I wanted to do a midweek update on my progress and to think about how to spend the rest of the week…

So far, I have spent the week making a proof of concept for the third performance, which I had initially anticipated would be pretty light on tech. However, last week I had the idea to use image classification to add another layer to the performance. Not only would it be commenting on the ridiculous hoops people will jump through to save money on behavior-based insurance discounts, but also about how shitty most consumer visual ML applications are at this point. I came across Kaia health, which is an insurance plan that also works with a AI physical therapy application to calculate your flexibility and give you live feedback on your exercises. I am not sure if your movement data is used to calculate your rates, however it certainly has some of the feel related to my thesis. I also looked at Muse, which is an EEG headband that partners with a meditation application to track your brain waves and make you more mindful. It is my understanding that most consumer EEG headbands are bullshit (according to my girlfriend who is a neuroscience major). All this to say that, for my third performance, while I am trying to balance working out, meditating, and having a healthy breakfast, I am considering adding on interactive technology or commenting on the use of interactive technology. Am I doing all of this just to a Kinect with the speculative understanding that in the future computer vis will be omniscient? Or should I incorporate other devices into the mix?

I also feel I have to make a choice about what this performance will be. Like Ballyturk, I had initially imagined showing an audience a cacophonous yet well choreographed morning routine that has one mis-step and falls apart. However Ellen showed me (at least I think it was Ellen) Let’s Paint, Exercise & Blend Drinks TV! and I really liked it. The main benefit of making something like Let’s Paint is that I don’t have to choreograph a 10 minute performance, just simply make a task difficult enough for me to show up as a performer. However, I will have to allow the technology to be responsive to any unplanned mistakes and happenings, either using ML or a stage manager with a large amount of controls. This week, I worked on training a neural network with ml5 based off this tutorial that would be able to monitor and respond to whether I was successfully executing exercises in my workout. The result was mixed success and I don’t have any other experience training a computer vis model. It may be very time consuming to train a model that is accurate enough and which doesn’t make the performance all about how shitty the model is. Moving forward with the build, I think that I should start with the physical performance and layer on the technical elements as I go, incorporating fake interactivity before actual interactivity. I need to keep telling myself that the theatrical gesture is the most important, the theatrical gesture is the most important, the theatrical gesture is the most important.

Here’s me trying to balance meditation, working out, and making some avocado toast: