I began my teaching career in 1968! Yikes! In the last 25 years I have seen a myriad of teaching tools and systems come and go. In the Us we are constantly looking for the easy way to do something and in terms of teaching there is no easy way. Most of my years teaching involved the lower grades K-5. Teachers at this level need to know how to teach mathematics. They need help in the classroom from a person who can both teach them and teach them how to teach their students. AI cannot do this. We need to put our energy and money into supplying trained PEOPLE at the elementary level to work side by side with teachers demonstrating, showing , supporting, and talking with them as they watch and see what happens as their children become engaged learners. This takes time. Lesson planning is time intensive and requires the planner to have intimate knowledge of each pupil in their class. This cannot be turned over to technology. As a math specialist who spent 95% of my time in classrooms working alongside teachers I KNOW this way produces the results we are looking for. My teachers became teachers who loved teaching math and their students became active and productive math learners.
100% co-signed. What you describe are extremely time and labor-intensive processes, relational and unscripted. I'm convinced a great deal of the enthusiasm around generative AI in the classroom is the hope for some kind of cheat code around those processes.
Not talking past each other in conversation, basically. You and I meet to talk about education. We both need to size each other up and figure out what common experiences and language we have that can serve our conversation. In computer-based math education, the computer makes almost no effort to figure out what the kid knows and how the kid thinks and talks about it. That results in a cognitive tax the kid and the kid alone has to pay.
More here if you want to cmd f for "cognitive tax".
I agree, but don’t you think we’ll get closer to this (part of the burden will be on the tool) once the tool has a memory and can, over time, develop its own (technical) understanding of the child’s accumulated knowledge? This doesn’t seem to be out of the realm of possibility. I grant that this only is only part of the problem you’re describing.
I cannot imagine this, no. Which could be a failure of my imagination. But I visited a classroom yesterday and upon walking through the classroom door I was instantly flooded with petabytes of data about the learning environment, the people in it, and their relationships. The same with a single interaction I had with a struggling student. Most of that data came through my eyes and ears, which are sensory systems that have been evolving for millions of years. They're extremely sensitive and accurate. I wish the computers best of luck here.
I echo the other comments about content vs connection. I was working on the draft of a paper this weekend talking about that too often teaching is about slicing (thanks to behaviorism) and reducing. Reducing students to objects who behavior must be contained and controlled and rich content that must be sliced into "manageable" pieces.
Not meaning to flatter the host, but when I think of "students and teachers who return to use these tools week after week and day after day" I think of the Desmos graphing calculator, which gets a workout in every one of my classes, every day. And then what strikes me is the "toolness" of it, that it doesn't do a damn thing unless you pick it up, and then it only does what you tell it to do, whether "what you tell it to do" is what you *really need* it to do. Just like a hammer doesn't sit me down for instruction on how to build a house.
Dan, I enjoyed your newsletter this morning. I have a question. In it, you mentioned that you used AI to help with such things as writing the newsletter. My AI detector says that your newsletter is 100% AI generated. Is that accurate, or is my extension mistaken? (I am looking for some calibration here). Thanks!
Content vs connection is spot on. I often wonder if a part of the problem lies in the nature of the software creators. The industry seems to have a high proportion of neurodivergent people who, while very high functioning, may have more difficulty with human connections than neurotypical folks. I recognize the hazards of applying a stereotype here, but the factor should be considered.
I don't know about that, but if we're talking specifically about math instruction, I think it's fair to say these are people for whom traditional methods of math instruction - sit quietly in your seat, watch a lecture, work out these problems and then check against the answer key - worked just fine, thank you very much. So not people who would see a need for any serious rethinking of the curriculum or instruction methods, except for techbro ways to do traditional instruction. Personally, I'd love to see a curriculum designed by people who hated math in school. Might not use all of it, but I bet there would at least be some interesting ideas there.
This goes hand-in-hand with a recurring theme in MIT Prof. Justin Reich’s 2020 book Failure to Disrupt: Why Technology Alone Can’t Transform Education (p. 6):
“In the decades ahead, educators can expect to hear a new generation of product pitches about the transformative potential of new technologies for school systems: how artificial intelligence or virtual reality or brain scanners are the innovations that, this time, will actually lead to profound changes in education. These pitches will also be wrong—these new technologies will not reinvent existing school systems (though some of them may make valuable incremental improvements)—and this book is an effort to explain why.
I have two main arguments, corresponding to the two sections of this book. First, new technologies in education are not, in fact, wholly new; they build on a long history of education innovations. Second, there are certain basic obstacles that time and time again have tripped up the introduction of large-scale learning systems.”
The keyword from the book title is “alone”—it is technology *combined with* good teaching practices, passionate educators, forward-thinking policy makers, dedicated researchers, supportive parents and communities… that can, must, and I believe will transform education.
I’m fundamentally an optimist at heart, how about you?
