The first hackathon I can remember participating in was the Ghostly Visual Music Collaborative hosted at Eyebeam in New York City.1 It was about a week. It consisted of staying up late, eating pizza, getting to know one another, exploring a huge digital archive of music, learning from pioneers in the space, and pushing ourselves to make applications that could visually represent a song. It was electric. Over the years I participated in a few others. When I left for graduate school in Paris, I also left hackathons. Last week, Refik and his studio invited me to join a hackathon centered around geospatial data. The problematic: How could this data integrate into Dataland’s current exhibition Machine Dreams: Rainforest. As the two day event transpired, I was struck by something: due to AI the experience of the hackathon had changed. Let me elaborate.
Source code as source material
Like watching Iron Chef, hackathons are a race against the clock to achieve a goal. Chefs are given a secret ingredient. They have little time to prepare a menu. Then it comes down to execution and presentation in front of an audience. Similarly, we received access to datasets I never really looked at or knew about. We had little time to decide the direction we were going to go. Then we were off building. Rather than building ourselves, however, we had AI agents do much of the grunt work. We fired off prompts. Because of the speed and collaborative nature of the hackathon, I found myself being more carefree with the code being written than I am day to day. Participants were sharing code snippets and artifacts in fairly undocumented ways. Not a problem, because the AI agent could extrapolate and mock it up. In hackathons of the past, I would prep different project folders to be able to navigate and iterate quickly. For this hackathon, I felt more like I was assembling. I directed the AI agent to mix this code snippet and that code sketch. The result allowed me to converse more with my fellow participants and let the AI worry about the code.
More to show
In typical hackathons a lot can go wrong. The slightest typo and the whole application errors. Long hours increases mistakes. This is compounded when you use the graphics card or GPU. The errors are hidden. The visuals either do not show up or show up broken. It can take hours or days to remedy. Painful as it was, emergent designs can appear through this process. Personally, this is one of the satisfying moments of programming in pursuit of creating visuals. You get visual proof of a newfound understanding. Today, the AI agent is able to confirm errors and fix them before even presenting anything to you. This saves a lot of time. It also makes a lot more things to look at and review. Instead of spending late hours figuring out how to orient different spheres in some kind of physics relationship, I was sketching wireframes of how to present all the prototypes made.
These key differences made the entire experience different from what I was expecting. I had more of a creative director’s hat on than a technician’s. So, what did we make? Unfortunately, I cannot show you everything. But, I can share this initial prototype. Play with it directly by hitting the button above or view screen recorded video at the beginning of the post. It is best viewed on a computer, but does run on a phone. It combines a vortex algorithm to push particles around and the tendency for those particles to resolve and resemble the earth. The result is a kind of representation of the earth where displacement and movement of its surface could be attached to real-time storm data. Though that data is not assigned in the prototype. There are a few sliders at the bottom of the page where you can change the amount of vortices and its behaviors. You can also tap and drag to rotate the earth or pinch-and-zoom to get a closer look.
–Jono
Visual Music Collaborative Results. Creative Applications. 2010. https://www.creativeapplications.net/event/visual-music-collaborative-events-results/