Singapore's AI push needs a learning strategy, not just a productivity play

Singapore is betting big on AI as a national productivity engine, but a leading learning scientist warns that a narrow focus on output gains risks building a workforce fluent in using AI tools without truly understanding them. Manu Kapur, Director of the Singapore-ETH Centre and author of Productive Failure, argues that the very thing making AI attractive — instant results — could be undermining the deeper learning that Singapore's economy will need to stay competitive.

Kapur's central point is that AI adoption and capability building are two different things. He cites a study of 26,000 Chinese students over 30 months, where AI improved homework performance and reduced completion times but actually hurt exam scores — suggesting the tool masked rather than closed skill gaps. The same dynamic plays out in workplaces: a junior lawyer using an AI copilot to draft contracts looks productive, but without practising the craft, they never learn why an indemnity clause matters or when a confidentiality carve-out carries risk. The senior partner still has to carry the full burden of accountability.

To fix this, Kapur proposes a two-track adoption model for businesses. Track A — the performance track — deploys AI where speed and consistency matter, with proper governance. Track B — the learning track — deliberately creates settings where employees cannot outsource thinking: experimenting, critiquing AI outputs, running post-mortems, and building internal playbooks. Companies that run only Track A will get quick wins but hit a ceiling when the environment changes. Those that embrace both will build organisations that keep adapting. The warning extends to layoffs: cutting headcount after AI deployment creates "capability debt," leaving firms with fewer people who deeply understand the workflow, exceptions, and customers.

Why it matters for Singapore: Singapore's National AI Council is well positioned to coordinate strategy across sectors, but adoption rates and near-term productivity gains are a misleading scorecard. The real measure of success is whether Singapore builds a workforce that can critically evaluate, challenge, and improve AI outputs — not just generate them faster. As Kapur puts it, treating AI only as a productivity machine risks scaling "unproductive success" — impressive output today that leads to brittle, unadaptive capability tomorrow. For a small, high-cost economy that competes on talent and trust, that distinction matters enormously.