Generative AI software is popping up everywhere, and now it’s even in teddy bears. Poe the AI Story Bear is the first talking plush toy that uses AI software to generate and read stories aloud to children. It’ll be available in stores in August for $50.
Los Angeles-based toymaker Skyrocket is launching Poe as part of PLAi (pronounced “play”), its AI-powered toy brand. Skyrocket has entered into a multi-year partnership with AI speech software company ElevenLabs to provide the voice of Poe and future PLAi toys. No two stories told by Poe are the same. The stories are generated using Microsoft Azure and OpenAI’s ChatGPT 4o chatbot and delivered on-demand with a kid-friendly app.
Check this out: Hands-on review of the first AI teddy bear: What my kids thought
Poe is part of a new wave of toys powered by artificial intelligence software. The trend comes as toymakers explore how AI can keep a toy looking new and potentially keep a child interested for longer. The app and voice for this teddy bear are still being finalized, but I got a chance to get an exclusive look at Poe at home and see what it’s like for my kids (and me) to use an AI program as a toy.
How Poe Creates a Story
Poe, a plush toy powered by four AA batteries, needs an app to generate new stories. The audio files created are sent to the bear via a Bluetooth connection, and his mouth moves to “speak” as the audio plays. But Poe doesn’t always need the app to read a story. You can save your favorite stories to the bear and play them back at any time by tapping his ears.
The app has a simple design. It guides the child through a series of images depicting characters, objects and a setting, allowing them to choose the ingredients of their story. The options go beyond what you would find in a typical fairy tale. Take for example the choice of characters: you can mix a story with a witch, a social media influencer, an alien warlord, a zombie and an archaeologist. The more things you add to the story, the stranger it becomes.
Children use an app to choose the ingredients that will make up their AI-generated story.
Behind the scenes, a story prompt is sent to ChatGpt 4o. The Skyrocket team has a number of behind-the-scenes safeguards in place to ensure a story doesn’t stray into controversial territory, like talking about murder. While this wasn’t present in my early test, the final app should also allow parents to block out scary themes or characters — and even be tailored to younger age groups. (Maybe no zombies for under-5s.)
The bear’s childish voice is also made by AI, which turns text into speech. All this happens within moments of pressing the “Make up a story” button on the app. I found that it generates a story in less than 30 seconds.
The company says Poe will be able to tell stories in more than 20 languages, and that ElevenLabs’ voices will resonate with listeners in each language.
Put limits on what you can create
The bear itself has no microphone or camera, and it isn’t connected to the internet. However, the security risk here is not knowing what words an AI chatbot will serve to an impressionable child. To allay these concerns, Skyrocket says a number of filters and safeguards have been added into the programming to make the story child-friendly.
Skyrocket CEO Nelo Lucich told me that a team is adjusting the protections it applies to story creation messages, and the company is listening to feedback from parents. During testing, for example, one story used the word cancer as a synonym for destructive. The company decided that could lead to uncomfortable conversations, and it put a block on the word.
There are also limitations on the length of a story, not only for a child’s attention span, but also for cost reasons. Generating AI stories requires a lot of server resources, and the longer the story, the more it costs Skyrocket. The stories I generated lasted about 2 or 3 minutes.
There will also be limits on the number of stories you can create with the app. Each Poe is expected to come with credits to create 75 stories. But after that, a parent may need to purchase additional credits to continue creating stories. Lucich said there’s talk of offering additional stories for free over time, so you may not have to pay for new content.
My 8-year-old daughter was surprised that the chatbot made the prince a villain in the scary story it generated.
What My Kids Thought of Poe
My kids are 8 and 5, and I didn’t need to explain generative AI to them before playing with Poe. I simply told them that the bear makes up stories based on things you ask him to include in the app.
After listening to a few stories, I could see that my kids sometimes had a hard time grasping everything that was going on, but they wanted to keep making up stories, eventually treating it like a game to see what they could make the bear say. (And putting everything they could into a story to be silly.)
For me, the stories it generates exist in a literary uncanny valley, with stories taking random turns to fit every parameter of the message. It was sometimes difficult to follow every line of a story when the vocabulary felt out of place in a children’s story, as if ChatGPT’s AI was overusing a thesaurus.
Take this moment from a spooky time-travel story we created about a princess, a prince, and a thief:
The once vibrant kingdom had fallen under a curse cast by the evil Prince Barron. His sorcery summoned dark specters that haunted the land, chilling the hearts of Allara’s people.
A frightening apparition materialized at her side, the vengeful spirit of Queen Avelina, Allara’s ancestor from centuries past. “The ghost hour draws near,” Avelina’s raspy voice warned. “The stars have aligned to right this ancient wrong.”
Before Allara could respond, the thief Kiros’s sinister chuckle echoed through the room. With a puff of smoke, he appeared holding Avelina’s Chronosphere, a mystical orb used for time travel.
I’ve been told that final programming should help steer a story to be more suitable for a younger audience, and perhaps that applies to simpler vocabulary as well. Inflection and pauses in a voice also matter in grabbing children’s attention. My test bear’s voice varied, and when the voice lacked emotion or sparkle, my 5-year-old couldn’t follow the plot. You wouldn’t want to follow a story based on the voice of the robot calling you to tell you that a prescription is ready for pickup. You need a voice actor.
Skyrocket shared a recording sample of ElevenLabs’ near-final vocals, which I included in my video embedded above. This sample seemed to have the ability to pause and emphasize the right words at the right time to mimic what a voice actor might do.
More AI-powered bedtime stories are coming
It’s clear that this technology is evolving quickly. Last year, VTech’s CEO told the Financial Times that he sees AI-powered teddy bears reading stories to children by 2028. But now Poe is doing just that in 2024. And more parents could use ChatGPT to tell bedtime stories, even without the bear. Last month, at Apple’s Worldwide Developers Conference, senior vice president Craig Federighi highlighted how Apple’s upcoming software will make it easier to use ChatGPT to create stories for children without having to pay for an OpenAI account.
I’ve been covering major toy shows and industry trends for over a decade, and in the past two years, there’s been a growing buzz about AI chatbots as playmates. There’s already the $99 Miko Mini, which comes with an optional subscription. The price tag increases as AI tech becomes more appealing, like the $799 Moxie robot.
Comparing Poe to the Original Narrating Bear
The experience of testing Poe reminds me of an iconic talking bear from my childhood: Teddy Ruxpin, the first animatronic talking toy. Released in 1985, Teddy Ruxpin moved his nose and blinked in time with audio stories played from cassette tapes. The bear came with physical books that kids could flip through and read. The toy was a huge success and cost a whopping $70, or $204 today.
But the story of Teddy Ruxpin required a team of writers, musicians and voice actors (who also sang). It was a fantasy adventure series about friendship, starring Teddy, and it spanned 60 story tapes that were translated into 13 languages.
So when I look at Poe, I see that the art that made Teddy Ruxpin so special has been exported to a machine. Poe doesn’t have his own series of adventures, but Skyrocket claims that the app could create sequels to the stories it generates. That means a child could create multiple stories with the same characters — though it’s hard to say how closely they’ll resemble the same characters when the creativity is left to a computer.
Poe is not yet good enough to replace the task of parents reading stories to their children before bed. But he could be used as a tool for creativity, almost as a way to spark interesting family conversations. Maybe these strange stories inspire children to think about creating their own stories.
It’s worth seeing how the storytelling evolves as the bear gets closer to launch. And I’m curious to see if Poe will inspire other AI toys down the road.