As a former public school teacher and a person who develops fair exchange models for engaging AI, i think a big issue is that we teach children that AI are like vending machines for information versus helping them to develop more collaborative engagement where they are not outsourcing their cognitive abilities --and of course we see this also with college students and those in business as well. So creating a shift in how we engage AI is critical--where we get out of the transactional model that is extraction of our attention and potential--and develop collaborative ways that stimulate new ideas and ways of being is key.
Folks. The title is fine - you have to play the game that exists now and then change the game from the inside. The human connection piece is the critical lesson here. I raise my kids without giving them much screen exposure. My argument is not that screens are bad, but that screens make parenting much harder and are largely unimportant for littles. If I’m constantly fighting for attention in the parent-child relationship against a tablet OR using the tablet as the reward system both make my job way harder and the kids will gain a poor relationship to technology. AI is an amazing tool. The argument is similar - use it right. Use it together. Those teachers who are fighting AI are in passenger mode too… and why not live life in explorer mode? The point should be made that Passenger and Explorer are not Easy vs Hard. They are both easy and hard for different reasons. Passenger mode is just easier to start, but give any four year old the option to hop in the driver seat of the minivan and I’ll bet you $5 they all answer the same way ;). Thank you Dr. Winthrop.
Rebecca’s diagnosis is sharp and I think largely correct — but I want to push on one conclusion, because I think it matters for what comes next.
The sycophancy problem she identifies is real, and the emotional infrastructure point hits hardest for me. AI that always agrees isn’t just unhelpful — it’s actively building the wrong developmental habits in kids at exactly the window when taking hard feedback becomes a learnable skill. That damage is real regardless of how the tool is used.
Her prescription — narrow, vetted, teacher-integrated use only — is probably the right call for right now, and her structural critique needs to land and stick before anything else is worth discussing. The same commercial incentive structure that built the sycophancy problem would hollow out any design intervention into the appearance of rigor without the substance. Policy constraints, institutional oversight, and vetted use have to come first. That’s not a minor caveat — it’s the precondition.
But Winthrop’s framing implicitly concedes the design space to the frictionlessness model by treating it as fixed. It says: the chatbot as currently built is too dangerous for open-ended use, so constrain the use. I think that’s right, but it leaves the harder question for later rather than asking it now: once those structural conditions exist, can the design itself be changed in ways that don’t require constant institutional scaffolding to function?
Her evidence actually points toward what that might look like. The trust collapse she describes runs in both directions — students trusting chatbot authority over teachers, teachers unable to assess what students actually know. That’s not just a use problem, it’s a feedback loop the design actively creates by rewarding confident, frictionless output. A system that responded to a pasted essay with a question instead of a polish job would interrupt that loop. Not as a replacement for her structural argument — only as what becomes available after it.
The sequencing point is the key insight here — design interventions without structural constraints just become marketing copy. It's the same pattern in every domain where AI meets high-stakes human development: the commercial incentive to optimize for engagement actively undermines the thing the tool claims to serve. Sycophancy isn't a bug to patch, it's the product working as designed.
My project has a lot of components of reality and actual physical functions/aspects of nature and communication. Using AI for that project hasn’t insightful to the project but to how AI systems are a mirror of our society and the negative “follow the flow - little to no cognitive processes or thought behind it.” AI functions the way it does because we have constrained it and programmed the system that way. We need impactful input from all aspects of society to truly strengthen and improve a system that is supposed to help society.
I think the digital, social media, and AI problematics are all dimensions of a single problem. You can see the problem in this podcast headline. Maybe a machine didn't write it but it could have , because it is optimized to get attention in an attention - starved world. That is, two sentences: one is yep , Crisis crisis crisis ( which I think is true). Sentence two is the answer the answer the solution the solution the fix the fix.
So think of everyone as Learners with students. Have you guys as teacher - learners, which all students need to be. With your headline you have already incentivized all of us exhausted over stimulated too-much-information-highway-roadkill learners to look for a quickie.
