My Work

<aside> 💡

1. Considering the Model and Data Biography reflect on the the following questions:
   • What questions do you still have about the model and the associated data? Are there elements you would propose including in the biography?
   • How does understanding the provenance of the model and its data inform your creative process? </aside>

Questions and Propositions

• Brings me back to what I was reminded of during week 2: The quote “Innovation before regulation” It seems that this is how the field of Machine Learning has been operating, and recently, at larger than ever scales than before. This has lead to a lot of progress and the accomplishment of many extraordinary tasks.
  • My question is if there exist merits of operating under “Innovation before regulation.” And if so, then what are they.
  • Additionally, should we as a society allow some leeway on these ethical matters associated with ML model training and dataset creation. If we do, then where should the line be drawn?
  • In a reality where we never allowed ml research to conduct the ethical violations they have so far, then what would the field of machine learning and AI look like now?
  • If we proceed from this day on strictly enforcing regulation, what might the future look like? Would there be anything lost?
• If a model trains on data generated by users of the model or by other machine learning models, or even data it itself generated, then its very likely that these factors will impact biases and continue to build upon and perhaps amplify racist, harmful or non-ethical conduct.
  • Should we, and if so, how should we regulate who get’s to use a model and how their data should contribute to the model’s learning. If we are not careful, this could end up like the AI policing example where data and usage of the model might impact others in unexpected and unfair ways.
  • What about if the models are impacted by other models?
  • In both cases, how should responsibility be split in the case that something illegal or harmful occurs? For example, if self driving cars train on their user’s driving data to continuously improve, then in the case of a car accident, can and should the people who contributed to the training of the model (the users of the car) be held responsible?
  • And what if in the self driving cars example, the dominant population of users are drivers vs. Bicyclists, it’s possible that driver’s priorities and driving styles reflect their preference of cars, and therefore cause unexpected bias or behavior towards cyclists. And in this case, likely the people who can afford self driving cars inform how they behave whereas the people impacted could be part of a different group of people in terms of income, location, race, sex, etc.

Impact on My Creative Process

• As we’ve discussed and read about, model and data biography can cause very unexpected results and biases when used. For example, even something as simple as distance from a the camera when training (like on Teachable Machine), can impact the model’s performance.
• Additionally, as we saw from the video clip we watched in class, a lot of times models are not accessible by minority groups. In the case of speech recognition, up to recently, the technology was inaccessible to groups with different levels / abilities of speech (ie. people with multiple scoliosis).
  • Understanding model and data biography helps inform me of how I should design my product to be more inclusive and accessible. It is impossible to make something completely accessible to everyone, but when you understand what a model can and cannot do, it can help you utilize the model in the most efficient ways, and identify the target audience of your product.

My Sketch Documentation

p5js Sketch Link: https://editor.p5js.org/conniehu25/sketches/quZDoQrKY

Drum Machine on Github: https://github.com/ch3926/interactive-drum-machine/tree/main

Idea Overview :

Last week, I experimented with adding sound classification functionality to a p5 project called ‘Drum Machine.’ I decided to continue expanding on this project using poseNet.

Link to Original Drum Machine: https://editor.p5js.org/conniehu25/sketches/0oEH5eBZr

Component 1 - Adding poseNet Functionality

• I wanted to allow viewers of my sketch to be able to control elements on the sketch with their body movements. I chose to keep mouse functionality because I did not have time to make every element of the sketch interactive.