Develop AI systems to efficiently learn new skills and solve open-ended problems, rather than depend exclusively on AI systems trained with extensive datasets
We're going to need to get an algorithm to figure this out...
I feel like the physics based approach works well here again. We talked about how the grids are actually 3D spaces, and that color and depth are interchangeable. So you could have a form inhabit the same (X,Y) coordinate and that'd be no problem because they are differentiated by the Z-axis.
The pattern is very easy to see here, but the idea of getting an AI to perform this task feels very challenging.
I like the idea of simulating the gird, and that the final output is computed (take the top-down 2D representation) based on the 3D world. In this example, we see how the input forms "lose" their bottom right square because the next form overlaps it, but in the simulation there's no change of form, just a change in what we can "see".
Discover other posts like this one