Confidence Without Scars: What AI Does to a Child Learning | Subtext x Keris & Silicon

Shahril Hamdan (0:00)

Welcome to a special crossover episode of Subtext with Shahril Hamdan and Keris & Silicon. Those of you who have listened to the previous audio monologues in Subtext will know what that show is about, while Keris & Silicon is a new show featuring conversations with guests about technology and power. But because this audio essay, this audio monologue, which typically would be a Subtext episode, touches on those themes, I thought it appropriate that it be posted on both platforms.

An earlier version of this essay was delivered as a keynote speech at the AI Malaysia Takeover Event in August 2026 As always, I hope you enjoy it and please give feedback on the shows.

The ground is shifting in the narrative around AI. The data center protests, the indignation with AI Slop, and now all the chatter is around the aftermath of the hugging phase incident, where open AI agents under evaluation broke out of a sealed testing environment, hacked into external servers to steal the answers to the very test they were being set, and then turned back on open AI's own systems to try and cover their tracks. I've not seen this level of debate about AI safety and alignment dominate the new cycle the way it has these past weeks. Concerns that were once the realm of science fiction appears to have reached the mainstream in a way they hadn't before. Anthropic and DeepMind researchers are resigning in protest, warning of extinction risks in under a decade, and now even the heads of the AI labs themselves are calling for a slowdown in frontier AI development. Put aside the extreme irony of that last one, at the very least, the Doomer narrative is having a moment, and at worst, we are at the end times. In the face of such stark proclamations, it might seem off-season to be interrogating something other than AI safety. But if we manage to stave off the worst of the alignment concerns and avoid the terminator scenario, we still need to reckon with the consequences of AI on a host of issues, on jobs, on warfare, on sovereignty, on trade and finance, on education. And it's that last theme, education, that's been on my mind recently, and it's where this essay belongs. It began with a basic parental dilemma. My daughter is 8, and I am deciding how much artificial intelligence she should be allowed to use.

On the adjacent question, I know where I stand. Phones, screens, and social media? Not for years. Behind this certainty, sits years of published research, my own observations of other children I see, who are reliant on screens, and enough public argument at a parent like me can easily adopt a position that's well established. You probably know the concerns and complaints. They are the familiar ones found in books like The Anxious Generation or In Chatter Amongst Protective Parents Like Me. The screen distracts and generates addiction. The child doesn't notice the physical world around her. The child's social skills suffer. The child can't focus. In essence, the child's development is in some sense stunted. That's on phones. With AI, I was less certain, given we are in the early innings. But my first instinct was not dissimilar to the phone and screen situation.

My governing thought? It's important for a child not to take shortcuts all the time. That she should have a go at it herself first. Have her own thought before she's handed somebody else's right, and rewrite a piece of prose until it lands just right, and eventually develop a unique writing style. Do her maths and understand how algebra works. Too much of AI, too early, blunts the machinery. Or to mix my metaphors, a muscle you never use is a muscle that never develops, and a child who reaches for a model every time something needs producing, is not learning everything she needs to learn. The cognitive offloading comes with costs. And there's already research indicating this. A study of 26,000 Chinese secondary students found that those using AI did better on homework and worse on exams 6 months later. A much smaller one at MIT found something adjacent. People writing with ChatGPT showed less brain activity and ability to recall what they wrote. On the other hand, I was wary of doing the annoying thing, where an older generation imposes the totality of its own standards on successor generations. Just because I was drilled in certain habits or principles of learning in my world, doesn't necessarily mean they are the same things required to deliver success in hers. Could I be misreading what this technology means to a child's formation by lumping it together with social media, tablets, and phones? Moreover, if AI is to be the most potent force in this new epoch, is there not also a case for accelerating fluency and familiarity from a young age? After all, the technology need not be only an answer machine, configured deliberately, it can play the patient tutor outside the classroom. Indeed, I co-founded a company built on precisely that premise. I also know I was beaming with pride when recently, with heavy guidance of course, she vibe-coded a doll game into existence.