All we have is our limited cognitive abilities. With those limited cognitive abilities, we create machines with limited cognitive abilities. The limitations are enfolded within us and remain when expanded via limited cognitive abilities to broader and/or more precise mechanisms. This to me means that we may build machines that are capable of anything we can imagine but are also only capable of anything we can imagine. It’s a wall no single human may reach but collectively it’s there. Thing is, we won’t know it. So what is reality but what we can map? What’s unreachable will always be invisible.
Thankfully, there’s more than enough to keep me going. I do a lot of experiments to understand things and to try to go places I never saw anyone else do myself.
Last handful of days I was obsessed with trying to make videos from an old movie utilizing mathematical notions but using found tools (and no math myself).
Outlining with minimal surface areas
Voronoi (skeletonization) – finding medians (not averages)
finally: Labeling (grouping).
I didn’t know this was my quest: I just allowed my interests to lead and knew I was done when I said to myself, “That was satisfying”.
Hindsight provided a reason. A need to justify “following my bliss” prompted me to create a story to tell myself after the fact.
Now I feel as if I understand concepts on a visceral level – via my participation in following my curiousities.
And… I can turn any movie into stick figure movies (sorta) if I wanted
Oh gosh no. I wanted to do everything _automated_ – to have a process. But fast. I don’t have the patience for programming anything more than a macro or batch file anymore But finding a “thinning” algorihm (Voronoi / skeletonization) in an open source medical imaging software (ImageJ / fiji) was a super treat.
What I’m doing is far more basic, but it’s just these kinds of things that AI is programmed to look for – and I wanted to see what limitations / assumptions are getting baked into things. This is basic labeling / grouping of unknowns.
Classic outlining – but a good outlining. So with this trio there’s: outline, “shortest paths” (cellular / travelling salesman stuff, structure) and grouping (hierarchy, layers, objects separated in some way), it’s enough to construct a system of any complexity with this as primitives,
Problem is: It’s all based on binary. I literally had to do a binary mask PRIOR to each of these.
So EVEN WITH finer and finer gradients, the binary dichotomy is there and how it’s chosen is either arbitrary or hand-picked, but none are automatically correct.
That’s Simpson’s Paradox..
A lot of what I’m doing is combating my own prejudices. I do not like hierarchies, object types (or typed programming languages), etc, as I know that ANY can be transformed into any other SOMEHOW, no matter how tenuous the connection.
I love finding what’s happening in that transformation space where the orthogonal happens.
I find I am disappointed when the answer ends up being: Us.
I already know it is. But I want to find something more technical, can be described, a metaphor that aids understanding.
That’s a pain but satisfying when I find it.
For example, in programming, any language can be created in untyped lambda calculus but you can’t get untyped from typed UNLESS you treat untyped as a single type.
But untyped is “dangerous”. TRUE=FALSE is ok.
So, the quest for the perfect typed language continues on in computer science, adding problem when its trying to solve them.
So, like that binary image of you, a cutoff threshold value has to be established. Is it 255? 0? 128? Which gives the most value?
Arbitrary. Simpson’s Paradox. Arbitrary. It’s “what do I want to see”.
But “what do I want to
As long as we check and recheck results, compare and contrast, finding the proper cutoff thresholds (what to put into which buckets) for classification and categorization in order to abstract for useful purposes, we’ll be alright.
As a guy who can get his head lost in the clouds, I’m constantly comparing to see if I’ve gone *too* abstract. Thanks for the post and conversation. Sugihara’s focus on geometrical forms is fascinating and certainly a change from today’s heavy leaning on triangular surfaces.
(outline -> inkscape -> Ctrl-A –> Path: Trace Bitmap –> Simplify (repeat a few times) -> Export Bitmap –> Upload to FB). explicability (from latin for Unfold)
Influential paper! I go to scholar to look at influence and to see “how far we’ve come” and “how long we have yet to go”.