December 31, 2022
Written by 
Christie Pang

Should AI be part of the village that raises our children? Our Collective in Asia gave it a try.

At the height of school closures due to the Pandemic, caregivers had to, one way or the other, “home-school” their children while many worked-from-home at the same time. This created an incredible amount of stress for families. The US Census Bureau published an article in August 2020 that detailed many statistics showing the transition to online schooling, and stay-at-home orders required at least one adult in the home to focus on the children — helping them with schoolwork and supervising them all day. 1 in 5 working-age adults said they stopped working because the Pandemic “disrupted their childcare arrangements”. In Asia’s urban cities like Hong Kong, that usually means within the confines of less than 600 square feet. Stressed out parents were trying to get their children to sit for “Zoom school” for hours, blocking off video games and YouTube from laptops and iPads, while giving up in the evening to unlimited and further screen-time. All this seemed to have exacerbated any attention deficit and hyperactivity tendencies in many who found it hard to release pent-up emotions and energy.

Parents, in particular working moms, were among the unsung heroes of this crisis. They have adapted their households and juggled work, children’s schooling, and other household needs. Even in homes with extra hands to help, mothers bore the brunt of the tantrums, fought against learning loss by becoming personal tutors, and lost sleep scrolling through articles about how their Covid Babies will have delayed communication skills.

This is me trying to act out our collective Mom (and Dad) guilt.

It was during this time that Clement and I decided to take a long bet on society and take a stab at shaping how shiny new Generative AI and NLP capabilities are going to impact our next generation’s cognitive, social, and emotional development. We spent 3 days in 2020 holed up in a conference room and came up with an adaptive, AI-enabled multi-media platform prototype, based on the collective product insights we gathered from our 200+ parents and educators in an earlier survey. Within a week, dozens of families decided to bring their children in-person to our Hong Kong office (all in N-95 masks and with repeatedly-sanitised hands obviously) to try out what we called a “pretotype” - yes, we were big fans of Alberto Savoia’s book The Right It.

After learning from 120 hours of children interaction, we started to see unique patterns of engagement, multi-modal learning, and screen-off movement breaks emerge for all developmental ages, with the goal of optimizing each child’s mastery of subjects.

Observation One: AI as a companion that encourages and guides

Most children intuitively participate and respond to questions, call-to-actions, and object manipulation within videos or songs, even when the video itself has no interactive elements built-in (it’s like a mind-trick if the content’s engaging enough!). This is further encouraged by their learning companion (be it a co-viewing teacher, nanny, or in our case, an AI-enabled chatbot that hypes it up). The “hype-man” role a GPT-powered chatbot can play is unique - when given the right prompts about skills the child has mastered, books they’ve just read, problems they are working on or struggled with, the bot can speak unique 10-second prompts to a child that helps them recall what they’ve learned in the past. Coupled with engagement and learning engines, GPT can encourage children to participate with personalized hints and rewards. It can even DJ and get children to stand up and dance every 20 minutes so that they get a screen-break!

Observation Two: AI as a personalized librarian

Without boxing any child into a certain “learning style”, patterns began to point towards custom “blends” with mixed modalities for each child - from audio books, nursery rhymes, teacher videos, and even rap songs (Jack Hartmann was a huge favorite in Asia!) - all contributing to that unique child’s progress in mastering whatever topic is at hand. There are also measurable mastery improvements when AI mixes it up and prescribes classroom-style material interspersed with reinforcement entertainment. Like a customised mixtape!

Observation Three: AI did not think kids master topics in a linear, curriculum-aligned fashion (duh)

That is probably over-stated in teaching, but yet a hard reality to face and scale for educators and curriculum designers in practice. Looking back at the “playlist” of content and assessments that were played on our pretotype, every single hour of interaction a child had with the platform differed in terms of:

  • Most featured video channels
  • Proportion of classroom-style vs. entertainment-style content to keep engagement high (measured by observed body-language and reaction times of children)
  • Number of times a child is “assessed” before the learning model confirms they have mastered a subject (simple multiple choice questions were used in pretotype)

Perhaps the most exciting to us, is the curation and calculation of what exactly made a child master a subject, in the shortest amount of screen-time necessary, while having the most fun they could possibly have.

We experimented further with dozens of engagement metrics, mastery probability indicators, and feedback mechanisms that rate segments of each song, book, video, and podcast on its ability to improve learning outcomes.

Want to know what parents and grandparents said about the observations and their children’s mastery of new topics after playing with the prototype for a month?

Stay tuned for the next Collective post!