[FRIAM] exploiting the uncanny valley...

[uǝlƃ ↙↙↙]

I think the low dim representations do exist, but are social constructions [⛧]. Dark Souls is interesting because it gave rise to the "souls-like" genre. I think part of the reason the DS games were labeled notoriously hard is because the representation did not exist as an intersubjective thing. Your Zoom avatar(s) could be constructed either by profiling you and mimicking that *or* by profiling everyone else in lots of Zoom meetings (subtracting you out), setting constraints within which your avatar randomly wiggles. I think the latter would produce a much freer/schizoidal avatar than the former. So, I think that argues for the existence of those low dimensional representations, even if they're fragile or somewhat incoherent.

How latent they are depends, I'm guessing, on iteration. How often does one have the same (type of) interaction with the same set of people so that the pattern of the interactions becomes ingrained or pushed "down" into habit? And does variation decrease (monotonically) as iterations pile up? When I first started brushing my teeth with my left hand, I was terrible at it. I tried to get my left hand to mimic my right hand. But that was simply the wrong way to do it. I finally gave up on mimicry and allowed my left hand to wiggle whatever way it wanted in order to focus on the objective of brushing. Now I'm quite good at it. But the way I brush with my left hand is different from the way I brush with my right, much more arm movement and much less wrist movement, the way I grip the brush differs, etc. And that experience helped with a badly acute case of tennis and golfer's elbow after chemo when I was trying to build my strength back. I got tennis elbow on my right side and golfer's on my left. The way our stabilizer muscles work seems to be a good example of latent low dimensional representation. I'm not sure where the data stands these days on innate handedness, though.


[⛧] That whole argument that the OSX UI is/was somehow more intuitive than, say, Windows is just nonsense to me. I explicitly bought an Apple laptop in 2000 *just* to train myself. I used it for something like 3 years on a daily basis. I never grokked it. It was similarly difficult to get a hold on Ubuntu Unity and Gnome 3. But because I could escape to the command-line any time, I stuck with it longer and finally grokked it. (Had to give up Gnome 3 for Xfce4, though. I guess my age is seriously impacting my willingness to engage unfriendly interfaces.)

On 1/21/21 4:12 PM, Marcus Daniels wrote:
> My peeve of late is with people that talk about latent low dimensional representations as if it were a given that they exist and are generally valid to use.  Of course, these animations hold most variables constant about their subjects and change some coarse or local geometry.   They are not generative models trained on observed behavior.   I dream of an avatar that can reproduce me on Zoom calls for a few days so I can go to the park and play with my dog.   I'd like to have training sets like "in a good mood and feeling generous" which I could replay again and again and again coupled to whatever topic.  Eliza like stuff that just responds to the speaker with remarks like "What a good idea?  Can you tell me more?"   


-- 
↙↙↙ uǝlƃ