Project Update Anugrah
Problem/Issue being addressed
When walking through the AI exhibit at the MIT museum I was quite unconvinced that it could substantively educate people about how modern AI works. The parts that I connected with most were the gigantic physical machines and not the small AI writeups. I think part of the issue lies in novelty - now that ChatGPT exists and is so easy to use, interactive displays (like the poem writing one) don’t actually add anything meaningful that you couldn’t just accomplish at home.
However I think it is quite possible to create a more informative addition to the AI exhibit that explains the rough methods behind the high level ideas in modern machine learning/AI instead of just explaining the high level ideas so quickly that the viewer just gets lost.
Envisioned audience
The intended audience for such an exhibit would probably be an adult or a highschool student. I think having some rough background in what modern technology looks like would be helpful, and I don’t know if someone too much younger could take much away in terms of the high level details.
Potential museum collaborators
I think a project along these lines would really only work at someplace like the MIT museum. Throughout its history, MIT has been pretty tightly connected to the growth of technology, and I think such additions would fit naturally into the AI exhibit.
Technological/spatial approach
I think the ideal presentation of this project would probably be something like 2 smaller informative demos with sparse text that visualize what a neural network is, and how they can be applied to images (i.e. adding in a few things like convolutions on images). Then we could have a cutting edge demo that shows some piece of technology that has been big on the internet in the recent few months and rotates every once in a while. For example, given the current internet interest about image generation models like DALL-E, we could have an interactive demo that somehow works in real-time e.g. a local server that hosts a StableDiffusion model and allows people to play around with different parameters.
Skill sets needed
Primarily this sub-exhibit would require knowledge of modern machine learning, some UI/UX design for creating the actual interfaces, and some basic web scripting experience.