Whenever I hear people say something like "this will replace 90% of the administrative work teachers have to do so they can focus on student-centered tasks" I'm reminded that the real reason teachers feel overburdened with administrative tasks is a simple one: they have too many classes with too many students, and definitely way too many parents. We make these capitalist assumptions that the key to solving education's problems lies with new technology, when really we could just choose to fund education -- hire more teachers, reduce class sizes and course loads, and voilà suddenly there is plenty of time to focus on students.
One thing that teachers seem to have lost touch with is that post-COVID, students perceive watching a video as work. I literally had an eighth grader wail, "The video is FOUR MINUTES?!" in my class this morning. A screen doesn't make a task inherently engaging the way it used to. A chatbot referring them to a video is a double dose of I'm-not-doing-that.
This matches my experience as well, and is under-discussed within edtech to an almost shocking degree. There was a time where pulling out the laptops was a novelty. Something exciting and different was going to happen. Now, especially post-COVID, the allure simply isn't there. Screens have to bring SO much more value than they used to.
When ChatGPT 4o first came out and Sal Khan's video with his son went viral, I tried my hand with getting help on a simple two-step linear equation. I tried to have it explain a particular step in different ways, because I pretended to not understand the first way it showed me, and got frustrated because it kept repeating itself. So, to use a phrase I heard a student use, "B----, I told you I don't understand!" That's all I needed to say. It responded back with "I'm sorry, you are correct", and then let me continue to make a couple of mistakes, getting the answer wrong, but having it praise me for a job well done and that my final incorrect answer was correct.
When all the incentives for students are in completing the assignment, getting the grade, and moving on; when all the incentives for school and district administers are around compliance, test scores, and funding/spending then no tool is going to make a material difference on improving education. IMO
"Kids do not seem like they want to talk to LLM chatbots about academic content,"
Like you, I was an AI skeptic and to a certain degree still am.I did a summer internship the machine learning and I'm well aware of how the AI modules come up with their knowledge. It's very unnerving.
However I needed to do some makeup p d hours and attended a session on Flint, which is based on Claude. I got the idea of building these interactive sessions that are not about a worksheet per se but about allowing the student to direct Flint to get better at a specific skill they choose.
I am teaching 2 classes this year ( Discovering Geometry and algebra support) with students who struggle in math and in both classes there are students who really would like to get better at math, hard working kids who seriously lack skill. I also have students who completely goof around during math class because long ago they just checked out and don't know even basic skills like addition error multiplication math facts.
I am not a teacher who puts laptops in front of students other than desmos activities that I either built or found that I think will engage and teach them.But I have found significant success with 30-40% of these students who see this as a tool not as an answer generating machine.
They've learned to say things like "stop.You're giving me too much information.Keep it to a top three short list." Or "can you give me a similar problem with only numbers and not variables so I can work out the pattern of how to solve this problem?"Sometimes I let Flint direct the problem set with explicit instructions on the types of problems I want assigned (stressing when in doubt lighten digmffuculty) but always allowing student direction. Other times they're working with the worksheet and describe the problem to flint and ask for advice on approach . I can track the history of all their activity. And absolutely there are some kids who don't use it or who just sit there and do nothing and goof around until I find them.I don't want to make it sound like a perfect engagement tool.But it is doing something that I find valuable which is giving students who want it an opportunity to direct their learning.
I have also had two extremely specific cases of students who goof around all the time and just finally confessed to me that they did not know any of their math facts.So I set up an AI program for that could learn their math facts.And it has been a remarkable success.Both of them got significantly better at their mathematics and also are now willing to work acquiring new knowledge. They ask AI to drill them on a specific group of math facts.And they write down on paper.All the ones that they miss.Then they study those and they say okay.Now drill me on the ones that I miss. Both of them have gone from literally knowing no facts to close to complete mastery. In both cases, these are clearly relatively capable students who had just tuned out for all of middle school math, not ones with a cognitive issue with memorization, which flint would not help with much.
Some of them have come to me and ask if I could set up a session for them so they could use it in a writing class or in a science class. These are kids who know about chatgpt, who could if they wanted to just use AI to get answers or cheat and are instead using Flint to actually build their knowledge or skills.
