Director of Strategic Innovation Shaun T. Yorgey '97 showcased three interactive Artificial Intelligence projects in a hands-on maker space at the 2020 NAIS Annual Conference in Philadelphia. Developed along with Allison Rodgers, Coordinator of the Artificial Intelligence Institute, the projects demonstrated to participants how Artificial Intelligence works, while having some fun in between conference sessions.  

"We weren’t sure what to expect heading out to NAIS’s makerspace with a small suite of opportunities for folks to 'make' using artificial intelligence," says Mr. Yorgey. "The response was informative and reveals the next steps in mass adoption of making sure all of our students receive a base education in the field."

Perkiomen's maker space set up included three Mac Book Pros with web cameras, one digital keyboard, and a basket of plastic fruit. The three stations used free cloud-based software to allow participants to create with different aspects of A.I.

Teachable Machine
This station allowed participants to train a machine learning model based on images. We use the webcam of the computer that we are working with, and take lots and lots of snapshots of plastic fruits at different lighting and angles. We label these test images as we’re working on them. For example, if we’re using bananas, we label our samples so the program knows which bananas are not ripe yet, ripe, or overripe. The machine learning program then makes decisions about what makes these categories different. For certain characteristics, it might be more accurate than others.

Mr. Yorgey shares the following results:
"We launched Google’s free (& no account needed) Teachable Machine. Using the plastic fruit and a web camera, participants were able to photograph the different types of fruit categorizing them as pear or apple, or lime, or papaya, etc. By the end of the first day we had enough participants to take well over 10,000 images of nine different pieces of fruit. The A.I. struggled understanding pear from papaya as their coloration was very similar. As a result, participants trained about double the number of images for these two compared to the others. The final result was an A.I. trained to identify these nine pieces of fruit with a nearly 100% accuracy. This station demonstrates a technology that is utilized extensively in the medical community already in any areas that use a visual recognition benefit, for example, looking over a CT scan, identifying tumors, flagging skin cancer, etc. The average dermatologist sees about 12,000 incidences of skin cancer in her lifetime. Similarly trained A.I. has used a data set of hundreds of thousands of classified images. Another application one can imagine is searching through satellite imagery for industry and defensive purposes." 

“It’s Free?!” The fact that students can access these types of applications on the web for free was a complete surprise to most participants. Almost everyone who participated in one or more stations photographed the URL they were working from. Many sent it off to a colleague or two right away.

AI Duet
This station allowed participants to play a duet with the computer. The notes we play (as humans) into these programs are taken as inputs, and then the computer programs make decisions as to what our playing sounds like, and adjusts or adds to it to make it sound more musical.

"Another free (& no account needed) Google app is A.I. Duet," shares Mr. Yorgey.  "This station utilized a digital keyboard, though one is not necessary to participate. At this station, participants could play a song on the keyboard and the A.I. would jump in and play a duet to accompany the song being played. A.I. Duet has been trained on a massive set of audio data, and uses its training to determine how to best harmonize and complement what the participant is playing. For many participants it was interesting, but it became very intriguing when a capable piano player stepped up to the keyboard. A.I. Duet performed best when it had an increasingly complex player to partner with."

"I thought creativity would be one of the last places to impacted by A.I.," said one participant in reference to the performing and visual arts stations that allowed users to work with A.I. to be creative.  

Selfies with Deep Dream
This station will allowed participants to use the webcam on a computer to take a selfie. Using that selfie, folks can run their photos through the Deep Dream generator. This software was originally set up to help computer scientists have an idea of what a neural network sees when it is classifying photos. The strange edges (and stranger eye-shaped forms) that sometimes show up in these images make them different than other filters -- and also beautiful! Deep Dream generation has since become a wacky way to create new art. 

"The final station utilized Deep Dream A.I. to show the complexity of what the machine is 'seeing' when it interprets imagery," says Mr. Yorgey. "There are several ways to approach Deep Dream, but most participants opted to filter a selfie or other image through a second image. For example, one participant took a selfie as the base image (providing the composition framework of the new image). As a filter image, he choose Van Gogh’s Starry Night. This provided a stylistic guide for coloration, fidelity, and overall interpretation of the base image. The result is that the A.I. made the decision of how to preserve the selfie as identifiable as the participant while making the image distinctly Van Gogh.  The results are awesome. Similarly we ran an image of a red stag through a famous Der Blaue Reiter painting to make a distinct image that could easily be recognized as part of the movement." (pictured below)

"A recurring theme in user's reactions to using A.I. is how approachable it can be," shares Mr. Yorgey. 'I can’t believe we were able to do this without coding,' said one participant.  In each of these stations they were able to see the real-world use and most found it surprising that they didn’t need a background in computer science to understand or use it."

Special thanks to NAIS for the opportunity to participate in the maker space and to Mr. Yorgey for sharing his experience. 

The Institutes at Perkiomen School       Inquire      

By Bernadette Kovaleski, M.J., Director of Marketing & Communications, Perkiomen School

Bernadette Kovaleski earned bachelor’s and master's degrees in Journalism and Communications and delights in telling the transformational stories set at Perkiomen School. Her experience includes advancement, development, and communications roles in higher education and the non-profit sector before coming to Perkiomen School in 2016.