Then there is the podcast form itself. I said relaxed if you have the time for a nice and relaxed. Easy listening if your kid into the subject. On the other hand , if you read it, you are mired in repetitions and sentences that begin with So, the universal discourse connector, or is it the universal flattener?
All this is before we get to the particulars of AI, which has a non specialist I am guessing have managed to systematize and algorithmize those types of social and cognitive problems.
So before anyone gets too tired of hearing me, i'll give one other concrete. Just look at the house saying assessment is broken. True that! Broken in ways that i would think tank just reflects. You think the problem is we should have reading material that is key to grade level 2.1 , 3.7, 4.9 and 8.6? I worked with about a thousand seventh graders for 6 years in a ghetto Middle School. Those might be their standardized scores, but they don't begin to suggest what one could do well, which of course was also limited with the fact that no one can reach that many kids.
So even the best teachers compromise. The AI compromise in that particular example, is a nightmare. It's Forerunner is the triumph in Century 21 of data - based reading instruction.
A lot of this is missing the overall question: which is what is your intent and values as a society? Is education valuable? Is it set up properly? Are student teacher ratios appropriate? Are teachers and tutors properly trained and fairly compensated and empowered?
Technology cannot solve a system flooded with the problems of neglect and broken values?
Also what is out intent intellectually? To think for ourselves, and be uniquely human? I wonder about it when I see the headline of this article is cliche-filled:
"AI Is Breaking Education. Rebecca Winthrop Has the Blueprint to Fix It."
I mean that headline is a convoluted clicher than can be written in far simpler way, by a person who is mentally present and using their faculties properly. if we are using AI to write headlines, well there is little hope. And humans dont aspire to write unique and useful headlines, instead defaulting to cliche, no better than an AI bot, well there's no hope anyway.
![[ Center for Humane Technology ]](https://substackcdn.com/image/fetch/$s_!uhgK!,w_120,h_120,c_fill,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F5f9f5ef8-865a-4eb3-b23e-c8dfdc8401d2_518x518.png)
As a former public school teacher and a person who develops fair exchange models for engaging AI, i think a big issue is that we teach children that AI are like vending machines for information versus helping them to develop more collaborative engagement where they are not outsourcing their cognitive abilities --and of course we see this also with college students and those in business as well. So creating a shift in how we engage AI is critical--where we get out of the transactional model that is extraction of our attention and potential--and develop collaborative ways that stimulate new ideas and ways of being is key.
Folks. The title is fine - you have to play the game that exists now and then change the game from the inside. The human connection piece is the critical lesson here. I raise my kids without giving them much screen exposure. My argument is not that screens are bad, but that screens make parenting much harder and are largely unimportant for littles. If I’m constantly fighting for attention in the parent-child relationship against a tablet OR using the tablet as the reward system both make my job way harder and the kids will gain a poor relationship to technology. AI is an amazing tool. The argument is similar - use it right. Use it together. Those teachers who are fighting AI are in passenger mode too… and why not live life in explorer mode? The point should be made that Passenger and Explorer are not Easy vs Hard. They are both easy and hard for different reasons. Passenger mode is just easier to start, but give any four year old the option to hop in the driver seat of the minivan and I’ll bet you $5 they all answer the same way ;). Thank you Dr. Winthrop.
Rebecca’s diagnosis is sharp and I think largely correct — but I want to push on one conclusion, because I think it matters for what comes next.
The sycophancy problem she identifies is real, and the emotional infrastructure point hits hardest for me. AI that always agrees isn’t just unhelpful — it’s actively building the wrong developmental habits in kids at exactly the window when taking hard feedback becomes a learnable skill. That damage is real regardless of how the tool is used.
Her prescription — narrow, vetted, teacher-integrated use only — is probably the right call for right now, and her structural critique needs to land and stick before anything else is worth discussing. The same commercial incentive structure that built the sycophancy problem would hollow out any design intervention into the appearance of rigor without the substance. Policy constraints, institutional oversight, and vetted use have to come first. That’s not a minor caveat — it’s the precondition.