I have also used it in two other settings.I teach engineering and have found it very difficult to teach analytical writing. For example they watched a documentary on the challenger and I wanted them to present an opinionated analysis finding fault for the explosion. I've been doing this activity for several years and consistently got mostly essays describing the events without any analysis or state their opinion again without analysis. So I decided to try using it with this product Flint and got again a remarkable improvement. I set up an activity telling Flint to argue with the students. Don't be nice , don't always compliment their ideas. Disagree and push them to improve. So flint in obnoxious mode was very helpful in driving the kids to not only improve their writing but also learn that flint was sometimes being unreasonable or inaccurate. It was really funny.Students were getting actively frustrated with Flint, telling him that he didn't know what he was talking about that Alan McDonald wasn't an engineer working on the boosters, that it wasn't fair that larry mulloy used his funding power to override Morton Thiokol, that this was an indisputable element in assigning blame. And while sometimes they would change their opinion other times they just learned how to make their opinion stronger. For the first time the results from even the weaker, less engaged students showed they had engaged with the facts rather than just repeat what they saw on the screen.
My favorite example is from an admittedly extremely bright student who argued that the only way the boosters would ever have been fixed was for an accident like this to happen, that the only way the public would support fixing the boosters was if people died. He used the funding history from those years to prove his point. Totally upset Flint; I
Told the student to remember to remind flint that his analysis doesn't mean he approves of this outcome just as an unfortunately reality and to show empathy in his writing to make sure that people understood this. They argued about it for about an hour and at the end of it Flint acknowledged that the student had convinced him. The students said o k now give me three paragraphs restating my opinion. He then showed it to me and asked if he could submit it.I said hell no , but he could use it as a source to rewrite his own thoughts.
I've also used it in stats with seniors who are totally in checkout mode. But I am also teaching them excel which they realize they're going to need in college. So for example.
I could tell flint to give them problems that they could solve on excel in binomial distribution. First, they learned how to create a worksheet with variables that reset based on the trials and the probability of success this. We did as a class. Then I could tell Flint to give them a bunch of problems of differing levels of difficulty and allow them to plug it in and see what worked. And then if they were having problems, they could upload what they had from the spreadsheet and flip could look at it and suggest what they were doing wrong.This has been very useful and the students are interested in working with it. And once again it's not perfect. There were certainly students who are sitting there doing nothing. But most of the students saw a significant improvement in their understanding and also in the use of excel.
The most notable successes have been in my discovering geometry class as these students have a long history of math struggle. Many of them would love to get better. I don't use it as much in my algebra support class.Because most of these students simply need to be kicked into doing their homework but a few of the more highly motivated ones who often have trouble translating their knowledge into application have found it very useful and I often find them using it without my having even assigned it. I'm able to review their results I can see how they are asking for help with a particular subject that they find tricky.
If I don't use Flint as a tutor but more as a friend, an opinionated smart but occasionally wrong friend and allowed the students to direct their learning it has been a useful tool. The key issue here is agency. Every time rhey use AI I tell the students that the goal is not to get the answers.The goal s to learn how to direct this tool to help them gain skills not only in math but in asking for help in meeting their goals, not mine.
But I do wonder if the massive investments already made do create a desperate need to see this tech as "transformative" just like the airplane or television or the internet.
If I put billions of dollars into a hammer factory, education (and everything else) would start to look a little nailish.
And education has about $1T worth of nails annually. 80% of them are teachers, though. So to the extent they are looking to recoup their investment in hardware & compute through education revenue, they need it to come from districts hiring many, many fewer teachers.
It's a pretty basic math problem, isn't it? If you're not firing all the teachers, you're expecting schools to come up with lots of EXTRA MONEY to pay for the AI whiz-bang? And when have schools ever had extra money?
I'm just wondering if a human tutor can suggest materials to study or taking a break...
On the other hand, prompt engineering sooner or later should be replaced with some models. And I believe (actually, I'm almost sure) these models will have a more diverse approach to students than even the best teacher ever.
I really would be happy to know how much such cases the chatbots have. If it's just edge cases or not...
seems like an easy fix. pretty soon, there will only be a tiny % of human tutors better than an ai. next generation will be more comfortable with ai tutor than an annoying human.
You are correct...."pretty soon there will only be a tiny % of human tutors" and this breaks my heart. All of the potentially great tutors and teachers will be gone. I suggest we stop paying the older generation to teach the younger generation today so we can put an end to all this silly "tutoring and teaching". AI can replace the expensive salaries and pensions and we can all live in the AI generated universe. I just hope AI is willing to pay all of the taxes.
I think this happened with tomatoes, didn't it? Flood the market with factory-farm-produced flavorless tomatoes, after a while people forget what real tomatoes taste like, they stop expecting tomatoes to taste like anything, it's a garnish you put on something for color.
