I think its product has a profound democratizing effect. In theory, a kid sitting in a provincial town in rural Brazil should be able to receive the same responsive interaction with the Efekta AI teacher as someone living in Mayfair.
Is anything lost by the introduction of AI to the classroom? Will we end up with a generation of students who use chatbots as a crutch—to draft essays, solve problems, and so on?
They’ll do that, anyway. Trying to shut out AI from schools is senseless. It’s about how you incorporate AI into education. Bad teachers will use it badly, and good teachers will use it very well—as they did whiteboards and calculators.
But we’re talking about a more fundamental change. I’m asking what it might mean for students not to develop foundational skills.
If you go back to the time when calculators were invented, [people thought that] kids are never going to be able to do mental arithmetic. But that didn’t turn out to be the case. It will have an effect, of course. But I think the net effect should be positive in terms of educational performance.
Children are probably uniquely vulnerable to the kinds of dangers associated with chatbots. How do you think about those risks?
Of course there are perils—particularly, vulnerable adults and children becoming emotionally dependent and invested in a relationship with something that has an avatar, humanoid presence in their lives.
At a societal level, we should take a very precautionary approach. I think you should have clear age-gating on how agentic AIs are made available to young people.
Like Australia’s social media ban for under-16s?
There’s no point in having a ban if you can’t measure people’s age. That’s where policymakers rush to catch headlines about bans and don’t quite think through the quite-difficult stuff. Unless you want all these platforms to, what, hold everyone’s passport details? My view for a long time has been that the only way to do that is through the choke points of iOS and Android, at an [app store] level.
But in principle, I think you should take a similarly precautionary approach. The susceptibility to becoming highly emotionally invested in and perhaps unduly influenced by your relationship with a kind, patient, 24-hour voice who’s listening to you all the time is a very real one.
I don’t think it’s a risk at all with the kind of products that Efekta produces, though.
Even though the AI is literally assuming the role of the teacher?
Well, no—because it is not. These agentic AIs produced by companies like Efekta are not going to have some sort of surreptitious midnight relationship where they say all sorts of ghastly things to a pupil. It’s a teacher-controlled experience.
You spent almost seven years at Meta. In that time, AI became the frontier technology. I’m curious how your experience at Meta colored your perspective on the opportunities, the risks, and limits of AI—and the quest for superintelligence.
If you ask three people at the same organization what superintelligence is, you’ll get three different answers. I get the impression that everyone in Silicon Valley has to say they’re within touching distance of artificial general intelligence or superintelligence, because that’s the way to attract the best data scientists. I find it difficult to grapple with a concept as hand-wavy as that.
