Bridging the Achievement Gap with AI
Today, more than 50 million American children are sitting in K-12 classrooms. They’re learning to read, conducting science experiments, and falling asleep to the soothing sounds of differential calculus. Do you know what most of them aren’t doing? Learning about the technologies that will have the most significant impact on their lives.
In Artificially Human, I wrote about how artificial intelligence (AI) is set to disrupt higher education. I avoided talking about K-12 education for two reasons. First, it’s not my area of expertise. I roll my eyes when parents say they want to “take control” of their children’s education. The little I know about psychology and neuroscience is enough to realize I should probably stay out of the way and leave education to the experts.
Second, I generally agree with what others have to say. You can find plenty of content about AI disrupting K-12 education. There are valid points regarding the benefits (e.g., increased personalization) and risks (e.g., reinforcing biases). I was less impressed with the conversation around higher education, so I covered that in my book.
Then, I read some interesting research about how AI affects worker performance in the private sector. It changed how I think about the benefits of adopting AI in K-12 education. I’m no longer convinced that we’re having the right conversation.
Mind the Gap
What is education? It’s a system for building cognitive, social, and emotional intelligence in humans.
What is AI? At a basic level, it’s manufactured intelligence. The word “intelligence” is literally in the name of the technology.
Why is the education system constrained to human intelligence? In the future, each person's potential will be dictated by their biological intelligence PLUS the artificial intelligence accessible to them.
Cleo Abram explains this concept in her recent video. She states, “[AI] can help us shrink the gap between the ideas in our heads and what we can make.” Abram is talking about music, but the idea applies generally.
We all have ideas. We’re only paid for our ideas if we can translate them into a product or service others will buy. In some cases, we sell products and services directly to others. More frequently, we sell them to an employer for a steady paycheck.
Today, the ability to translate ideas into outputs is unevenly distributed. Say you grow up in a household where your physical or psychological safety is constantly under threat. Many of your 80+ billion neurons will be dedicated to keeping you safe. Can you sell your safety as a good or service? Probably not. Your output isn’t valuable to others.
The brain is malleable, but we can’t expect teachers to bridge achievement gaps through instruction alone. Critical neural pathways are established before students set foot in a classroom and are reinforced by ongoing trauma outside school. Educators will tell you they can shift a student’s trajectory, but persistent achievement gaps are frustratingly difficult to close.
Students begin school from different starting points and have unique experiences outside the classroom. Believing we can close achievement gaps purely by investing in human intelligence is a pipe dream. We need another way to close the gap between the ideas in students’ heads and what they can make.
Boosting Achievement with AI
It’s too early to tell how AI might change education. School boards, teachers, and administrators are understandably skeptical. Ill-informed experimentation can have disastrous consequences.
That said, we can glean insights from early experiments in the private sector. In one study, AI tools provided to 5,179 customer support agents increased productivity, as measured by issues resolved per hour, by an average of 14%.
Another study involved 758 consultants divided into three groups. Those researchers found “consultants using AI were significantly more productive (they completed 12.2% more tasks on average, and completed tasks 25.1% more quickly), and produced significantly higher quality results (more than 40% higher quality compared to a control group).”
Impressive results, but the impact wasn’t evenly distributed. In both cases, the lowest-performing workers improved the most. Novice and low-skilled customer support agents saw a 35% improvement compared with minimal impact on high-skilled workers. In the second study, all consultants benefited from AI, but above-average performers only improved by 17%, whereas below-average performers improved by a whopping 43%.
I often hear AI compared to the introduction of the calculator in the context of K-12 education. That’s misleading. A calculator is a tool. It allows you to add, subtract, multiply, and divide numbers more efficiently. AI isn’t a tool. It’s augmented intelligence.
AI models are typically trained on data generated by top-performing humans. Nobody trains AI models on my B- high school English papers. They train them on published works that meet a higher standard.
That’s why we don’t see as much of a boost for top performers. AI may help the most skilled agents and consultants do their work a little faster, but the significant gains come from boosting the performance of the workers struggling most.
If this level of AI performance boost is possible in the workplace, why aren’t we discussing AI’s role in closing achievement gaps in education? Limiting AI access in schools isn’t hurting the students already doing well. It’s hurting the students who can’t keep up.
A common rebuttal is that using AI is “cheating.” Why should a student using AI to write a book report be entitled to the same grade as a student who does it entirely with biological brainpower? It’s the same reason you shouldn’t mind if I use AI to edit this article. Would this article be better if I was typing away on a typewriter?
Markets don’t care about the learning process. They care about the goods and services people produce. Sending students into the world without the ability to access the power of AI is borderline malpractice. The privileged children will be OK. They have access to capital and resources. It’s everybody else who suffers.
Diverse Disruption
I’ve mentioned markets a few times. What do markets have to do with education? We’re talking about children, not demand curves.
There are reasons for education that have nothing to do with markets. We want to live in a functioning democracy and experience the joy of a thriving arts culture. However, if we’re being honest, the primary role of education is supplying human capital for the labor market.
Today, the supply of human capital is uneven. The probability of success isn’t randomly distributed. Your economic future has more to do with the zip code where you were born than your abilities. At this point, the American dream of social mobility is propped up more by anecdotes than data.