Of course, we didn't ALL forget what real tomatoes tasted like, and now real tomatoes can be back as a niche, farmer's market/Whole Foods product - for more money of course.
I suppose the same thing could happen with teaching, children of the working class never know there was something better, those who can pay get artisanal teaching.
Sadly, in my area.the "fancy" farmers market (where everyone brings their dog??) has a wonderful organic family run farm and they get....about 1.99/lb for the delicious tomatoes. No one seems willing to pay the actual cost for "real goods" anymore, cheaper substitutes in large numbers seem to be the standard protocol.
In education, most of the money is spent on an army of poorly paid aides, substitutes and software. A substitute teacher example is here.....
paying a reasonable salary to a substitute is out of the question.
Working class or upper middle class.... no one gets a well paid teacher in their classroom supporting students anymore. Money is going to an army of barely paid contingent staff that won't complain about ANYTHING and the students get left behind.
Your analysis of the current state always seems incisively honest and your predictions (or the predictions of others you platform) about the fundamental limitations of this technology seem short-sighted and hubristic. I may have missed it, but have you written about how despite the exponential improvements expected by the average AI researcher, they actually will not lead to significantly different outcomes from the current status quo?
Whenever I hear people say something like "this will replace 90% of the administrative work teachers have to do so they can focus on student-centered tasks" I'm reminded that the real reason teachers feel overburdened with administrative tasks is a simple one: they have too many classes with too many students, and definitely way too many parents. We make these capitalist assumptions that the key to solving education's problems lies with new technology, when really we could just choose to fund education -- hire more teachers, reduce class sizes and course loads, and voilà suddenly there is plenty of time to focus on students.
🎯
e.g. "AI can help you grade 125 student essays" without wondering why we ask teachers to grade 125 student essays.
One thing that teachers seem to have lost touch with is that post-COVID, students perceive watching a video as work. I literally had an eighth grader wail, "The video is FOUR MINUTES?!" in my class this morning. A screen doesn't make a task inherently engaging the way it used to. A chatbot referring them to a video is a double dose of I'm-not-doing-that.
This matches my experience as well, and is under-discussed within edtech to an almost shocking degree. There was a time where pulling out the laptops was a novelty. Something exciting and different was going to happen. Now, especially post-COVID, the allure simply isn't there. Screens have to bring SO much more value than they used to.
When ChatGPT 4o first came out and Sal Khan's video with his son went viral, I tried my hand with getting help on a simple two-step linear equation. I tried to have it explain a particular step in different ways, because I pretended to not understand the first way it showed me, and got frustrated because it kept repeating itself. So, to use a phrase I heard a student use, "B----, I told you I don't understand!" That's all I needed to say. It responded back with "I'm sorry, you are correct", and then let me continue to make a couple of mistakes, getting the answer wrong, but having it praise me for a job well done and that my final incorrect answer was correct.
https://chatgpt.com/share/c2b9c6be-487a-4cf8-9eea-f99fd0042b9e
When all the incentives for students are in completing the assignment, getting the grade, and moving on; when all the incentives for school and district administers are around compliance, test scores, and funding/spending then no tool is going to make a material difference on improving education. IMO
"Kids do not seem like they want to talk to LLM chatbots about academic content,"
Like you, I was an AI skeptic and to a certain degree still am.I did a summer internship the machine learning and I'm well aware of how the AI modules come up with their knowledge. It's very unnerving.
However I needed to do some makeup p d hours and attended a session on Flint, which is based on Claude. I got the idea of building these interactive sessions that are not about a worksheet per se but about allowing the student to direct Flint to get better at a specific skill they choose.
I am teaching 2 classes this year ( Discovering Geometry and algebra support) with students who struggle in math and in both classes there are students who really would like to get better at math, hard working kids who seriously lack skill. I also have students who completely goof around during math class because long ago they just checked out and don't know even basic skills like addition error multiplication math facts.
I am not a teacher who puts laptops in front of students other than desmos activities that I either built or found that I think will engage and teach them.But I have found significant success with 30-40% of these students who see this as a tool not as an answer generating machine.
They've learned to say things like "stop.You're giving me too much information.Keep it to a top three short list." Or "can you give me a similar problem with only numbers and not variables so I can work out the pattern of how to solve this problem?"Sometimes I let Flint direct the problem set with explicit instructions on the types of problems I want assigned (stressing when in doubt lighten digmffuculty) but always allowing student direction. Other times they're working with the worksheet and describe the problem to flint and ask for advice on approach . I can track the history of all their activity. And absolutely there are some kids who don't use it or who just sit there and do nothing and goof around until I find them.I don't want to make it sound like a perfect engagement tool.But it is doing something that I find valuable which is giving students who want it an opportunity to direct their learning.
