Happy Friday!
What a week for AI and parenting. I was very much coming here with the plan to write about the AI companion elements of the government’s latest press release following on from the social media ban announcement in June, but then Ofcom dropped an investigation into TikTok yesterday – so I’m going to dive a bit into that, because it has a very heavy reliance on AI, and that’s the bit that’s caused the investigation. Later on in the newsletter I explain a bit about how that AI works and why it’s triggered the investigation.
This weekend’s screen-free AI activity
Fancy having something on hand as a response to those never-ending “I’m booooorrrrreeeddd” calls from the kids? This one is particularly good because it’s also delivered in lots of short bursts, so hopefully they don’t lose interest.
Let me introduce you to the Boredom Buster Jar! AI generates a bunch of age-appropriate tasks to keep your kids busy and each time they want a challenge they stick their hand into the jar and pull a new one out. And the best bit is that if you have more than one similar-aged child then you can make them competitive – “who can make a paper aeroplane that flies the furthest”.
There’s a step-by-step guide over on our blog – with a handy pdf download.
And as a thank you for subscribing you can access the entire screen-free summer activity series already.
What’s going on with TikTok
TikTok has been using an AI tool to decide how old users are, known as Age Inference. When your child first signs up for TikTok they are asked for their age, but as you’d expect, not every child is entirely truthful. So even if they give themselves an adult date of birth their usage over time decides whether they should be given the “child” label.
To do that, the system looks at what videos your child is watching, which accounts they follow and interact with, the language they use in comments and captions, and the content in their own profile to decide whether they are a child.
Once an account is labelled as a child, that label sticks unless the account holder appeals with a human moderator using proof of age, such as ID. If your child’s account goes undetected as a child, then nothing similar happens, they can just continue using the platform as an adult – because no child who has lied about their age on signup is going to voluntarily request the “child” label on their account.
Ofcom’s investigation has raised two issues with this. The first being that the system needs time to label your child’s account. So in the time before they are considered a child they are not being treated as such. And secondly, and in my opinion more importantly, if your child is watching more mature videos or is particularly advanced with language, they may be labelled as an adult account.
Behind all this is an AI categorisation algorithm that will have been trained on data that is known to be a child or an adult and the system looks whether the patterns of behaviour in your child’s account best fits with the known child or adult accounts. This data comes from accounts that have voluntarily identified themselves as children – which arguably could be a very different personality to the children who are lying. And, as I never stop banging on about, AI is just looking for patterns in the data it has, so if the training data contains no data similar to your child’s behaviour then they may well not be detected.
On top of this, AI gets things wrong. AI algorithms are probabilistic, which means the algorithm will output something like “63% confidence that it’s a child”. Then it’s up to the people who have designed this feature in TikTok to decide which percentage constitutes a child account. I talk about how that threshold is decided in a deep dive into how the AI behind age inference works.
But the thing to take away from this is that AI algorithms are not black and white, they are all shades of grey, and even if it’s a tiny percentage that are wrong we’re talking about the risks on children here, and these are real people, not numbers. In my opinion even just one child being shown harmful content is not good enough. This is the bit that has triggered the Ofcom investigation – they are concerned that it’s not a tiny percentage, and that TikTok is failing to identify child accounts using this method.
Over to you
Ooof so after that heavy newsletter topic – sorry about that, but it’s important – it’s time to wrap up. You may have spotted my little news videos on social, that’s a new thing I’m trying out, so I’d love your thoughts. The idea is that I flag anything relevant in the news you may have missed, but also then give a little extra “AI expert” sprinkle on top.
As always if you have any questions, please do send them over and I’ll do my best to get you an answer.
We’re on Instagram, Facebook, TikTok and YouTube.
Have a great weekend,
Fiona