Critical Response #2: "Power to the People / Humans in the Loop"

by Andrew Lee

CS 470: Music and AI

Part 1

In response to "Humans in the Loop: The Design of Interactive AI Systems," I agree in challenging the modern view of fully-automated AI systems. Back in November last quarter, I began to see posts of AI art arise in my Instagram feed. Frankly, I was terrified. While the high quality of the works was scary, what frightened me more though were the comments: "This is so cool!", "Lowkey better than what the artist created", "LOVE THIS OMG", etc. Although these opinions only accumulated a small portion of the comment section, and there were many opposing viewpoints, I was worried that society would begin to accept such works, artistic pieces created fully by AI and had no human involvement. This was actually one of the main reasons I decided to take this class.

From a philosophical discussion I had with a friend recently, I now think that the meaning of a piece of art lies in its process. For example, if I had to compare two sunset paintings, one being an aesthetically pleasing piece made by an AI, and the other being an mediocre-looking piece made by a novice artist who went to a beach to attempt to capture it, I'd find the work by the novice artist to be more valuable. If someone told me to judge the two pieces simply by the appearances, I'd question them, "can an artwork really be understood and appreciated simply by its surface?" Rather, it is the process of the piece's creation that determines its value.

Because of this, I think human-AI interactive systems are much more valuable than fully-AI systems. I felt that what we, as a class, have came up with through Etude 1 and Project 2 to be incredibly innovative and thought-provoking. Compared to MusicLM, which only generates prompt-based music, we developed a whole new domain of music creation! Isn't that much more impressive? Similarly, we have developed incredible systems such as AlphaGo and speech-based language translators. However, that doesn't mean humans will stop playing Go or they'll stop learning foreign languages to communicate with others. Instead of developing more AI systems for replace and automating human activities, I'm much more interested in developing interactive systems that can contribute to new things for us to do.

For instance, as a composer, what if I wrote a new musical piece by selecting and combining fragments from multiple AI-generated pieces? Or perhaps, I could collaborate with a music generator to write a piece, where we switch off between who composes it every few measures. And this doesn't have to limit to art-related tasks. For the case of self-driving cars, I'd personally feel safer sitting in an Uber that's both operated by a human and an autopilot system than just an AI system. It goes beyond trusting an AI system too; I feel that human drivers can provide a sense of flexibility, compassion, and understanding that is not exhibited by an autopilot system, making it much more enjoyable for me as a passenger. Ultimately, there's something about that humans and experiences that just make some activities irreplaceable, and we don't need to try to make systems that replace them. Instead, AI should be a tool to help us create new ways to enjoy these activities.

Part 2
  1. Self-driving systems as an interactive, educational tool to teach humans to drive
  2. Using music generators to provide chord-progression options for a melody, helping music composers brainstorm
  3. performing improvised music as an ensemble of both human and AI musicians, where AI plays in response to the human, or vice versa
  4. A collaborative painting: a human starts off drawing on a canvas for 10 minutes, and then the AI adds to it for 2 minutes, and then the human painter adds more, and they continue to switch painting until the piece is complete.
  5. Show recommendation systems that allow for more abstract user feedback, such that it could understand feedback like "Give me more Romance movies" or "Stop suggesting me movies like Twight".
  6. An interactive, educational music compositional AI that can teach a human to compose in certain desired styles by providing short motif or rhythmic phrases to their current ideas.
  7. A ChatGPT that asks its user for more details and feedback whenever it receives a query. For exmapkle, if I ask it write me a poem, it could ask me for feedback on its generations or additional clarifications, rather than just giving me a single answer and end the conversation.
  8. An prompt-based sketches generator that gives you simple sketch ideas for a simple art prompt, so that you can get some inspiration and develop the drawing yourself using these suggestions.
  9. A medical diagnosis AI that works along with a doctor, such that the two can reach a decision on a patient's condition based on both exisiting data and personal professional experiences not found it past data.
  10. Cooking AI that allow users to specify which steps they want the AI to do, while they perform the other steps themselves. This allows for more personalization and flexibility in cooking different recipes.