What if that wasn’t the case? What if the early results we see in the private sector also manifest in education? What if each student graduated with similar skills, even if students from disadvantaged backgrounds relied more on AI?
Disruption is coming whether we like it or not. Incumbents powered by human labor will give way to new entrants built on a foundation of machine labor. The question is whether we empower children from all backgrounds to participate in the disruption.
If you want proof that disruption can be evenly distributed, look at the creator economy. The people running YouTube channels are more diverse than traditional media companies. YouTube provides low-cost tools to anybody wanting to publish video content. AI does the same, but the potential applications extend far beyond a single industry.
Equality of Access
The traditional approach to closing the achievement gap has been increasing access to high-quality instruction and resources. That’s an admirable goal, but I’m not sure it solves the underlying problem.
The first issue is that traditional models aren’t scalable. New schools are expensive to build and operate. Demand is highest where resources are most scarce. Manual processes become highly variable and more entrenched with scale.
The second issue is that equal access is not equal opportunity. Suppose your brain is hardwired during your formative years to deal with physical, emotional, and psychological trauma. It’s unreasonable to expect you to achieve the same results because you have the same resources. There are limits to neuroplasticity, and we can’t expect students to jump on a moving train without a running start.
Fortunately, AI scales better. Creating models is expensive, but the marginal cost of deploying them is small. For example, I am a heavy user of OpenAI’s GPT-4 platform. I use it daily to turn research papers into podcasts and help me code. My average monthly bill for OpenAI is about $100. That may sound like a lot of money, but there’s no way a human would do what I need for even five times the pay.
I’m not advocating for every kindergartner to trade their ABCs for 0s and 1s. The fundamentals of what we teach children need not change. There are opportunities to weave AI into existing curricula. For example, socialization is a critical part of kindergarten. Introducing five-year-olds to “AI friends” who can answer questions about dinosaurs could aid in socialization, especially given advancements in AI empathy.
There are promising signs that powerful AI models will become available to everybody. I didn’t have “Facebook…I mean Meta…open-sourcing their large language model” on my bingo card. That said, models are only one piece of the accessibility puzzle.
Here are three other areas where the outlook is less rosy:
- Broadband: Nearly every American has at least one internet service provider. That’s a positive first step toward ubiquitous AI access. More concerning is how few people in the lower income brackets have a broadband connection at home. Just over half of households earning less than $30,000 annually have one.
- Hardware: According to the U.S. Census, most American households have computers. Unfortunately, that “computer” is often a smartphone, especially for households earning less than $25,000 annually. Smartphones are primarily consumption devices for AI. If you want to create something with AI, you need a computer or tablet.
- Language: It’s difficult to overstate the importance of learning to program. You can consume AI content without knowing how to code. However, to “employ” AI workers, you must know how to speak to them. According to a 2019 study, less than half of American high schools teach computer science. In contrast, over 90% of high schools teach foreign languages. I feel strongly that we’re teaching our children the wrong languages for the future.
These barriers to access may seem daunting, but technology gets better, faster, and cheaper each year. We’re not talking about providing every student with a gigabit broadband connection, a $3,000 laptop, and $1,500 annually to cover API fees. Students need a 100-megabit broadband connection, a cheap laptop, and a guide to where to find open-source models. There’s no reason universal AI access should cost more than $1,000 per child annually. That’s about $50 billion annually, a portion of which is already included in our education budgets. For reference, annual spending on K-12 education in America is nearly $800 billion.
I recognize that I’m oversimplifying the problem. This isn’t a matter of providing access and hoping for the best. Schools and teachers will be instrumental in translating access into outcomes. However, let me provide one last illustration of why access matters.
The decision to market the personal computer as a gaming platform aimed at boys has shaped the field of computer science and the technology industry to this day. A seemingly trivial decision four decades ago produced a drain on our collective potential. We can’t afford to make the same mistake. AI is not only for boys. AI is not only for the wealthy. AI is not only for the tech-savvy. AI must be for everyone.
Speed > Perfection
I cringe when I hear techno-optimists insist that all technological progress is good. The e/acc crowd is even worse. The “move fast and break things” mantra is catchy until you realize that sometimes the things you break are people’s lives.
I know there will be unintended consequences from deploying AI in the classroom, and we’ll make mistakes. Unfortunately, technology doesn’t progress on a linear path. Computing power, data, and other inputs are growing exponentially. We don’t have decades to study the problem.
Furthermore, the consequences of inaction will probably be worse than the alternative. Restricting AI access primarily hurts children trying to move up the socioeconomic ladder. If the only place children can access AI is outside of school, who stands to benefit? It’s the people who can afford to provide access outside the classroom.
Fortunately, we already have experts in place to guide the deployment of AI in the classroom. Teachers know their students and are best positioned to assess what’s working and what’s not. Any push to introduce AI into schools can’t come at the expense of teacher primacy.
AI is not the modern equivalent of the calculator. It’s an extension of human intelligence. Students need to maximize total intelligence — human and machine. That requires a K-12 education system that looks quite different from today.
This is a scary time. The question is whether you believe the intrinsic potential of every student is equal. If so, we must do everything possible to close the achievement gap using AI. I’m not confident it’ll work, but we have to try.
