Sweaty Machine Learning
My mind constantly orients itself to make any necessary task a more creative one, but despite my best efforts it’s still hard to design while riding a bicycle. While the past 16 months and thousands of cycled kilometres have been the best way to clear my head and keep fit, I found myself dreaming up ways to disrupt a circular task.
Inspired by friends generating intriguing images using AI engines, I collected stats from my Wahoo bike trainer and Zwift virtual cycling app, and transformed them into Python Script parameters for VQGAN+CLIP, running on Google Collab’s virtual processors.
I first translated my cycling data into a series of prompts to guide AI creation; for example, if I cycled a distance of 25.3 km, I’d have the AI complete 253 000 iterations of the assigned prompt. I interpreted additional factors like average speed, watts generated, ride duration, vertical climb distance, and scenic details of the virtual world.
Each goal yields unique results, and generates a trippy, abstracted representation of my accomplishments, like cycling a virtual Tour de France or climbing the equivalent of Everest in vertical metres. Using this goal-driven art generated in achieving these goals felt natural to craft custom cycling attire and accessories that reflected what I did to get there.
For me, cycling continues to be the best way to organize my thoughts, and collaborating is the best way to push beyond my comfort and perspective. I didn’t expect machine learning to become my latest accomplice, but with every ride offering a chance to generate a new piece of art, I now have even more motivation to climb that next hill.
