By Fiona Morgan

It’s 2 am and everyone in the house is fast asleep except your four-year-old, who is having yet another nightmare. It’s the third night in a row and the parents’ WhatsApp group didn’t have any helpful advice last night, so this time you turn to ChatGPT. Everyone has done it. In fact, many other parents will be doing the same thing at 2 am, right alongside you. And there’s nothing wrong with asking Artificial Intelligence (AI) questions about your children, if you do it the right way.

I’m going to talk you through the five things that I think parents should know about using AI.

1. Write your question in a neutral way

AI companies improve chatbot responses by having humans rate them, and they feed their ratings back into the AI so it gets better at giving higher-rated answers. This is called “reinforcement learning.” 

This might sound like I’m heading into a technical explanation, but reinforcement learning actually explains something about AI that all parents should be aware of: the agreeableness of AI.

Turns out that the real humans rating the AI answers gave higher ratings to answers that aligned with their opinions, so the AI learned that agreeing is better than pushing back. So when asking questions, try to remove your feelings and any suggestion of what you think you should do, and you’ll get a more neutral answer.

For example, try “what is this rash?” instead of “I’m really worried about this rash. Do I need to take my daughter to the ER?” or “This is just heat rash, right?”

2. Be specific about what you’re looking for 

The information that the AI “knows,” or can access, comes from its training data and the internet. It is well-reported that this data doesn’t represent all cultures and languages equally, so AI will often default to the information it has most of, which tends to be American.

If you are looking for an answer specific to your own culture or situation, then include that in your question. Specify that you’re interested in vaccinations given in Thailand or details on the UK school curriculum, and you’ll get a far more targeted and useful response instead of something built from assumptions.

You can even specify the output structure you want. So if you’re using the response to write an email and want some bullet points, just ask for them in your question. 

3. Don’t put private information into AI

In point number two, I said to be more specific, but don’t be too specific. Unless you change the settings, some AI tools use the messages you type to train future AI models. A few years ago, we saw no real issues with uploading our children’s pictures to Facebook, but these days it’s considered a significant privacy risk. So it’s best to be cautious.

It’s difficult to know exactly what kinds of risks there are, or will be, if AI knows all the details about your family. But what I do know is that most of the time your question can be worded in a way that shields the most private information. There’s never any need to use your child’s name or school information; locations can be vague, medical details anonymous.

If you’ve got an AI account that you are logged into, you may also find that it has a memory. So it can very easily start to piece together information about your family and make assumptions. ChatGPT has a feature called “Temporary Chat” and Claude has “Incognito Mode.” These don’t save anything to the account memory, so it’s best to use them if you’re using the tools casually.

4. Always review the source

As well as being skewed toward particular cultures, AI’s source material includes online forums such as Mumsnet and Reddit. There’s nothing wrong with these websites—I visit them myself all the time—but literally anyone can post. I could write a long piece about how crispy pork is healthier than Som Tam and AI could pick this up and present it as truth. It is, right?

AI is better than it used to be at showing you the source of the information, but if it doesn’t do this, then ask for it. Adding “include website links for the source of all information” to your message will really help. Oh, and read them: I often find that the “sources” AI presents are broken links or misrepresented.

So now you’ve got a beautifully written message to send to AI. It includes no identifiable information, it’s got a neutral tone, you’ve been clear about wanting a response relevant to being an immigrant in Bangkok, and you’ve asked for sources. Time to send it.

And this leads me to my final point. The most important one.

5. Critically evaluate the answer and think for yourself

AI does not “understand” what it says. It doesn’t care, nor does it not care; it’s not a human. So if you get poor advice from AI, no one will be held responsible. And I really can’t stress this enough. Humans have morals guiding the advice they give. Computers don’t. And sometimes AI makes stuff up.

Read the response and take the time to think about it for yourself. You can try asking another AI tool and comparing the answers. You can get a second opinion from someone you trust. But please don’t just follow the response blindly. 

You and the other adults in your child’s life are the ones best placed to make decisions about what is right, even if it goes against what AI suggests.

About the author

Fiona Morgan has spent over ten years working in AI, advising businesses on how they use it, and taught secondary school math before that. She co-hosts “Raising Generation AI,” a podcast for parents, and writes a weekly email explaining current AI news stories and what they mean for families. Visit raisinggenerationai.com