I’ve spent some time probing ChatGPT and have decided that while it may be able to provide information in a more clever way than Google, it is not capable of leading students to new understanding. I found that by asking it to explain a concept. I used Newton’s second law because I have taught physics. What I got back was a description, not and explanation. I probed further looking to description vs explanation. The system told me it could only respond to my questions, but could not say more than that. I found it was not able to ask leading questions which could lead to better understanding. As such it is not a very good teacher.
By the way it is terrible at the New York Times Spelling Bee. You get nonsense.
I guess the futurists bet is that generating and delivering the right content to the right students will increase the engagement and create this connection by delivering more personalized (less alienating content). Timeline algorithms like TicTock and FB are already really good at hooking people for longer and longer periods of time to useless content, so why would they not have a chance to succeed in education?
Don't know much about TikTok, but what I find addictive about Substack isn't so much the great stuff I get to read (classify that as "interesting but not addictive") but the opportunity to add something to the conversation myself and then see how other real humans respond to that. If the AI math tutor is treating students as empty vessels in need of filling, the parallel for that would be a read-only Substack, or maybe one where you can comment and then get back a "correct/incorrect" response from a machine. No addiction potential in that.
Well I know it's unpolite to ask a chatbot its age, but I think in reality ChatGPT just turned...254! That is, if we trace its origins back to Wolfgang von Kempelen's chess-playing "Mechanical Turk." True, that device turned out to be a bit of a ruse, but perhaps that only underscores the through lines to today's AI efforts.
For a wonderful history of automatons that goes further back than Audrey Watter's (excellent) effort, I recommend historian Jessica Riskin's "The Restless Clock: A History of the Centuries-Long Argument over What Makes Living Things Tick." Includes history of a defecating mechanical duck.
I began my teaching career in 1968! Yikes! In the last 25 years I have seen a myriad of teaching tools and systems come and go. In the Us we are constantly looking for the easy way to do something and in terms of teaching there is no easy way. Most of my years teaching involved the lower grades K-5. Teachers at this level need to know how to teach mathematics. They need help in the classroom from a person who can both teach them and teach them how to teach their students. AI cannot do this. We need to put our energy and money into supplying trained PEOPLE at the elementary level to work side by side with teachers demonstrating, showing , supporting, and talking with them as they watch and see what happens as their children become engaged learners. This takes time. Lesson planning is time intensive and requires the planner to have intimate knowledge of each pupil in their class. This cannot be turned over to technology. As a math specialist who spent 95% of my time in classrooms working alongside teachers I KNOW this way produces the results we are looking for. My teachers became teachers who loved teaching math and their students became active and productive math learners.
100% co-signed. What you describe are extremely time and labor-intensive processes, relational and unscripted. I'm convinced a great deal of the enthusiasm around generative AI in the classroom is the hope for some kind of cheat code around those processes.
Couple to typos there...In the last 55 years! And in the US as in United States.
I have read quite a few takes on the one-year anniversary of ChatGPT and even written one myself. This is the best one by far!
>and for the new ideas to meet the student much more than halfway.
What does this mean?
Not talking past each other in conversation, basically. You and I meet to talk about education. We both need to size each other up and figure out what common experiences and language we have that can serve our conversation. In computer-based math education, the computer makes almost no effort to figure out what the kid knows and how the kid thinks and talks about it. That results in a cognitive tax the kid and the kid alone has to pay.
More here if you want to cmd f for "cognitive tax".
https://danmeyer.substack.com/p/chatbots-have-a-math-problem-and
I appreciate the explanation, thank you.
I agree, but don’t you think we’ll get closer to this (part of the burden will be on the tool) once the tool has a memory and can, over time, develop its own (technical) understanding of the child’s accumulated knowledge? This doesn’t seem to be out of the realm of possibility. I grant that this only is only part of the problem you’re describing.
Speculating, with respect to the “people problem” outlined in your attached post, I could imagine an AI tool, like above, but this time it holds the knowledge of all of the students in a specific classroom, giving it the ability engage the students in the class as a group, drawing on each students’ knowledge (something like you describe Liz Clark-Garvey doing here: https://open.substack.com/pub/danmeyer/p/the-misunderstanding-about-education?r=184vh&selection=3758646b-3cc3-47c5-a347-9d7dafc287ce&utm_campaign=post-share-selection&utm_medium=web).
I cannot imagine this, no. Which could be a failure of my imagination. But I visited a classroom yesterday and upon walking through the classroom door I was instantly flooded with petabytes of data about the learning environment, the people in it, and their relationships. The same with a single interaction I had with a struggling student. Most of that data came through my eyes and ears, which are sensory systems that have been evolving for millions of years. They're extremely sensitive and accurate. I wish the computers best of luck here.
I echo the other comments about content vs connection. I was working on the draft of a paper this weekend talking about that too often teaching is about slicing (thanks to behaviorism) and reducing. Reducing students to objects who behavior must be contained and controlled and rich content that must be sliced into "manageable" pieces.
Not meaning to flatter the host, but when I think of "students and teachers who return to use these tools week after week and day after day" I think of the Desmos graphing calculator, which gets a workout in every one of my classes, every day. And then what strikes me is the "toolness" of it, that it doesn't do a damn thing unless you pick it up, and then it only does what you tell it to do, whether "what you tell it to do" is what you *really need* it to do. Just like a hammer doesn't sit me down for instruction on how to build a house.
