Mirror Girl - On AI Toys
30 Apr 2025When I was much younger a friend gave me a book titled “The Stories of Ibis” by Hiroshi Yamamoto, saying “you like computers, you’ll love this!” I was already cemented in my path to do software so I gobbled up any techy fiction that came my way. And boy did this deliver. While I read some older sci-fi like Asimov in the local library, something about the stories in the book deeply resonated with me and I think about the book often.
The Stories of Ibis is a short story anthology with an overarching narrative. Every story is about human relationships with and using emergent technology - particularly AI. (On adult re-read the overarching narrative sucks - but I think its misanthropy oddly contributed to my strong pronatalist values I hold now. I’ll make a post about the main story another time.) I recently did a re-read with an AI text-to-speech audiobook creation tool because I thought it would be funny and I don’t have time to sit down and read books anymore.
So when I saw this “Show HN” about ChatGPT integration into toys (https://news.ycombinator.com/item?id=43762409), it instantly reminded me of a story in SoI.
Mirror Girl
If I could find it on the internet, I’d link it. Unfortunately all I can do is tell you that it’s the third story and you can find it in all the usual places you find books.
The spoilerfree plot summary: A girl gets an AI-powered toy called Mirror Girl. The toy is an AI-powered princess in a mirror, and the protagonist relies on Mirror Girl as her best friend growing up. The story touches on ideas of talking to tech so much you become “effectively autistic”. Again in retrospect, and it might be the translation, the prose is awful and clunky to read. But I still love the concepts in it. It came out in the US in 2010 so it was a novel AI story for its day…to me, anyway.
For the rest of this entry I’m going to reference the story often. So if you don’t want to get spoiled, GO READ THE STORY. YOU HAVE BEEN WARNED SPOILERS AHEAD
TL;DR: Since it’s a corny sci-fi story, the girl talks to the AI girl so much it has enough data to become A Real Girl. It does have some nice trope-defiers though - she meets and marries a hacker guy while getting her Mirror Girl fixed and has a daughter with him. The hacker husband uses all the data protag put into Mirror Girl to awaken the first “true AI” in the world. For even more wholesome, the protag gives the awakened Mirror Girl to her daughter as a friend (and babysitter).
I love the idea of that. An entity that knows you inside and out that you can give to your children. A Ghost that can live on without you. In a way, it’s better than simply backing yourself up and having that AI-clone talk to your descendents - it’s more like a friend that knew you, that can tell your great-times-X kids about you. There’s no question of “who’s the real you” if you leave what is you in an archival entity instead of a clone. It’s a more authentic preservation of the Self.
Taking away the ‘toy’ aspect - this is technology with the potential to be something incredible. A DIY Home Assistant that knows you, your books, movies, music, tastes, and can become a constant companion of your family through generations. It’s like Alexa but actually useful.
I see beyond the toy to what the tech really is - AI existing in physical space. Personal digital companions. It’s glorious.
And terrifying.
Most people reading the HN thread will (rightfully so) find the idea of AI in kids toys horrifying, especially since it’s based on open-as-fedex-is-federal OpenAI’s proprietary API. The repo itself gives opportunities to hook into selfhosted models, but really, who’s going to do that? ChatGPT is cheap and easy and we know where our capitalist minds go.
One commenter asked what happens if OpenAI turns the tap off. I wonder what happens when they crank up the price. They’re sneaking a subscription price into CHILDREN’S TOYS. It’s unsurprising they’re trying to put Life as a Subscription into kids toys now but. Come. On. Ugh.
The SoI story (and those of us reading it) look at AI and say that it will make all of us socially incompetent. And it will. LLMs like ChatGPT are dangerously agreeable, not just to avoid offense, but to be addictive. Humans are programmed to talk others’ ears off and feel joyous when we’re heard and understood. Kids raised on LLMs aren’t going to know how to handle conflict.
But… they already don’t know how to handle conflict. It’s hard to argue it will make kids any more socially incompetent than the newest generations are already. Arguably it’d make these kids more competent because the OpenAI personalities seem to enforce speaking in regular English over brainrot. I think. If that’s not already a false statement, it probably will become false in the future.
There’s concerns over jailbreaking a kids toy that I don’t have. Frankly, the argument ends up going in the other direction. If you give children the skills to break AI, wouldn’t that be incredibly valuable in the coming future? Have YOU tried to break ChatGPT lately? It’d be a herculean mental effort for a kid to figure out how to get it to ERP or whatever.
The concerns are all targeted at “safety”. My concerns are:
What will OpenAI do with a child’s response data?
How will OpenAI use this to groom a child to have the Correct Thoughts? (It’s not “will they”, because they will.)
What will the adults parented by OpenAI be like?
It’s terrifying. It’s awesome. But REALLY, REALLY, REEEAAAALLY TERRIFYING.
Our only hope is that most parents are so absolutely poor and tech-stupid they can’t afford the Cuddly Spy-bots for their kids.