I have also had two extremely specific cases of students who goof around all the time and just finally confessed to me that they did not know any of their math facts.So I set up an AI program for that could learn their math facts.And it has been a remarkable success.Both of them got significantly better at their mathematics and also are now willing to work acquiring new knowledge. They ask AI to drill them on a specific group of math facts.And they write down on paper.All the ones that they miss.Then they study those and they say okay.Now drill me on the ones that I miss. Both of them have gone from literally knowing no facts to close to complete mastery. In both cases, these are clearly relatively capable students who had just tuned out for all of middle school math, not ones with a cognitive issue with memorization, which flint would not help with much.
Some of them have come to me and ask if I could set up a session for them so they could use it in a writing class or in a science class. These are kids who know about chatgpt, who could if they wanted to just use AI to get answers or cheat and are instead using Flint to actually build their knowledge or skills.
I have also used it in two other settings.I teach engineering and have found it very difficult to teach analytical writing. For example they watched a documentary on the challenger and I wanted them to present an opinionated analysis finding fault for the explosion. I've been doing this activity for several years and consistently got mostly essays describing the events without any analysis or state their opinion again without analysis. So I decided to try using it with this product Flint and got again a remarkable improvement. I set up an activity telling Flint to argue with the students. Don't be nice , don't always compliment their ideas. Disagree and push them to improve. So flint in obnoxious mode was very helpful in driving the kids to not only improve their writing but also learn that flint was sometimes being unreasonable or inaccurate. It was really funny.Students were getting actively frustrated with Flint, telling him that he didn't know what he was talking about that Alan McDonald wasn't an engineer working on the boosters, that it wasn't fair that larry mulloy used his funding power to override Morton Thiokol, that this was an indisputable element in assigning blame. And while sometimes they would change their opinion other times they just learned how to make their opinion stronger. For the first time the results from even the weaker, less engaged students showed they had engaged with the facts rather than just repeat what they saw on the screen.
My favorite example is from an admittedly extremely bright student who argued that the only way the boosters would ever have been fixed was for an accident like this to happen, that the only way the public would support fixing the boosters was if people died. He used the funding history from those years to prove his point. Totally upset Flint; I
Told the student to remember to remind flint that his analysis doesn't mean he approves of this outcome just as an unfortunately reality and to show empathy in his writing to make sure that people understood this. They argued about it for about an hour and at the end of it Flint acknowledged that the student had convinced him. The students said o k now give me three paragraphs restating my opinion. He then showed it to me and asked if he could submit it.I said hell no , but he could use it as a source to rewrite his own thoughts.
I've also used it in stats with seniors who are totally in checkout mode. But I am also teaching them excel which they realize they're going to need in college. So for example.
I could tell flint to give them problems that they could solve on excel in binomial distribution. First, they learned how to create a worksheet with variables that reset based on the trials and the probability of success this. We did as a class. Then I could tell Flint to give them a bunch of problems of differing levels of difficulty and allow them to plug it in and see what worked. And then if they were having problems, they could upload what they had from the spreadsheet and flip could look at it and suggest what they were doing wrong.This has been very useful and the students are interested in working with it. And once again it's not perfect. There were certainly students who are sitting there doing nothing. But most of the students saw a significant improvement in their understanding and also in the use of excel.
The most notable successes have been in my discovering geometry class as these students have a long history of math struggle. Many of them would love to get better. I don't use it as much in my algebra support class.Because most of these students simply need to be kicked into doing their homework but a few of the more highly motivated ones who often have trouble translating their knowledge into application have found it very useful and I often find them using it without my having even assigned it. I'm able to review their results I can see how they are asking for help with a particular subject that they find tricky.
If I don't use Flint as a tutor but more as a friend, an opinionated smart but occasionally wrong friend and allowed the students to direct their learning it has been a useful tool. The key issue here is agency. Every time rhey use AI I tell the students that the goal is not to get the answers.The goal s to learn how to direct this tool to help them gain skills not only in math but in asking for help in meeting their goals, not mine.
Thanks for taking the time to write this long comment!
Though AI evangelists over promise and under deliver, it's also incorrect to dismiss AI.