Nice spin on the all the birthday announcements! Well played!
Thanks. I'm pretty sure I'll never troll harder than the title of this post.
No worries. The history lesson is a much-needed reminder.
Love what you do by the way!!!
I want to live in that space-age classroom.
Dan, I enjoyed your newsletter this morning. I have a question. In it, you mentioned that you used AI to help with such things as writing the newsletter. My AI detector says that your newsletter is 100% AI generated. Is that accurate, or is my extension mistaken? (I am looking for some calibration here). Thanks!
Hi Linda, I didn't say that I used AI to help write this newsletter. It's 0% AI generated FWIW.
I'm glad to hear that! So much for my extension. I was wondering how accurate it was. . .
Excited to see what AI looks like in the Desmos Curriculum! 👀
Ha thanks but this is nothing like that for now. Just an experimental project.
Darn, I was really looking forward to increasing our state assessment scores with AI. 😢
Content vs connection is spot on. I often wonder if a part of the problem lies in the nature of the software creators. The industry seems to have a high proportion of neurodivergent people who, while very high functioning, may have more difficulty with human connections than neurotypical folks. I recognize the hazards of applying a stereotype here, but the factor should be considered.
I don't know about that, but if we're talking specifically about math instruction, I think it's fair to say these are people for whom traditional methods of math instruction - sit quietly in your seat, watch a lecture, work out these problems and then check against the answer key - worked just fine, thank you very much. So not people who would see a need for any serious rethinking of the curriculum or instruction methods, except for techbro ways to do traditional instruction. Personally, I'd love to see a curriculum designed by people who hated math in school. Might not use all of it, but I bet there would at least be some interesting ideas there.
This goes hand-in-hand with a recurring theme in MIT Prof. Justin Reich’s 2020 book Failure to Disrupt: Why Technology Alone Can’t Transform Education (p. 6):
“In the decades ahead, educators can expect to hear a new generation of product pitches about the transformative potential of new technologies for school systems: how artificial intelligence or virtual reality or brain scanners are the innovations that, this time, will actually lead to profound changes in education. These pitches will also be wrong—these new technologies will not reinvent existing school systems (though some of them may make valuable incremental improvements)—and this book is an effort to explain why.
I have two main arguments, corresponding to the two sections of this book. First, new technologies in education are not, in fact, wholly new; they build on a long history of education innovations. Second, there are certain basic obstacles that time and time again have tripped up the introduction of large-scale learning systems.”
The keyword from the book title is “alone”—it is technology *combined with* good teaching practices, passionate educators, forward-thinking policy makers, dedicated researchers, supportive parents and communities… that can, must, and I believe will transform education.
I’m fundamentally an optimist at heart, how about you?
Follow me on LinkedIn for more https://www.linkedin.com/in/haihaoliu (Substack content coming soon too! https://hhliu.online)
Original post https://www.linkedin.com/posts/haihaoliu_ai-has-changed-a-lot-but-it-hasnt-changed-activity-7135789583228653569-1QOe
I’ve spent some time probing ChatGPT and have decided that while it may be able to provide information in a more clever way than Google, it is not capable of leading students to new understanding. I found that by asking it to explain a concept. I used Newton’s second law because I have taught physics. What I got back was a description, not and explanation. I probed further looking to description vs explanation. The system told me it could only respond to my questions, but could not say more than that. I found it was not able to ask leading questions which could lead to better understanding. As such it is not a very good teacher.
By the way it is terrible at the New York Times Spelling Bee. You get nonsense.
best,
joe
"this new fidget spinner"
Wow, judging from that slam, I'll bet Dan dealt out some harsh barbs when he was in the classroom. Just when you think you know someone . . . 😊
I guess the futurists bet is that generating and delivering the right content to the right students will increase the engagement and create this connection by delivering more personalized (less alienating content). Timeline algorithms like TicTock and FB are already really good at hooking people for longer and longer periods of time to useless content, so why would they not have a chance to succeed in education?
Don't know much about TikTok, but what I find addictive about Substack isn't so much the great stuff I get to read (classify that as "interesting but not addictive") but the opportunity to add something to the conversation myself and then see how other real humans respond to that. If the AI math tutor is treating students as empty vessels in need of filling, the parallel for that would be a read-only Substack, or maybe one where you can comment and then get back a "correct/incorrect" response from a machine. No addiction potential in that.
Well I know it's unpolite to ask a chatbot its age, but I think in reality ChatGPT just turned...254! That is, if we trace its origins back to Wolfgang von Kempelen's chess-playing "Mechanical Turk." True, that device turned out to be a bit of a ruse, but perhaps that only underscores the through lines to today's AI efforts.
For a wonderful history of automatons that goes further back than Audrey Watter's (excellent) effort, I recommend historian Jessica Riskin's "The Restless Clock: A History of the Centuries-Long Argument over What Makes Living Things Tick." Includes history of a defecating mechanical duck.