I thought it was interesting how the imageClassification program from class often gets the image wrong if it’s slightly more obscure, so I wanted to play with the idea of artificial “un”intelligence by coding an unsure, apologetic AI who’s just trying their best.
In order to do this, I changed the “gotResult” function, shown here in the original code:
By adding a variable “resulting_label” and adding numerous replace functions to the label, as well as text for the label and probability results:
The replace functions include several letters swapped for other letters that produce “dumber” results, like replacing u with oo and s with th. The results are some pretty unintelligent and unsure answers from a poor AI who’s just making their best guess on some confusing images…
This project was challenging, especially with getting started and exploring the possibilities of editing a preexisting in the first place, but I enjoyed thinking about ways to add personality to AI and exploring how even wrong results can produce a funny, interactive program. I felt like the AI programs we worked with in class can seem very robotic and lack personality or voice, so I wanted to humanize it a bit by making the wrong answers seem acceptable because the AI had tried their best! I had to simplify my original idea into one I could code, but I think there are many possibilities for using machine learning technologies outside of their intended functions but still producing entertaining programs, and I’d love to go further with this with my future projects!
My code: https://gist.github.com/jujucodes/df555376b27a34734a279dcdefe8cb63