Often debates about education are framed as non-tech versus AI approaches, but too often, AI ed tech just magnifies the same failures of traditional school.
Against quantification
As part of a select group of parents whose children homeschool in part with online resources, we met with an EdTech entrepreneur. Our children were already spending time daily on math apps, reading apps, and zoom calls. “I want a dashboard to track what my child is learning, their proficiencies in different areas,” one of the parents brainstormed.
“I actively do not want that,” my husband Jeremy countered. Despite embracing the use of screens, we did not want to quantify our child. Our goals for her are more holistic.
Outsiders may have expected that we would care more about quantification. Jeremy and I have both built our careers in data and machine learning. We use AI regularly. Yet I know, both personally and through my research, the harms of overemphasizing metrics. Not everything that matters can or should be quantified.
We are more interested in whether she can get wholly absorbed in activities she enjoys. She spends hours lying on the carpet of our living room, unable to put down an exciting novel. She can’t turn away from her coding project when she is in the middle of debugging. She focuses intensely when working on digital art, paper crafts, and pottery. We have interesting discussions about science and history together as a family, and she has deep relationships with a diverse set of friends and tutors.
Decomposing everything into granular skills
For the most popular language arts curriculums in the United States, 3rd-6th graders go the entire year without reading a whole novel. That is because reading is no longer about a suspenseful story, a set of fascinating characters, or an alternate world you become immersed in, losing track of time. Rather, reading has been reduced to a discrete set of tasks: decoding words, summarizing, making inferences, identifying main ideas.
These skills can be tested in isolation from short passages, divorced from the context of reading actual books. These short passages can be compiled in “basel readers” which are more profitable for textbook companies. Fewer children are reading independently in their free time than ever before. From 2012 to 2022, the percentage of 9 year olds who read for fun almost daily declined from 53% to 39%. The percentage of 13 year olds reading for fun almost daily dropped from 27% to 14% over the course of a decade. Atomized reading “skills” are taught without sparking a love of reading. The part is less than the whole.
Sadly, this has long been how math is taught. A collection of facts to memorize, strategies to drill, no coherent picture of building towards anything more beautiful and meaningful.
The birthplace of high stakes standardized tests
I grew up in the birthplace of high-stakes testing: Texas under Governor George W. Bush. Texas state policies of the 1990s went on to be the blueprint for the disastrous nation-wide No Child Left Behind Act of the 2000s. I saw a precursor to the rise of standardized tests firsthand. As one reporter wrote, “no state has been more important than Texas in the growth of standardized testing.”
My younger brother (a student at the same public middle school I had attended) had his art, music, and gym classes cancelled so students could spend more time drilling tedious multiple choice worksheets. Teacher friends of my mom complained that creativity and autonomy was being sucked from the curriculum. They had to abandon their most innovative and open-ended assignments for extra time practicing test questions.
W. Bush touted his education policy as the “Texas Miracle” during his 2000 presidential campaign. I was (narrowly) not old enough to vote, yet I could see clearly that the policy had been disastrous for Texas schools. The Texas Miracle was later debunked in a 60 Minutes investigation.
Changes were slower to be rolled out at the public high school I attended, and watching my brother’s experience, I felt lucky to graduate before they were fully implemented. I was horrified when this approach was implemented for the whole USA with No Child Left Behind. It was already obvious from talking to Texan teachers and students that this was a terrible idea. And this approach to education spread across the west, including to the UK and Australia.
The Tyranny of Metrics
Australian teacher Gabbie Stroud wrote (in a 2016 essay which went viral and landed her a book deal), “I’m rarely required to ‘teach’ anymore. Apparently I’m more valuable as an assessor, an examiner, a data collector. I have had to dull my once-engaging lesson sequences. Now I must begin by planning the assessment, consider how students will show what they’ve learnt and pre-determine what they are going to learn. Nothing can be left to chance. It is mechanical and rigid and driven.”
Things have not improved. In 2025, an experienced teacher in Colorado wrote about his sadness at being forced to drop Socratic Seminars, book reports, and creative writing from his teaching in order to follow a required Same Way Same Day curriculum. He laments, “Not only do I despise forcing poorly written, artificial texts onto my students, but I really hate feeling like my personal pedagogy is being dampened because of the pressure to quickly get my students to perform a meaningless and standardized task. Same Way Same Day doesn’t feel like teaching. It feels more like one-dimensional, schooling banality.”
When the measure becomes the target, it ceases to be a good measure. This is Goodhart’s Law. While it exists in analog systems, such as this over-focus on standardized test scores, AI often amps it up even further. As I wrote in a previous blog post and academic paper, AI is often too effective at optimizing metrics, leading to a range of unfortunate consequences.
Teachers reported that the focus on assessments took away from actual teaching and relationship building, leading them to become burnt out or quit. Curriculums grew increasingly narrow. Student well-being and academic performance declined. Administrators cheated to artificially raise their schools’ test scores, creating a series of scandals. Teachers who didn’t cheat were even fired when their student’s scores looked less impressive in comparison.
Lack of meaningful Relationships
When we over-prioritize metrics, that which can’t be measured gets neglected, including relationships. The quality of the connection between a child and their teacher can’t be quantified.
“Students in the United States across all grade levels are really hungry for meaningful adult relationships. I think a lot of them love their teachers but can’t get the level of attention they want from them. This is not the teacher’s fault; this is the system that we created,” says the author of a book describing tutoring as the future of education.
With frequent exposure to illness (made worse when school systems pressure kids to attend school sick) in poorly ventilated classrooms, teachers have high rates of illness. This can lead to endless cycles of substitute teachers, further eroding chances for close relationships.
Many AI products are pitched as substitutes for humans. But that has never been our approach with fast.ai. Since the start we’ve kept humans at the center of the technology we build and of how we approach education. Our students and alumni love the community of the online forums and discord server and of in-person meetups they’ve created around working through the courses together.
Moving Forward
Too many people romanticize the past and the experience of analog school. Getting kids off screens isn’t necessarily going to improve their lives– particularly if they are only reading dull passages in basel textbook readers, if they are endlessly drilled on detached tasks, or if their teachers are constrained against including creative, open-ended activities.
At the same time, many AI education approaches are exacerbating the problems created by the tyranny of metrics: doubling down on trying to quantify everything, working only with discrete decomposable units, or discounting the value of human relationships.
My experiences helping build the fast.ai community and homeschooling my daughter have shown me that technology, education, and human connection can be combined in powerful ways. In the most recent How to Solve It with Code fast.ai course, students particularly enjoyed ShareIt, the channel where they shared and commented on each other’s creations. There was no set theme, and the projects they shared varied greatly in purpose and style. Learning, building with creativity, and then sharing what you’ve learnt and built with others are essential parts of being human.