The Blind Spot in AI Tutors Holding Students and Poverty Alleviation Back

The education industry tells us to care about cognitive skills. Abstract thinking, setting our schedule, recalling facts, and other measurable skills will take us far. Every project brief or job description I see has the phrase "data-driven" in it. Data is very, very important. 

We should be content with any form of improving our cognitive skills. When Massive Open Online Courses (MOOC) platforms like Coursera launched, I was excited about learning from universities I saw in movies. I could finally access the best education. A better future was one math test away.

The Standards Movement of the 1990s made it easy to measure the quality of education. The best student scored the highest in measurable skills, and the best teacher taught the students with the highest scores. I often speak with educational non-profits, EdTech, and institution leaders, and they agree that we must move away from standardization.

This is not a modern stance. Education critics from the 20th century opposed standardized delivery methods, curriculums, and classroom tests. Jonathan Kozol said they conditioned conformism, hindering creativity. Paul Goodman warned how planning the same future for all children would lead to idle consumers in a rat race of competitive, meaningless work. Similar debates happen today daily on and off campuses; if only we had more numbers to finally listen. 

These leaders also highlight another skill set crucial for educational and socio-emotional success: non-cognitive skills. Non-cognitive skills include perseverance, ability to act on long-term plans, socio-emotional regulation, attentiveness, and self-confidence. They impact earnings, employment, labor experience, college attendance, teenage pregnancy, participation in risky activities, compliance with health protocols, and participation in crime. 

In 2013, Economist James Heckman found:

"One percent increase in either type of ability has roughly equal effects on outcomes across the full distribution of abilities." 

Now that many educators agree that non-cognitive skills can be equal or more important predictors of success, it doesn't make sense (for-profit decisions aside) not to walk the talk.

It is frustrating that many people in leadership positions recognize these issues but can't disrupt how their institutions operate. There is too much bureaucracy. Higher-ups show up for unscheduled meetings to talk about profit. Alternative education methods, such as those from Waldorf, Montessori, and St. John's, seem to work on developing cognitive and non-cognitive skills, yet nothing beats the reliable "strategy" of running summer classes.

AI tutors and MOOCs seem to solve so many issues in traditional education. Khan Academy's AI tutor, Khanmigo, offers self-paced, personalized tutoring. Another, Synthesis Tutor, teaches children about Egyptian hieroglyphics and algebra while weaving in Drake memes. There can be a lesson plan, but it's flexible and adapted to each student. Expanding access to education yields large returns on well-being, health, and financial future. Often, any education is better than no education; it is better if personalized. 

However, AI tutors, at least for now, don't address any of the problems they say they want to solve or imply they are solving. AI isn't a "great equalizer" simply because it increases the supply of cognitive skills at a lower cost and on a bigger scale. Believing this diminishes the influence of crucial variables affecting someone's long-term future. These factors include parents and children who do not value education, being born into families without both parents or abusive parents, lack of encouragement to learn from parents, teachers, and institutions, biases towards specific knowledge and people deserving of education, competition between siblings, living in high-risk areas and other unfavorable social environments.

It seems that the current versions of AI tutors could actually further widen the gap between students who wait to be told what to do (most) and those who can own their self-education (minority). Most of last century's students didn't have these non-cognitive skills, as Education Researcher Tony Wagner argues, and current ones don't.

Wagner assessed classroom rigor and college readiness during a visit to early colleges funded by the Bill & Melinda Gates Foundation. In a chemistry course, there were three students whose mixture was the only one giving off a spiral of smoke. Wagner asks what went wrong, and they say, "Donno, we must have mixed it wrong." They don't dare to hypothesize about what went wrong or even know what a hypothesis is. All they knew was that they had followed a recipe, something had gone wrong, and an authoritative figure could fix it.

It's nice I can ask a 24/7 AI tutor what went wrong and how to be guided to an answer. It's nice to feel like there is always an answer. Nevertheless, even if there were always one, most people would only seek it if told to do so.

There are exceptions; there are always exceptions.

The founder of an alternative education school for children shared that some of its alums formed a study group to complete an Ivy League Computer Science program independently. They taught themselves through peer-to-peer tutoring, discussions, and accountability and later got jobs at enviable startups.

I would be more convinced by the great equalizer narrative if everyone possessed the non-cognitive skills of these teenagers, but most of us lack this inner motivation. At 19, I began to self-teach myself communications and went on to work for around fifty startups. I'm a nerd who reads about five books a month. I can easily schedule every hour of my day. Not everyone is like me, and I am not a sample of someone with incredible non-cognitive skills. I do not enroll in MOOCs, use AI tutors to build curriculums of what I'm learning, or have Socratic dialogues with ChatGPT. I know these tools exist, but I don't use them for self-learning. I, a privileged nerd devoted to literacy, resist these tools. How could we assume a child living with one parent in rural Peru who has to work after classes would do any better just because they have 5G?

I do not expect infant versions of AI tutors to solve global educational and wealth inequalities. I do not expect them to ever target foundational causes of poverty, such as malnutrition, the cycle of mental illness, divorce rates, generational poverty's correlation to the present one, lack of affordable housing, etc. However, some advocates for AI tutors frame them as if they could solve these issues, and I am saying no, their current versions can't.

We must find a middle ground when discussing the benefits of AI education. While we should acknowledge AI's potential and encourage researchers, educators, and scientists to continue improving it, we must also avoid over-glorifying the benefits that, while valuable, are mostly limited to the scaling of education supply. Most of us agree that we are more than just numbers on a board and require unmeasurable skills to navigate the world. Only by acting as such can we truly educate more people.