With schools, businesses and travel shutting down around the world to slow the spread of COVID-19, many families are facing weeks or even months at home. Unlike regular seasonal school breaks, the sudden suspensions of classes and extracurricular activities has caught many parents unprepared. With parents trying their best to work from home — many doing so for the first time — balancing work with looking after kids can be exhausting.
Can AI help?
Synced looked into smart solutions that could help keep kids engaged and entertained, and might even be educational. Here a few of the AI-powered websites and experiments we found.
Google Senior Engineer Monica Dinculescu built the Magic Sketchpad to enable an easy and fun way to doodle on screens. A machine learning algorithm attempts to finish each stroke to match a category selected by the user. The Magic Sketchpad however seems a bit low-tech and visually sparse for today’s kids, and might instead end up as a nostalgic pastime for parents.
Quick, Draw! is another game built by Google with machine learning. On the Quick, Draw! website, a neural network attempts to guess what the user is drawing. Although it doesn’t always work, its responses can be fun and the accuracy improves the more you draw and play.
If regular toys don’t keep your musically inclined kid entertained, how about giving them an entire orchestra to play with? Standing in front of a web camera, kids of all ages can conduct their own virtual orchestras through the Semi-Conductor website. Grand, conductor-style arm and body movements are used to change the tempo, volume, and instrumentation of a performance. Google researchers created the game using PoseNet, a machine learning library that map movements through webcam input.
A computer performs piano duets based on the notes a user plays on a virtual piano keyboard. Coder and musician Yotam Mann created AI Duet to allows users to play a duet with a machine. Using neural networks, AI Duet listened to countless melodies and learned structural elements such as the relationships between notes and timing to build its own music understanding, generation and collaboration ability. You hit notes and the program responds to what you’re doing — so the better you play, the better the duet!
Pattern Radio: Whale Songs is an interactive AI-powered website that enables users to explore thousands of hours of undersea audio in search of humpback whale songs. Researchers from Google and National Oceanic and Atmospheric Administration (NOAA) Pacific Islands Fisheries Science Center (PIFSC) visualized a year’s worth of underwater recordings in a spectrogram.
Users can zoom in to identify individual sounds, not only of humpback songs but also of passing ships and fish and even some unusual and unknown sounds; or zoom out to observe months of sound at a location. An AI heat map predicts where whale songs are most likely to be found, highlight bars help visualize repetitions and patterns within the songs, and guided tours from whale song experts point out moments of interest in the data. Users can easily share links to interesting sounds with their friends.
This Google website showcases a collection of simple AI experiments that enable anyone interested in machine learning to explore projects that focus on art, technology, design and culture. The experiment collection includes projects covering AI + Drawing, AI + Learning, AI + Music, and AI + Writing. There’s even a “FreddieMeter” AI-powered singing challenge that rates how closely your singing matches the voice of Freddie Mercury!
Journalist: Fangyu Cai | Editor: Michael Sarazen