The families with the most resources are buying more human support, not less.
GUEST COLUMN | by Alicia Kacharia
There is a lot of excitement about how, for the first time, every student can access a personalized tutor regardless of income, location, or level. To be fair, AI is great at delivering detailed, personalized content. I believe AI can personalize better than any teacher managing 50 students. So, for a motivated student conjugating French at 11 p.m., AI is an incredibly useful tool.
‘…for a motivated student conjugating French at 11 p.m., AI is an incredibly useful tool.’
That being said, for many of the students who need the most support, content was never the binding constraint. Rather, three conditions often determine whether learning begins in the first place:
1. Stability: Is the student fed, housed, and safe enough to think past tomorrow?
2. Motivation: Does a student believe effort will change their outcome?
3. Accountability: Does an adult expect something from them, and will they notice if the student disappears?
The third condition is one of the most important ways teachers support their students. And it is a condition that AI does not currently—and should not—replace.
AI Works Best When the Conditions for Learning Already Exist
AI tutoring startups are founded by Oxford and Stanford CS grads who are “building the platform they wish they had.”
Many of these very talented founders are at the top of their class: motivated, stable, and held to high expectations by the adults around them. For a student like this, content may genuinely be the constraint: they need better explanations, more practice, or the ability to move faster. This issue often does not generalize to students experiencing persistent educational instability.
The global private tutoring market was worth roughly $92 billion when ChatGPT launched and is projected to reach $155 billion by 2030. In the United States, it is growing at 11% a year, adding nearly $29 billion by 2029. All of that is happening alongside the most aggressive deployment of AI tutoring in history.
If AI has made high-quality personalized instruction essentially free, why does demand for expensive human tutoring remain so strong among families who can afford both? When families who can afford both AI and human tutoring continue to pay for the human, it suggests they are paying for something beyond personalized content.
Content Is Not the Limiting Factor
There is a massive amount of information online. Although AI is helpful in finding, processing, and explaining, that is not enough to make a good teacher.
‘Although AI is helpful in finding, processing, and explaining, that is not enough to make a good teacher.’
For example, one of the biggest successes in edtech, Khan Academy, found that students who use it more than thirty minutes a week improve. However, even Khan Academy’s own research emphasizes the role of educators in getting students to use the tool consistently.
The obvious objection is that AI is getting better at simulating accountability: it checks in, follows up, adjusts its tone. Accountability, however, goes beyond the act of following up. Students need to know that someone is invested in what happens to them.
An instructor at my company once worked with a student on agmatine. No high schooler is naturally passionate about agmatine, but the instructor was visibly excited by the research and the student caught it. That student showed up for the instructor, not the topic, and over time caught the topic.
AI Should Do the Noticing. A Person Should Do the Following Up.
One-on-one human tutoring has been shown to produce gains that were two standard deviations above traditional classrooms. We have spent billions of dollars across decades trying to replicate those gains with software by personalizing content delivery at scale. Meanwhile, the income achievement gap has widened by roughly 40% and now exceeds the Black-white gap. We may have been scaling the wrong part of tutoring.
We have spent decades trying to make the benefits of one-on-one tutoring scalable without actually scaling the one thing that made it powerful: the human relationship.
‘We have spent decades trying to make the benefits of one-on-one tutoring scalable without actually scaling the one thing that made it powerful: the human relationship.’
A harder question is how to get more adults into the lives of students who do not have enough of them. We know that works: high-dosage tutoring has produced some of the largest learning gains on record, Chicago’s SAGA program doubled or tripled a year of math, and Mississippi’s statewide literacy program runs about $32 per student.
However, the above costs money, and budgets often move the other way. Given this, a more realistic goal is not more adults but better-aimed ones. AI can help with this.
Going back to the third condition, a student needs someone who will notice if they disappear. AI can do the noticing and spot the student whose submission pattern changed or who went quiet in week three. But AI should not be the one who follows up, engages, and gets to know the student. Students need to know that someone cares about their success.
Many products detect disengagement and then resolve it themselves with an automated nudge. The better architecture routes to a teacher. So for anyone buying, make sure to ask: does this tool put a teacher in front of a student, or stand in for one?
The digital divide of the 1990s was about who had a computer. The one we are building now is about who gets a human. Wealthy students will have teachers, tutors, and mentors, and AI on top of all of it. Other students could end up with fewer caring adults instead.
0 Comments