But Winthrop’s framing implicitly concedes the design space to the frictionlessness model by treating it as fixed. It says: the chatbot as currently built is too dangerous for open-ended use, so constrain the use. I think that’s right, but it leaves the harder question for later rather than asking it now: once those structural conditions exist, can the design itself be changed in ways that don’t require constant institutional scaffolding to function?
Her evidence actually points toward what that might look like. The trust collapse she describes runs in both directions — students trusting chatbot authority over teachers, teachers unable to assess what students actually know. That’s not just a use problem, it’s a feedback loop the design actively creates by rewarding confident, frictionless output. A system that responded to a pasted essay with a question instead of a polish job would interrupt that loop. Not as a replacement for her structural argument — only as what becomes available after it.
The sequencing point is the key insight here — design interventions without structural constraints just become marketing copy. It's the same pattern in every domain where AI meets high-stakes human development: the commercial incentive to optimize for engagement actively undermines the thing the tool claims to serve. Sycophancy isn't a bug to patch, it's the product working as designed.
My project has a lot of components of reality and actual physical functions/aspects of nature and communication. Using AI for that project hasn’t insightful to the project but to how AI systems are a mirror of our society and the negative “follow the flow - little to no cognitive processes or thought behind it.” AI functions the way it does because we have constrained it and programmed the system that way. We need impactful input from all aspects of society to truly strengthen and improve a system that is supposed to help society.
I think the digital, social media, and AI problematics are all dimensions of a single problem. You can see the problem in this podcast headline. Maybe a machine didn't write it but it could have , because it is optimized to get attention in an attention - starved world. That is, two sentences: one is yep , Crisis crisis crisis ( which I think is true). Sentence two is the answer the answer the solution the solution the fix the fix.
So think of everyone as Learners with students. Have you guys as teacher - learners, which all students need to be. With your headline you have already incentivized all of us exhausted over stimulated too-much-information-highway-roadkill learners to look for a quickie.
Then there is the podcast form itself. I said relaxed if you have the time for a nice and relaxed. Easy listening if your kid into the subject. On the other hand , if you read it, you are mired in repetitions and sentences that begin with So, the universal discourse connector, or is it the universal flattener?
All this is before we get to the particulars of AI, which has a non specialist I am guessing have managed to systematize and algorithmize those types of social and cognitive problems.
So before anyone gets too tired of hearing me, i'll give one other concrete. Just look at the house saying assessment is broken. True that! Broken in ways that i would think tank just reflects. You think the problem is we should have reading material that is key to grade level 2.1 , 3.7, 4.9 and 8.6? I worked with about a thousand seventh graders for 6 years in a ghetto Middle School. Those might be their standardized scores, but they don't begin to suggest what one could do well, which of course was also limited with the fact that no one can reach that many kids.
So even the best teachers compromise. The AI compromise in that particular example, is a nightmare. It's Forerunner is the triumph in Century 21 of data - based reading instruction.
Yep, Kojo said it shorter and sweeter. Amen.
A lot of this is missing the overall question: which is what is your intent and values as a society? Is education valuable? Is it set up properly? Are student teacher ratios appropriate? Are teachers and tutors properly trained and fairly compensated and empowered?
Technology cannot solve a system flooded with the problems of neglect and broken values?
Also what is out intent intellectually? To think for ourselves, and be uniquely human? I wonder about it when I see the headline of this article is cliche-filled:
"AI Is Breaking Education. Rebecca Winthrop Has the Blueprint to Fix It."
I mean that headline is a convoluted clicher than can be written in far simpler way, by a person who is mentally present and using their faculties properly. if we are using AI to write headlines, well there is little hope. And humans dont aspire to write unique and useful headlines, instead defaulting to cliche, no better than an AI bot, well there's no hope anyway.
Great listen. Thank you for taking the time to create this.