I like that you explain how even in its flawed state, it can still be a useful resource and help motivate students.
Could I quote you in an article I'm writing with a nuanced take on the current state of AI in education?
Dan, I’ve long been a fan of yours and this is truly excellent! Great reading and makes me wonder where we’re going these days with
teaching.... So many want a quick, easy answer on how to teach math, but a lot of it comes down to personality, communication
between teacher and students, and acceptance among all that it takes different approaches to solving problems.... I believe more
needs to be done on this in the teacher credential programs.... It’s one thing to know the math, but way more to know how to connect
and identify with students so that they feel their thoughts and explanations are relevant and worth sharing .... And, of course, that
mistakes are a big part of the learning process.... It really builds trust and encourages effective “struggle”.... I’m not at all convinced
that AI will always have the ability or the answer on how to do this.... It’s a good start for some things, but we need to keep the
“human” connection alive in education.... Thanks so much for sharing this! Jeanne Lazzarini
I don't know if AI truly is a bubble, I don't follow the stock market and don't much care:
https://prospect.org/power/2025-03-25-bubble-trouble-ai-threat/
But I do wonder if the massive investments already made do create a desperate need to see this tech as "transformative" just like the airplane or television or the internet.
If I put billions of dollars into a hammer factory, education (and everything else) would start to look a little nailish.
And education has about $1T worth of nails annually. 80% of them are teachers, though. So to the extent they are looking to recoup their investment in hardware & compute through education revenue, they need it to come from districts hiring many, many fewer teachers.
It's a pretty basic math problem, isn't it? If you're not firing all the teachers, you're expecting schools to come up with lots of EXTRA MONEY to pay for the AI whiz-bang? And when have schools ever had extra money?
Gosh, I'm gonna be rich! Thanks for the tip!
I'm just wondering if a human tutor can suggest materials to study or taking a break...
On the other hand, prompt engineering sooner or later should be replaced with some models. And I believe (actually, I'm almost sure) these models will have a more diverse approach to students than even the best teacher ever.
I really would be happy to know how much such cases the chatbots have. If it's just edge cases or not...
seems like an easy fix. pretty soon, there will only be a tiny % of human tutors better than an ai. next generation will be more comfortable with ai tutor than an annoying human.
You are correct...."pretty soon there will only be a tiny % of human tutors" and this breaks my heart. All of the potentially great tutors and teachers will be gone. I suggest we stop paying the older generation to teach the younger generation today so we can put an end to all this silly "tutoring and teaching". AI can replace the expensive salaries and pensions and we can all live in the AI generated universe. I just hope AI is willing to pay all of the taxes.
I think this happened with tomatoes, didn't it? Flood the market with factory-farm-produced flavorless tomatoes, after a while people forget what real tomatoes taste like, they stop expecting tomatoes to taste like anything, it's a garnish you put on something for color.
Of course, we didn't ALL forget what real tomatoes tasted like, and now real tomatoes can be back as a niche, farmer's market/Whole Foods product - for more money of course.
I suppose the same thing could happen with teaching, children of the working class never know there was something better, those who can pay get artisanal teaching.
I kinda thought that was the point 🤔
Sadly, in my area.the "fancy" farmers market (where everyone brings their dog??) has a wonderful organic family run farm and they get....about 1.99/lb for the delicious tomatoes. No one seems willing to pay the actual cost for "real goods" anymore, cheaper substitutes in large numbers seem to be the standard protocol.
In education, most of the money is spent on an army of poorly paid aides, substitutes and software. A substitute teacher example is here.....
https://www.indeed.com/cmp/Central-Bucks-School-District-1/salaries/Substitute-Teacher/Doylestown-PA.......$17.77 an hour...they can buy less than 10 lbs of tomatoes if they don't need to pay taxes for an hour of work.
With 30% of the staff "chronically absent" according to Propublica's Miseducation (more than 10 absences per teacher in 180 day school year) https://projects.propublica.org/miseducation/district/4205310
paying a reasonable salary to a substitute is out of the question.
Working class or upper middle class.... no one gets a well paid teacher in their classroom supporting students anymore. Money is going to an army of barely paid contingent staff that won't complain about ANYTHING and the students get left behind.
Your analysis of the current state always seems incisively honest and your predictions (or the predictions of others you platform) about the fundamental limitations of this technology seem short-sighted and hubristic. I may have missed it, but have you written about how despite the exponential improvements expected by the average AI researcher, they actually will not lead to significantly different outcomes from the current status quo?