Barbie's Stories
formerly WonderBarbie
Made with Rebecca Skurnik | Year: 2019
Took charge of research, coding and story scripts
A Barbie "bot" storyteller that engages little girls with stories of powerful women in history.
Tools: Javascript, Google Dialogflow, Google AIY Voice Kit and Home assistant
How can we raise the self esteem and interest in STEM subjects for little girls and make use of predefined role models like Barbie to do it? How can we use AI voice technologies like Amazon's Alexa and Google Home, which are now ubiquitous in homes all over the U.S. and the most of the developed world, to help us motivate our girls in the face of vulnerability?
Abstract, Research and Approach
On average, girls are more motivated than boys to perform well in school, at least during elementary school. By the time girls reach high school, however, some may try to downplay their own academic ability in order make themselves more likeable by both sexes [Maccoby, E. (2002). Gender and social exchange: A developmental perspective.]. Even if this occurs, though, it does not affect their grades: from kindergarten through twelfth grade, girls earn slightly higher average grades than boys. This fact does not lead to similar achievement, however, because as youngsters move into high school, they tend to choose courses or subjects conventionally associated with their gender—math and science for boys, in particular, and literature and the arts for girls. By the end of high school, this difference in course selection makes a measurable difference in boys’ and girls’ academic performance in these subjects.
Statics resources at National Council of Teachers of Mathematics
Researchers and child psychologists believe that it is somewhere around the age of six that little girls start to conclude that math and science are not feminine subjects, or are too difficult for girls to do well in, or make careers in. Despite high performance, the work force in the STEM fields have a gender imbalance.
Therefore role models with non-traditional careers and stories with women in positions of power which are commonly attributed to men are very important for girls to keep an open, unbiased and fearless mind about future career options.
Our vision was this:
A Barbie with embedded circuitry, a speaker and microphone could enable us to create a dialog between the doll and the child, with the purpose of gauging the child's self esteem and providing engaging stories to motivate them.
The doll could be connected to the cloud with a data bank, storing the child's information and specific questions to ask the child in order to gauge that information. In specific doll related activities, such questions can be assimilated (and monitored by parents), and important data can be collected by the child's responses in collaboration with a few other metrics like school grades, hobbies and family background.
With this data, we could have gauged metrics that lie on a spectrum, use sentiment analysis on the responses to understand their psyches, and finally present the doll's findings to a concerned parent: an app could be made that notifies parents to investigate any alarming issues with a doctor, and offers corresponding advice if the child seems to be unmotivated in pursuing goals
that may not align with the social construct of gender.
We understood that this approach may not be safe and/or ethical - more on that in the Process.
Process
To start off with the process, we examined the internal circuitry of Hello Barbie, a WiFi connected, interactive doll by Mattel, and used external circuitry and devices to create a connection and a database, along with voice detection and analysis. The hardware inside the Barbie was designed to communicate with an app, both of which have been discontinued. Because of this, we needed to replace the hardware with our own.
For connection to the cloud, we approached this project in
three ways:
1. Using the Particle
2. Using a Raspberry Pi + Google AIY Voice Kit
3. Using a Google Home with Dialogflow (and no data collection)
.png)
1. Using the Particle, an all-in-one IoT
platform -- embedded in the doll it can be used with a
microphone to send voice recordings to the Particle cloud,
from where its text version can be taken to a server.
This server can have access to the Google cloud machine
learning engine, where we can use the incoming responses
for our sentiment analysis and return it back to the Particle
in the doll using the Particle API. Text-to-speech can be done
on the hardware using a microprocessor.
We decided to use Particle Argon board to then communicate with the AI software that we will put on a cloud server. For some reason the board was not connecting to the ITP sandbox, so we had to setup the Particle.io board at home on my home wifi.
We spoke with Don Coleman who is very familiar with Particle boards. He told us that this board does not have text to speech capabilities, which is not ideal for our project, adding that we either need to use another board/component with this board, or perhaps use a Raspberry Pi.
The Raspberry Pi may be too large to fit inside our Barbie, but this issue can be solved if maybe we can make a backpack for her, or a Jeep, or some other accessory to hide the hardware.
