Inadmissible Evidence

A child’s absorption in a learning app tells you nothing about whether they’re learning.

Hugh Shepherd, who runs the learning app Liaura, “a private place where children, families and trusted friends share moments, play, learn and do good things together” has a complaint about the UK government’s proposed ban on under-16s using social media.

He is one of a number of edtech founders who are worried that their well-intentioned learning apps will be inadvertently caught up in the ban.

He told The Times last week that ministers “tend to bundle in the misuse of a technology feature with the feature itself.” He argues that his app uses the same tools as the social platforms but to better ends.

In itself, it’s a fair point and I had some sympathy with it until the defence undercut itself in the next breath:

“Depending on the severity, it could be existential. Things like push notifications, for example … if we were restricted from using them altogether, I think that could be deadly, actually. When you’re building a technology platform, particularly when you’re trying to get off the ground, habit formation is an essential part of it.”

There is a serious debate to be had about whether a state should ban children from social media. The proposed UK ban is driven, in part, by a worry about addiction, the sense that these platforms are designed to be compulsive and that children can’t defend themselves against that. Which is what makes Shepherd’s defence a little strange. Asked why his app should be spared, he points to its habit-forming design; he defends himself with the very feature the ban is built to catch.

But he can hardly be blamed for the reflex. Education has spent years treating engagement as a good in its own right. It became a proxy for learning so long ago that we stopped noticing it was ever a proxy at all and, once a metric is that firmly established, the only arguments left are about who gets to claim it. Which is exactly what’s happened here.

In the education context, the debate has become a question about where to draw the line. When is a social media app not a social media app? The government promises a “narrow” set of exemptions - music streaming, ecommerce, education. Understandably, the edtech founders worry about which side of the line they will fall on and argue that they need clarity in order to plan.

That is a real argument and the anxiety underneath it is real. But it is entirely a boundary dispute and, in a boundary dispute, protecting your territory tends to become your sole aim. Everyone’s arguing which apps are the good ones without asking what makes an app good.

This boundary dispute has a premise hidden inside it. To argue that education apps deserve an exemption, you have to believe that an education app educates. You have to believe that the worthy-purpose apps are worthy in their effects and not merely in their intentions.

The exemption for ‘education’ is only coherent if a learning app, by being a learning app, delivers learning.

The unexamined premise driving this entire argument is that absorption equals benefit; a child absorbed in a virtuous app is a child being served. The absorption is the evidence and that is clear in Hugh Shepherd’s assertion that ‘habit formation’ is ‘essential’.

Nobody checks whether any learning has taken place. The engagement is allowed to stand as its own proof.

You might ask why this is a problem. After all, if children are flocking to an app, spending hours on it and coming back day after day, doesn’t that tell us something? We trust engagement elsewhere. If you’re in a new town and looking for somewhere to eat, it’s a good rule of thumb to go to the busiest restaurant. A book you ‘can’t put down’ is generally accepted to be a good book.

Well, it matters because the features that keep a child returning to an app are designed to do exactly that. Push notifications, streaks, rewards are all skilfully and deliberately built to compel and the engineering is the same wherever you find it.

The loop that pulls a child back to a phonics game is the identical loop that pulls an adult back to a slot machine. It exploits the same reflexes and the same psychological levers.

And that mechanism has no knowledge of what it is attached to. It doesn’t behave one way when the purpose is noble and another way when the purpose is cynical. It cannot tell that Hugh Shepherd has good intentions. The machinery isn’t changed by the worthy purpose.

So Shepherd is right in one sense when he says the tool is neutral - it takes no side in what it’s used for, but it was built to compel regardless. So, he is wrong to think that it serves his cause. I have no doubt that his sincerity is real but it lives in his mission statement, in his head as the reason he built the app. It doesn’t travel down the line to the precisely timed push notification on the child’s phone, driving them back to the app. The good intention shapes why the tool was made. It has no effect on what it does once it has been made.

The loop does not know it is working for the good guys.

Again, you may well ask why this matters? Shouldn’t we judge the outcome? If your child gets hooked on a phonics app and comes out reading fluently, why would you care about the engineering that got them there?

Quite so. That is completely correct.

We mustn’t fall into the trap of thinking that engineered engagement is bad because it’s engineered. It might sit on top of something that is genuinely good. And that leads me to the point I want to make: if the mechanism is agnostic then the level of engagement cannot serve as evidence of worth.

We trust engagement normally because it is earned. A book grips us because it is good. The restaurant fills because the food is worth it. There, the level of engagement points directly to the quality of the thing we are interested in.

When the engagement has been deliberately engineered then it breaks that direct link. If a child can be hooked equally on an app that is good and an app that is useless, then the level of absorption tells us nothing. Quite simply, it is inadmissible as evidence.

So, Hugh Shepherd’s habit formation may well co-exist with real learning. But it cannot be the evidence of it.

To judge the value of a learning app, we have to look elsewhere.

And that means looking at the thing engineered engagement is designed to remove. Every streak, every notification, every frictionless loop exists to make the experience smoother, easier, harder to put down.

But learning has never been smooth. The moments that teach us most are usually the ones we’d rather skip - the effort, the difficulty, the point at which it stops being fun.

Which raises a possibility I’ll return to next week: that in engineering away the friction, we may have engineered away the learning.