Go ahead, adopt AI for education, but sharpen the critical capacity of students too

Its rise has already shaken India’s software sector, which needs full R&D mastery over it to make a proper AI pivot, even if adaptation is the urgency right now.

As this year’s Nobel award for economics reminds us, innovation-led growth is not costless. What’s creative can also be destructive, and we must do all we humanly can to ensure that its gains outweigh its costs.

AI uptake puts jobs at risk, but, as a report by Niti Aayog says, while a quarter of our tech sector’s 8 million roles may vanish by 2031, some 4 million new ones could arise too. Payrolls could go both ways. What AI implies for us beyond a five-year horizon matters even more, which is why an AI plan of the education ministry should make us sit up.

As reported, it may place AI on next year’s syllabus of schools under the Central Board of Secondary Education, starting with pupils of Class 3. China has taken a lead in raising kids as AI natives, but at what pace should India be trying to catch up?

While academia must adapt to AI, students must duly be forearmed with the key faculty of critical thinking. Classes must foster a spirit of inquiry that’s ready to probe everything that is pitched as truth. If social media yielded a ‘post-truth’ world, AI could amplify fakery. Given the real-world data that AI trains on, this is not a trivial concern.

Stanford AI researcher James Zou recently flagged some “troubling" emergent behaviour seen in large language models (LLMs). “When LLMs compete for social media likes, they start making things up," he said on X, “When they compete for votes, they turn inflammatory/populist."

A paper that he co-authored with Stanford’s Batu El on how LLMs can get misaligned in a contest for audiences calls this the ‘Moloch’s Bargain’: one that involves everyone, even though we would all be better off if it involved nobody.

Their lab tests found that even when explicitly asked to stay truthful, LLMs that were optimized for competitive success got deceptive under pressure to deliver on sales, vote-share and social media engagement goals. Clearly, AI adopters need to identify and squash such ‘alignment’ risks. Our education system, especially so.

True, the ‘post-truth’ world has defenders who call it the digital twin of a ‘post-modern’ dismissal of absolute truth. However, in the far less abstract realm that most of us inhabit online, we need good old truth to prevail. As we prepare to enrol AI for education, we must keep students updated on its fallibility, test all LLMs deployed for integrity, and make the artificiality—and amorality—of these clever bots amply clear in class.

Critically, the market for AI tools must never go the winner-take-all way. The profit motive of private enterprise is far more likely to keep the outputs of AI chatbots—and actions of AI agents—aligned with the interests of users if multiple tools vie for our usage. For this, we must avoid the trap of ‘network effects’: If everyone starts using the same tool everyone else is using (or is asked to), rivalry levels could fall, a monopoly may arise and users might lose their say in how AI evolves.

In education, like elsewhere, we need a strict vigil kept on the role of AI. Even if it’s only a tail-risk, going wrong would be far too costly. Let’s pace our adoption accordingly.