After recent conversations I find myself wondering what Forth running on this substrate would look like. This guy should try to collaborate with Chuck Moore on Green Arrays. cc @wtaysom @Garth Goldwater @fyr

extremely cool specific example for people with less time on their hands

there isn't a lot of talk about cellular automata for general purpose programming here, but I think in some domains it'd be worth exploring

I wanted to pitch\prototype a cellular automata based VR programming environment (basically a hybrid of MFM and Minecraft redstone), but it seemed too radical, getting to the system I've implemented now has required hours of meetings and 100s of pages of documentations with engineering managers and directors, so I gave up even bringing up more "radical" ideas

from the design system front eg with the VR environment example it seems like him working on it mostly alone has lead to way more productivity because he starts with super small pieces and just aims to figure out eg the two or three automata that could work together to sort rather than having to promise success and productivity within the next 3 years or whatever

lol you're right about the glue code-it glues new code into the spaces (which compose your infrastructure)

this is kind of like eventual consistency for behavior instead of data now that I think about it

Not sure you can run something bit-oriented on ulam exactly, short of reintroducing nand gates as discussed in the youtube comments. You could go slightly less far and represent the stack as a physically linked list of constellations of atoms, but that still seems like it's going against the intended massively distributed diffusion model, if that makes any sense.

I say “_may_” there because it also forces every programmer working at this level to rewrite security & reliability features for every project, rather than relying on existing components that gain maturity and thus trustworthiness.

I don't think Dave has published much about security as he seems to repeatedly emphasize how enormously early the theory is. in one video he presents a process (cell? actor? idk what to call them) that repeatedly forks itself being defeated by a faster-forking protector/immune cell. totally unclear how you could guarantee your defenses would fork faster than a malicious actor. seems like yo prove that you'd have to start enforcing some invariants about things that enter your chip vs things you create yourself which seems like it would open some serious holes in the theory—but I suspect there are holes in every theory, the trick is the size of the tradeoffs (in this case, unknown)

What do you think about his stance on determinism and scale?

its also interesting to see how much of his systems seems to me like a mechanism to describe constrains, at one point he speaks about degrees of an angle, its to bad that you have to describe the box in such a clumsy way

(by not using degrees that is)

maybe that is needed to get "robust-first" still I think there is improvement in the expression of this robustness

Re fyr's comment on the two lineages of computing: Yes exactly! A good way to think of this 'computational stack redesign exercise' is as bringing real time dynamical system control principles to modifiable, programmable, 'cognitive' systems.

Re Garth Goldwater's thought about inherent error->visual explanations. I think the key link is physically spatialized computations. You share work with your spatial neighbors, you ask your neighbors to cover for you when things go wrong, and so forth. RAM destroyed space, that's it convenience and it's sin. Respatializing computation is a basis for robustness AND visualization.

Re fyr comment about network griefing: At the moment the T2 tiles are designed not even to connect at the CA level unless they're using the same laws of physics. So fork bombs, say, can't come from up the net unless they were already possible locally.

incidentally @Dave Ackley what do you use for presentation software? I feel like your videos end up having a super conversational feel as a result of their style

or video editing—unclear

still looking but this is my late night explanation so far (probably not fully helpful): if you are serious about being able to scale a system indefinitely, you run into synchronization problems. you can think of this as like the asynchrony introduced when you start having multiple servers in multiple physical places. physical constraints mean you have latency, and so you can only reconstruct an artificial notion of absolute time by having some kind of reconciliation mechanism that resolves things like timing by way of a construction which is a leaky abstraction. so instead what you need is some form of relative time. that means you don't have a universal clock anywhere. you can't ask "what happened after 8:05 AM?" so you have to start asking your neighbors whether they observed you doing something after you observed them doing something. and if you need to ask neighbors, you need to know who your neighbors are, which is a spatial question so then instead of asking a traditional stepping debugger question like "which function created this value?" you have to ask questions like "which neighbor gave me this value? and why?" and the "and why" part of that question boils down to "well MY neighbor told me this value and I transformed it this way" and then chasing that value down becomes a spatial question and it's clear your tools will have to display data that way

I may wake up tomorrow and be certain I came off as a crazy person for explaining it that way

also I'm on my phone so sensible edits like "is this paragraph clear?" are like 10x as hard for me at the moment

And how did you implement the card sorter? 😊

They're very watchable videos!

Key figure from the paper.

This is a more strident introduction: . I like it. Bottom up engineering: when you have intention but don't have a way to specify it directly.

I don't know how to proceed from here but we are now at ~ 60 answers in this thread, could someone (@stevekrouse? ) please create a permanent public "robust-computing" channel?

A month or so there was this submission on hackernews: https://www.yodaiken.com/2019/02/20/von-neumanns-critique-of-automata-theory-and-logic-in-computer-science/

funnyly enough Dave mentions the exact same reference ( von neumann) in the above talk