Develop AI systems to efficiently learn new skills and solve open-ended problems, rather than depend exclusively on AI systems trained with extensive datasets
In this example, we identify horizontal columns as groupings that will persist. We notice the column colors change from input to output. We're unsure if it's the columns moving or if the colors are being translated, and how that translation takes place.
We test color translations, to see if any of them are consistent:
Order the colors in some way, and see if there is a fixed shift,
Order the colors in some way, and see if there is a shift based on column index
See if there is a consistent mapping between input and output color
This is found to be true
So we learn that green maps to yellow, red to magenta, blue to gray, cyan to maroon. We have multiple examples where we can verify these mappings are consistent.
Discover other posts like this one