Concerns with this approach
Complexity without value
We could make the Particle work, but we were not sure if it's adding any value to the project's functionality while adding more complexity, because we would need to go through the Particle cloud. Particle cloud messages are limited to 250 bytes so we would need to break the voice recordings into multiple messages (which is exactly what adds more complexity). We also might need a second micro-controller for recording the voice and text to voice.
Privacy
Learning from the fate of Hello Barbie, a similar concept which was labelled as a major privacy concern and an "eavesdropping toy" by almost every critic of tech ethics, parents and teachers, we concluded that storing any kind of data for a child who cannot consent to it is wrong. Moreover, the chances of such a product being abused were high, and so was the risk of a privacy breach. Instead, we decided to let the dialog be one-way: with simple questions like asking a child's name (in order to address her), and what kind of story she would like to hear, the system would only narrate stories and hope to motivate without storing feedback from the child.
2. Using a Raspberry Pi -- in combination with the Google
AIY Voice Kit, and using Google Dialogflow to create
conversation with the doll.
.png)
After learning, assembling and connecting our AIY Kit, we set up the assistant and used basic pre-set dialogflows to test our new circuitry. Here's a demo of the assistant in action, ready to be packed inside a doll:
.png)
Despite the fact that this worked, embedding this circuitry in a doll wasn't possible, and it gave no advantage over using a Google Home Mini for the same application, given that the overall size of the latter is smaller, and its functionality is not only faster, but also easier to use with Google Dialogflow in terms of the console and uploading the agent(s). Uploading onto the Voice Kit circuitry would limit our data (of the stories) to the storage capacity of the Raspberry Pi in any case.
3. Using Google Home Mini + Google Dialogflow
to tell stories: finally, we settled on the decision to
create a basic dialogflow and adding basic
interaction to it, in order to tell stories, but with a slight
feel of personalisation.
The Stories
Barbie's Stories comprise of a simple choice: choose
which story to listen to according to a career, each
career having stories of multiple powerful or iconic women
serving as role models, often with surprising overlaps (like the
one between a movie star and scientist: Hedy Lamarr). This
way, one established female role model like Barbie can illustrate the
stories of many, and hence the importance of having a deeper
purpose in life, no matter what her field. The stories are scripted to
emphasise the importance of value over looks, the
contributions of women to society, and more covertly, the lack
of male support in their greatness or achievements.
An example:
This is a story about Cleopatra, one of the most prominent rulers in history. Cleopatra was born in 70 or 69 BC and was the oldest daughter of the Pharaoh Ptolemy the 12th of Egypt. She was well educated and studied math, logic, debating, and science. She spoke no less than nine languages, and possibly more than even twelve. Because of this, she was able to address commanders and leaders of different nations without a translator or mediator which gave her an advantage. Cleopatra was known for her beauty, but she was always far more than just beautiful. Despite her romances with Caesar and Mark Anthony, two Kings of the Roman Empire, Cleopatra wanted Egypt to remain independent of Rome. She built up the Egyptian economy, establishing trade with many Arab nations. She was a popular ruler among the people of Egypt both because she embraced the Egyptian culture and because the country was prosperous during her rule. She is still celebrated to this day. We can all learn from her to be confident, strong and intelligent!
Dialogflow
A sample of the flow:
To create the conversation, we dived into learning about Dialogflow.
We created an agent called Barbie, although we used Google
Home assistant, in order to use the character of Barbie, and
prototype + user test with girls of different ages first.
After learning about intents, entities and fulfilment, and the
specific JavaScript code we needed to customise our "bot",
we deployed our Barbie's Stories onto the assistant and tested it
out with Eliza, a 6 year old girl in Parlin, New Jersey.
With this user test, we learned that despite the system
being a narration/mostly a one-way conversation, it was
not only engaging but truly empowering. By the next morning, all of
Eliza's friends and cousins (aged 6-9) had asked me to let them play
with our project. They remembered the details from the stories and
asked intelligent questions, while discussing their future careers.
For the next phase of development, we would need to:
1. Use micro-circuitry to embed this system into a real Barbie doll
2. Improve the Dialogflow by adding more natural conversational
elements, and the ability to loop the storytelling cycle with
the question, "Can I hear another one?"
3. Have different complexity levels for different ages
4. Have the persona of a male child in order to script it for both
genders, or another version altogether.