs weren't slowly weakening much prior to Bukele's crackdown, even if the murder rate was far below the peak of 2015 or so.
I said I'm skeptical of his educational achievements; incarcerating a large proportion of your population to decrease crime isn't exactly a galaxy brained idea. I myself am in favor of incarcerating a larger proportion of the American population... but not in gulags.
The sooner Bukkake gets Maduro'd by the Ossoff administration the better; man's gotta pay for the SECOT nonsense (we're lucky no Americans got sent thereO.
Have you spent any time in El Salvador and actually visited their prisons and jails? Did you care about the gangs and how the people were treated before Bukele became a cause celebre of sorts? I ask because Salvadoran prison/jail facilities were and still are in many places terrible and disgusting. People crammed into cages where they poop in baggies and throw that into a barrel. There are some nasty places and CECOT is not too bad in comparison. At least they are fed. Many places in El Sal the family needs to bring food for prisoners to keep them healthy. Is that good -- no -- but the way to fix it is increase state capacity and the underlying economy.
Right now El Salvador has about 1.7% of its population in prison, compared to the U.S. at .7%. (Roughly). So 2.5X. That's not a "large proportion of the population" in my mind. It's definitely high, but in return, you have a vastly safer country that can try to build an economy, and you have thousands of people a year not being murdered, and thousands not trying to flee to the United States.
IF El Sal can eradicate the gang virus (which was turbo charged by U.S. deportees) and build capacity, hopefully it can prevent a resurgence once people are inevitably let out. And figuring out how to try and rehabilitate some of the people, and find the ones who are wrongly imprisoned, is a noble goal. But sometime for society to survive extreme measures are warranted. And I think Bukele got it right -- even if he originally wanted to make deals with the gangs.
Is "build more housing to lower prices" a galaxy-brained idea? Was Matt's "1 billion American's" a galaxy brained idea?
Good government is about effectively implementing simple ideas. If Bukele had randomly tripled the incarceration rate, or did it in some other less effective way it wouldn't have accomplished anything. But he did it in a way that was very effective at destroying the violent gangs running half his country.
I think it helped a lot that the gangs that were terrorizing everyone conveniently marked themselves with specific tattoos, and enforced the correctness of the labeling by murdering any non-members with those tattoos. I have to guess that future Salvadoran gangs will look for less unambiguous ways to identify themselves and will make this kind of crackdown harder.
It did help, but he basically won that war. Maybe some day there will be new street gangs in El Salvador, but they are unlikely to be anywhere near as terrifying, powerful or violent as the gangs Bukele defeated.
It's fascist AF but it did solve crime, tbf.
It did; education is a tougher nut.
More to the point, it's a very different problem for which the useful tools and insight are probably unrelated to those needed to solve the "gangs have taken over the country" problem.
Unlikely! AI wont make people want to learn.
Did teachers make you want to learn in primary/secondary school? or was it fear of your parents, fear of falling behind your peers, wanting to have options as you grew up, etc
I don't see how that basic relationship changes.
I'm not familiar with this program but I imagine they're using standard teachers and AI tutors to augment that teaching. Having a tutor always on call as you struggle to understand something seems very useful to me.
I read the encyclopedia for fun in elementary school. I still read Wikipedia for fun today.
I mean, both. Some things were fascinating and I ate them up, and some things were boring af and I learned them to get a good report card.
If it worked everyone would be doing it. I find it unlikely that El Salvador has figured out some magic combination of teacher and AI tutor.
The models are moving so fast and the technology is new enough that I don't think we can say that definitively. Especially in the US where knowledge workers are sharply anti AI.
If it's true that with AI tutors El Salvador achieved German levels of performance, that's encouraging right? And I doubt the World Bank and PISA are putting out puff pieces. (But like I said i want to see more formal results).
I'm not all the way on Freddy's end but I do think the default assumption about any education breakthrough should be "it's 99% bull"
Seems kind of defeatist to assume that we already figured out education 100 years ago and there's no more improving it. I think we can what we can learn from the last 15-20 years in edtech is that introducing more screens is just more distractions actually hurts learning. But that doesn't mean there's not other solutions that are actually more effective
I think a common pattern is that the pilot program (done in well-resourced schools with dedicated handpicked teachers and often students with the problem kids filtered out) shows impressive results, but later attempts to apply the same techniques in normal schools with normal teachers and students don't.
For education, my understanding is that there is no technique that is remotely as powerful as selecting your students, so everyone kinda tries to do that whenever possible. Somehow get your students to be the children of wealthy accomplished parents, or children of parents who jumped through an elaborate and demanding set of hoops to get their kids into the program, or make the children jump through the hoops, or have a school reputation that convinces the badly behaved/unmotivated/dumb kids not even to apply, or just do the magnet school thing and give the kids (effectively) an IQ test and take only very smart kids. All that will give you big gains in test scores, graduation rates, AP classes, etc.
Other techniques may be useful, but they have like 5% of the impact of selecting your students, and they're much harder to implement.
