Let's keep artificial intelligence inside prisons and out of classrooms: Here's why

You don’t have to be a Luddite to see the problem: AI’s inherent promise is convenience, while learning requires effort. Those aims are fundamentally at odds. The technology does not belong in elementary school classrooms, and the later students encounter it, the better. A more effective place to focus would be teaching AI skills to adults with immediate vocational needs, such as prison inmates.

Last week, Singapore’s education minister made headlines by saying the city-state would introduce AI in the fourth year of primary school, although “under close supervision and with low exposure.” In Beijing, schools have already begun offering AI courses to primary and secondary students. But the push is colliding with reality. In South Korea, an AI learning plan was rolled back after just four months amid backlash from educators, parents and students.

This is a global debate. But since education has been central to Asia’s economic rise, the stakes are especially high. The region’s countries regularly dominate the OECD’s PISA rankings, which measure the performance of 15-year-olds in mathematics, reading and science. Singapore topped the latest round—even before introducing 10-year-olds to AI. The strength of Asia’s education systems comes from rigour and repetition, not removing friction.

Where academic pressures run high, so does the money and parental anxiety. Prior to Beijing’s 2021 crackdown on private tutoring, a Stanford study found that frugal Chinese households spent an average of more than 17% of their annual income on education. And in Singapore, the latest government data shows spending on education continues to climb.

It makes some of the broader regional so-called edtech initiatives look less like a public good than a lucrative captive market. Combined with government pressure to appear progressive, the rush by tech companies into schools deserves scrutiny.

This also distorts the definition of what success for these programmes would look like. Even if AI lifts short-term test performance, that may say little about whether students are actually learning.

One study found that those who used ChatGPT retained significantly less than those who relied on traditional methods such as study groups. On a surprise retention test 45 days later, ChatGPT users “scored significantly lower.” The researchers found that while AI assistance may ease initial study, “it appears to undermine the effortful process needed for robust learning.”

In other words, friction, repetition and struggle are how people learn. Removing too much of that may reshape how minds develop in ways we do not yet understand—even in adults. More long-term research is needed. But that uncertainty alone should be reason enough for caution before these tools are rolled out to children whose brains are still developing. The money pouring into pilot programmes might be better spent hiring human teachers.

This doesn’t mean AI has no sensible place in education. One Singapore initiative plans to offer AI literacy courses to elderly prison inmates. There, the students are adults and the goal is practical: equipping incarcerated people with skills that could help them find work and reenter society.

These key distinctions make highlight the real-world difference between pupils learning career-ready technical skills versus efforts to force the nascent technology into primary school classrooms. And a small US study found that expanding access to inmates’ digital education programmes was associated with lower recidivism.

The fears driving much of this agenda are easy to understand: children without AI knowledge may be left behind in tomorrow’s labour market. But the reality is that as automation spreads, skills like critical thinking, communication and emotional intelligence become more valuable. These cannot be instilled via a machine.

Policymakers are racing into the next classroom fad before reckoning with the damage from the last one. After years of proliferating screens and educational technology, a growing body of data indicates Gen Z has become the first generation less cognitively capable than the previous one. It’s hard to see how AI, which will only bind students more tightly to digital devices, will improve that trajectory.

Education exists to force the kind of mental effort AI’s optimization is designed to remove. AI training may make sense in prisons. But it has no place in primary schools. ©Bloomberg

The author is a Bloomberg Opinion columnist covering Asia tech.