AI, Education and Inequality

The debate on AI in schools is currently stuck in a binary trap.

One side wants to ban it to "preserve real learning". The other wants to surrender to it because "it's the future."

The problem is that both sides are right about what they fear, but its a false binary choice.

The "ban it" camp is right that indiscriminate use of AI degrades learning. Science has already shown that. If you outsource thinking to a machine before you've built the mental model, you get a hollow skill set. You get the "calculator dependency" nightmare scenario.

But the "adopt it" camp is right that ignoring AI is economic suicide. If you don't learn to use the leverage, you are competing with a shovel against a tractor.

The question isn't whether to use the tool. The question is how to handle a world where performance is now split into two distinct modes.

The calculator precedent

We have been here before.

When calculators arrived, math didn't end. We didn't stop teaching arithmetic, but we also didn't ban calculators.

Instead, we implicitly split the domain:

  • Understanding: Can you do the math with just your brain?
  • Leverage: Can you use the machine to solve hard problems fast?

We accepted that these are different skills. You need the raw mental model to know what to punch into the tool, but you need the tool to do it at scale.

The mistake is to think schools have to choose between these two. Education's job is not to pick one side of the tradeoff. It is to distribute both the skill and the leverage.

The floor vs. the ceiling

There is a popular narrative that AI is an equalizer because it raises the floor. The novice coder can now ship a feature. The bad writer can produce a readable email.

This is true, but it misses the main event.

While the floor rises linearly, the ceiling rises exponentially.

  • Give a calculator to someone who doesn't understand math, and they stop making basic arithmetic errors. That's a floor-raising effect.
  • Give it to a mathematician, and they build bridges. That's a leverage effect.

The tool is multiplicative, not additive.

We see this in coding already. AI helps a junior engineer get unstuck. But it turns a senior engineer into a department. The second one is MUCH more significant for the economy at large.

If you have high skill and high leverage, you pull away from the pack at a velocity that pure effort can never catch. Economic inequality comes from the extreme leverage at the top, not the lift at the bottom. The superstars who capture outsized gains are the winners, not the "mom and pop" shops that just get a little more efficient.

The race is on

This dynamic is what Goldin and Katz described in their 2008 book, The Race Between Education and Technology.

Their thesis was simple: inequality is the result of technology outrunning education. When technology creates new leverage, the people who master it first capture the gains. Education's job is to catch up, to distribute that mastery until the premium for having the skill drops.

The amount of inequality you get has two important multipliers:

  1. Speed of diffusion: How fast is the tech spreading?
  2. Generality: How many fields does it touch?

AI scores high on both. Today's inequality is a result of a very valuable but not very distributed skill: software. AI will be MUCH worse. Previous waves have been more specialized and slower to distribute, but AI will probably be neither.

If technology accelerates while education stands still (or bans the tool), the race is over. We get a caste system: a small group of super-productive, AI-leveraged elites, and a massive underclass that effectively cannot compete. Today's inequality on steroids.

We are seeing this now with software. A very valuable form of leverage exists, but it is still unevenly distributed, and the result is clear economic inequality. Those who can use software to make money flourish, those who can't flounder.

We have seen this before with reading and writing. Those too were once concentrated in a small elite, and the difference in power and quality of life was enormous. Mass schooling reduced that advantage by making literacy broadly distributed.

The same logic now applies to AI. When a new form of leverage appears, inequality depends on how quickly education turns it from an elite advantage into a mass capability.

The mandate

The job of schools today is not to pick a side. It is to build a system that can hold two contradictory truths at once.

We need to update our mental model of assessment to match the calculator precedent:

  • Test the raw mind: Can you think, write, and reason without the machine? (Protecting cognition)
  • Test the leveraged mind: Can you use the machine to produce high-quality work at scale? (Distributing power)

These are not opposing goals. They are the left and right hand of modern capability.

If we fail to teach both, we aren't protecting students. We are engineering an economic underclass.