My low-confidence impression is that this is one reason ed research seems like it usually doesn't yield much--there's this huge effect that confounds the tiny effects you actually want to measure, and that effect is hard to entirely avoid Worse, we only hear about the educational interventions that looked like they were really successful, but those are the ones that are most likely to have had some influence of the huge effect we don't want to measure (we already know a class full of kids with an average IQ of 120 will do well) that swamped the small effects of better educational techniques or slightly smaller class sizes or something.
I loved Circe!
You should also watch Lanterns.
I'm sort of torn on the issue of AI in K-12. Fundamentally I think addressing it at the level of law seems to be a bad idea.
In general I think unfettered access to a computer is a bad idea in grade school. I've fully been stripped of my techno-enthusiasm by the Chromebook experience. In my class my filter settings aren't you can go to any website except these it's you can only go to these handpicked websites. Plus physical proximity. There's like all kinds of ways that Ai can go bad.
But to say that there's never ever a productive use for AI seems ridiculous. As an autistic student with academic special interests I would have been all over brainstorming and research uses of AI. I admit this is an edge case but there's a lot of edge cases. I teach a gifted class who a small number of those students might benefit from a sprinkling of AI in their searches compared to google. It's not something I'd undertake using lightly. But it seems completely improbable there would be literally no use for some tasks that some children might do in an entire school system as big as New York or LA. I'd want edge cases to be able to be optimally reached as best as possible. At the extreme upper end I could see uses for it in a lot of enrichment projects. Gifted covers this incredible range from kids who are just kind of smart to a relatively profound group of people who honestly do need personalized attention that just won't ever scale to really reach their potential.
On the other other hand I've seen teachers who hand the kids Chromebooks and say go ahead and listen to music on you tube while you do your math problems and then are shocked when those math problems aren't done.
Bad cases make bad law. And a lot of the problems with the modern world is thinking that the existence of edge cases is a good way to set defaults.
Every single school district in the universe has IEPs that can cater for edge cases.
Just like IEPs allow all kinds of changes to testing that normally aren't allowed.
Someone formally diagnosed as autistic could get an IEP allowing use of AI. If, somehow, there was actual empirical evidence beyond just vibes, that was something beneficial to them.
Dunno why we'd skip past demanding empirical evidence of efficacy.
I don't think we should allow edge cases to set defaults. I think the default should be using AI with your students is a bad idea. I think law is a bad way to set norms.
I teach a class of exceptional students at the high end. Part of the tradeoff of that deal is finding ways to create opportunities for them to be challenged and I think careful small scale experiments make a lot of sense here without having to wait for permission from a bureaucracy. The whole reason they send their children to a charter is to avoid this kind of bureaucratic structure. I'm not sure how these laws apply to choice public schools and charters in New York and LA.
It reminds me a lot of the discussion around Khan academy. I'm pretty sure if Khan academy had been around when I was a kid and I'd had access to it, I'd have started learning calculus in the eighth grade or something. Because I was a weird kid who found that stuff really interesting. But it probably wouldn't have had nearly so big an impact on the median kid, who'd maybe look up something when they were confused in a math class or something, but not much more than that.
I agree it's something that's for unusual cases but it just so happens I teach a class of mostly gifted students with a small number of high achievers who are just shy of gifted. The median IQ in my class is like 140. I'm not even saying I have a ready made use for AI but it doesn't seem totally implausible that I would want to have a narrow targeted use.
I think k12 kids shouldn't have general internet access in public schools
It should be limited to research databases
I have developed some units I'm really proud of using internet resources and I think this is kind of insanely proscriptive. Hey I'm allowing you to go to these specific websites for this specific task really opens up opportunities for students to interact with materials.
Just an obvious for instance our state provides a list of whole books which they think are excellent for teaching many of the standards. They do not however provide the books and a google classroom with a PDF of the books. Or I have a unit about the development of American music and readings on Jazz and Blues followed by videos of examples of Louis Armstrong make it stick better than me talking about it ad infinitum.
There's a lot of useful things and the desire to exert uniform control seems pretty harmful especially to the gifted population I've been teaching the last two years.
There's an easy solution for that. Provide the books or provide a pdf of the books on the local database that doesn't have access to the broader internet.
Moreover, i'm a big fan of one standardized curriculum. Like that way, you can do standardized tests on not just reading in a mass. But on social studies, history, science
Etc
But kids cannot have unfettered access to the internet
And it's been proven time and time again. That whatever types of restrictions you put on them, the kids get around it
This goes along with my other idea that we are going to have to get rid of homework. And instead links in the school day so that all schoolwork is done, any proctored environment
Because otherwise kids are gonna have a AI do it
I think I agree with you on standardized curriculum but that’s not where we’re at all. And for advanced learners we’re given even less. Figure something out as long as you get growth.
Google Classroom is a really good bulletin board system. I really doubt the district would create one nearly as capable.
