There are many reasons it’s so tempting to project onto a single axis but maybe the foundational reason is the dichotomy between approaching and avoiding, or if you prefer, between positive and negative reward in reinforcement learning terms. This blows up into good vs. evil, friend vs. enemy, and so forth.
Edit: Also this is why Venkatesh Rao is much more sophisticated than we are; he does PCA but projects onto 2 axes and makes a 2x2 square.
I like this a lot.
There are many reasons it’s so tempting to project onto a single axis but maybe the foundational reason is the dichotomy between approaching and avoiding, or if you prefer, between positive and negative reward in reinforcement learning terms. This blows up into good vs. evil, friend vs. enemy, and so forth.
Edit: Also this is why Venkatesh Rao is much more sophisticated than we are; he does PCA but projects onto 2 axes and makes a 2x2 square.