With the New Year came a great new ChickTech workshop on artificial intelligence. On January 26th 2019, our amazing hosts from IPsoft, R&D Engineer Nick Corso-Passaro, and UX Designer, Ruth Ong, welcomed the girls to their magnificent downtown Manhattan office. IPsoft’s main AI technology, Amelia, helps businesses automate operational processes through a natural language interface, freeing employees and customers from repetitive tasks to focus on creative ones they enjoy.
The morning started with Nick introducing IPsoft and coding basics. ChickTech high schoolers learned the principles of coding through the lens of popular game Minecraft. Using drag and drop code, they advanced through 12 levels, starting with basic ‘left’ and ‘right’ commands to more advanced conditional statements and FOR loops, helping their Minecraft characters navigate different challenges. This fun exercise set the tone for the day, how machine solutions depend on basic logic to work, and that myriad ways of solving an engineering problem exist.
The workshop then segued into machine learning. Girls played a version of Nim using Jolly Ranchers to learn the principle of rule-based or supervised machine learning. Through trial and error, participants realized the mathematics behind winning the game, similar to how machines are ‘rewarded’ or ‘punished’ (supervised) to drive desired outcomes.
Next was an important aspect of machine learning called ‘feature extraction’. Two participants tried to guess a Disney character based on features that the rest of the group had extracted. Hilarity ensued when several characters were incorrectly identified, proving this simple exercise harder than the girls thought! This highlighted a common issue with machine learning—confounding. This happens when a machine’s accuracy is reduced due to inadequate feature extraction.
Participants then tested their newfound knowledge using Microsoft Cognitive Services. Participants first uploaded a set of celebrity photos to the system and assigned tags to each photo (e.g. “Selena Gomez” versus “not Selena Gomez”). After creating this simple 2-classifier model, participants moved on to more complex, multi-classifier models (e.g. “Selena Gomez”, “Cardi B”, “Vanessa Hudgens” vs “Not Selena Gomez, Cardi B or Vanessa Hudgens”). They uploaded more photos and manually tagged each photo to ‘teach’ the machine to identify their favorite celebrities. It became clear that machines still need humans to make well-designed and well-supervised classifiers to work as intended.
After lunch, Nick and Ruth surprised the girls by giving them a tour of the Amelia showroom. An entire floor was outfitted to showcase Amelia in various everyday life roles, such as a drive-thru attendant, retail associate or home assistant.
ChickTech would like to offer great thanks to our attendees, instructors, volunteers, and IPsoft for hosting the event.
Our next event will be Sunday, March 17 in Brooklyn. Stay tuned!
Written by Jaelynn Theobalds