I really think you’re underestimating the power of walking around the room. Wholly technical solutions do fail but the thing that doesn’t is mixing that with old fashioned class management tricks of walking around and paying attention to what kids are paying attention to. 3rd graders aren’t exactly subtle when they’re goofing off.
I agree it theoretically possible
But what's actually happening is kids begged the massive amount of time online
Or watching YouTube
Nice article but I don’t think I agree with this -
“While a student working in an engineering lab at M.I.T. probably should use A.I. to expand the possibilities of their studies and contributions to the world”
Yes that might be fine after they have learnt the fundamentals but not before. I really doubt most students even at top ranked schools will be able to use it that judiciously.
AI-enhanced search is really good at finding research in unusual places - that's the case I could see for a grad student at MIT. If a researcher in Ghana posted something on his department homepage, AI is more likely to find it than any other tool I know.
Good point. I was thinking more about undergrads. Grad students would be different yes
TIL: "big chunks of LI (including malls!) are still on septic! Part of the reason is lack of a sewer system makes it easier to stop any new construction"
https://witchsky.app/profile/did:plc:643p6z4ydk2xnwv77maa6eec/post/3murx3yuiuk2u
Suffolk's 2025 fair-housing analysis actually labels antiquated wastewater infrastructure an “impediment” to higher-density and affordable housing, because it makes that development either impossible or prohibitively expensive. It also says that “expanding access to sewers is a growth-inducing action” and that sewer construction creates the opportunity for increased development density.
https://www.suffolkcountyny.gov/Portals/0/formsdocs/ecodev/Community%20Development/2025_Suffolk_County_Analysis_of_Impediments_to_Fair_Housing_Appendices-Aug_2025-Final.pdf
Apparently the Smith Haven Mall had to build their own mini-wastewater treatment plant because of the lack of a sewer system! https://www.suffolkcountyny.gov/Portals/0/FormsDocs/publicworks/SCSA%20PDFs/Sewer_Agency_Minutes_November_25_2024.pdf
East Hampton's official waterfront plan states that development should continue to be planned so as to avoid the future need for sewers and sewage treatment.
https://dos.ny.gov/system/files/documents/2019/04/completelwrp.pdf
We don’t gotta hand it to anyone, but it genuinely is impressive the extent to which NIMBYs spent the last 50 years innovating how to stop development.
I sometimes wonder how this period is going to echo through history. I’m sure various civilizations have gone through NIMBY phases, but generally speaking the broad sweep of history has been relatively YIMBYtopian — people were usually always happy to expand and develop.
But this crisis was one that most of the West induced on itself for a very specific reason: the environmental backlash against industrial pollution and climate change. Its causality is thus probably pretty unique vs. most other NIMBY phases. And its sheer scope and scale is also pretty unique!
The correction we’re currently witnessing is just in its infancy. It’d be interesting to see what people think about the NIMBY mistake (if they even see it as one!) and how they view the correction to it in a century or two. Will it have been successful? Will it have required some larger, unexpected intervention that we haven’t seen yet? Will it take some sort of turn that we YIMBYs haven’t yet been able to conceive of?
I think about this all of the time. Having ceded so much authority, even veto power, to “the community,” how can even state-level government claw it back for the broader public interest? There are legal strategies that could be pursued but I’m not so sure about the political will that would be required.
Thanks for the excellent article, Halina. My wife teaches at an R1 public university and my daughter just started her first year at a small liberal arts college. In my experience, almost everyone at both types of institutions is hungry for answers to the questions you've raised.
I think the most likely outcome will emerge organically. It will be sloppy, confusing, and often contradictory. Institutions that don't take capabilities or changing incentives seriously are going to struggle, and probably disappear.
Agree, thanks for the great work. It underplayed what I suspect will be the load-bearing* issue: educational institutions are to a great extent *solutions* to the problem of lack of consensus among stakeholders. “The faculty want parking, the alumni want football, the students want sex,” etc. Legibility breaks the mechanism for achieving working consensus — it forces stakeholders to acknowledge tradeoffs among each other.
* yes, I used that AI cliché on purpose
The structure of most academic institutions -- a weak executive, consensus-focused committees, significant autonomy for academic departments -- is poorly suited to deal with the rapidly changing world of AI.
That structure has some upside for individual professors and departments, who can't be blocked or hamstrung by their institution. But a large share of those individuals have to make good choices, or the institution as a whole will lose credibility or become irrelevant.
None of which is to say that I know exactly what a good answer looks like, though I can tell some bad ones a mile away. No one can afford to stay ignorant of these changes, so the ostrich strategy is institutional suicide.
Buried the lede:
"Last month, the Massachusetts Institute of Technology published its committee report on “A.I. use in teaching, learning, and research training,” which is arguably the most exhaustive institutional self-examination on the subject to this point. M.I.T., a leader in both A.I. development and use, has produced a master key that, far beyond acting as a single elite institution’s one-size-fits-all prescription for all educators, is a thorough accounting of every value an educational institution could be organized around and the ways A.I.’s integration can advance or undermine the stated goal."