Simple Is Complex | Hacker News Reader

Simple Is Complex

avdi.codes

64 pointsvinnyglennon20 comments

Open articleView on HN

I like Rich Hickey's definitions

Simple (roots: sim, plex) – one fold or braid

Complex – braided or folded together

Easy – near to our capabilities

These are really articulate definitions that helped me as a programmer.

Emergent complexity from simple systems – it seems like there might be an insight here. Avdi says we can build complex systems out of simple components. He says complex systems are hard to understand. He says it's not feasible to understand the system or predict it. So they need to be measured and observed. That's interesting. I don't know if it's true that a system built out of simple components cannot be understood. How precise is the model? How many layers deep?

Complexity as defined in most scientific disciplines today refers to systems which are "greater than the sum of their parts," i.e., where the behavior of the whole system cannot be predicted by looking at each component in isolation, but only can be understood by the action and interaction between components and how that evolves over time. Most systems of interest are too big to enumerate all the possible n-ary interactions between constituent parts, so we must study them by other means: probe plausible simulations, define and compute statistics from observable data, or fit models which capture plausible correlations. To complicate this further, it usually takes just one cycle or feedback loop in the network / system to introduce enough complexity that most traditional analytical mathematical tools break down.

I studied complex (adaptive) systems as my Master's education and can confirm most techniques reduce to "linearize around the important bits and extrapolate from there".

Examples of complex systems include: social networks (IRL and online), economies, biological systems. Machine learning is one of the most promising tools for probing these because if we can fit an appropriate model, we can hopefully capture enough of the behavior that we can reliably extrapolate to unseen data / states.

It feels like the article ended just as it was getting warmed up towards something interesting! If any HN'ers have pointers/links to other interesting analyses of emergent complexity among simple interacting agents, I'd love to read 'em!

https://www.santafe.edu/

https://en.wikipedia.org/wiki/Santa_Fe_Institute

> The Santa Fe Institute was founded in 1984 ...

> SFI's original mission was to disseminate the notion of a new interdisciplinary research area called complexity theory or simply complex systems.

Unfortunately a pointer to the Santa Fe institute is a little too broad to my question (which was specifically about interacting agents, and I was implicitly thinking more about physical embodied beings, somewhat in the spirit of Rodney Brooks' robots). Also, I'm keen on concrete analyses rather than vague hand-wavy pontificating.

That said, the Santa Fe Institute is a great resource to be reminded of every once in a while. I'm particularly a fan of a line of research being pursued by Simon Dedeo (and others). eg: Optimal high-level descriptions of dynamical systems https://arxiv.org/abs/1409.7403

Sorry. I don't have any good references off the top of my head for that.

You know about "BEAM" robotics? https://en.wikipedia.org/wiki/BEAM_robotics

W. Grey Walter's cybernetics "turtles"? https://en.wikipedia.org/wiki/William_Grey_Walter#Robots https://en.wikipedia.org/wiki/Turtle_(robot)

Thanks, I’ll check those out! :-)

How about Conway Game of Life. The rules are simple but emergent pattern are complex enough that Wolfram could write a whole book about it.

https://web.archive.org/web/20190929151534/https://avdi.code...

Oliver de Weck from MIT has done a lot of work around complexity, including deriving definitions and measures. Turns out there is an objective measure as to how many microservices is optimal. The number comes to around 2.8 connections per component if I remember correctly. Can share the math, if anybody’s interested.

> Can share the math, if anybody’s interested.

Please do!

Definitely interested.

Couldn't really take much from this article, other than I should read that book, it sounds so interesting

TL;DR simple mechanisms can produce complex behaviours

My favourite example is the few simple rules of Go leading to possible extreme superhuman strategies.

What was the point again?

Simple isn't Complex any more than Love is War or Dark is Light.

System-wide interactions between any kind of components is about as difficult as it gets.

Second that. This conclusion "Simplicity leads to complexity." is just too much of a shortcut of the author's own train of thought. As you said, this sentence should read "Interaction and coupling leads to complex emergent behavior". I feel this summarizes this article better.

There’s a Alan Kay talk[0] that might illuminate for you that leap of logic about simplicity.

[0] https://youtu.be/NdSD07U5uBs

The article explained what that meant about as clearly as can be. If a pinch of poetic licence troubles you, skip those parts.

tldr; Micro-services architecture can start out simple at first, but when there are many of them, it will become complex even more so than the monolithic approach.

Don't put words in the author's mouth; nowhere do they contrast microservices with monoliths.

Also, one of Avdi's main points is that a proper microservice architecture is complex, but NOT complicated. And in my personal experiece, monoliths have a tendency to be complicated.

Reply on news.ycombinator.com