The trap of complexity


#1

To build something so ambiguous as The Machine a step to overcome would be the complex nature of itself. No matter if we are looking on TM as a literal physical machine or as a unification a of social and economic structure, it would have to overcome a problem nobody has overcome yet! There is a natural limit for the complexity of human made projects, where they become too inefficient until they fall apart.

So let’s assume that we have all the resources we need for achieving state of TM. Countless developrs ready to work. It’s almost as something as this great complexity should be supervised by someone else rather than humans. Even if there are the greatest managers and leaders, by scaling TM and acquiring new people to manage the increasing complexity, the bad sides of humans WILL manifest itself (greed, desire for power). And about this stage the projects start to fail. So is there anyone to help us? Extraterrestrial life has been rather silent so far ; ) Could AI be utilised for such tasks? Not the traditional programmed AI, but rather one which is capable of recursive self-improvement? (This introduces a whole new set of problems, but let’s skip that) However the technology for such AI is not yet here.


#2

Considering how advanced already existing “things” are, I don’t think complexity itself is what requires some kind of AI - rather, I think it’s the fact that a lot of the tasks and decisions such a system would have to make are simply beyond what can be hard-wired, and therefore would need some kind of higher level decision making.


#3

I’m definitely concerned with how the hell to manage this project. There are so many ways we can fail.

Still, I know it is in principle possible, so I’m willing to try.

I don’t have a lot of better answers right now except that I know we will have to be constantly aware of possible failure modes for the project as a whole and we will need to be willing to correct quickly as we see that happening.

I hope that if we keep a core philosophy of making profitable modules, people will maintain motivation to continue even after many errors.


#4

The top level comment by user philipkglass here was very interesting as it relates to project complexity.

https://news.ycombinator.com/item?id=13467572

Comment copied here:

"After announcing the project back in September, we have now learned that Tesla and Southern California Edison (SCE) have completed the massive 80 MWh energy storage station using Tesla’s new Powerpack 2 at the Mira Loma substation…While Tesla and SCE haven’t officially launched the new substation yet, sources familiar with the new Powerpack installation told Electrek that it was completed a few weeks back – late December – and brought online so that the electric utility can start using it to manage peak demand.Deploying this project in only 3 months is really impressive.I think there’s a pattern here: energy systems built from repeating small modules have systematically lower risks of cost/schedule escalation.

See for example “Construction Cost Overruns and Electricity Infrastructure: An Unavoidable Risk?”

Look at Figures 2 and 3 in particular. The mean overrun by project type goes:
solar < wind < thermal (fossil) < nuclear < hydroelectric.

That’s basically the same ordering as “minimum viable project size.” Coincidence? I don’t think so. A small solar project can be a handful of panels. If you want to scale up to hundreds of megawatts, just tile a
much larger area using the same panels and basic repeating structures. A lot of work can execute in parallel and sections can start generating output/revenue rapidly as they’re completed, sometimes years
before the project’s last panel is installed. No such luck with a nuclear or hydro project: you can be on year 7 of an 8 year schedule and it’s still generating zero watts because so little can happen in parallel. (This is also why I consider small modular reactors the last, best hope for a nuclear renaissance, but the latency for nuclear R&D is so high that renewables might “eat the world” first.) I rather expect that this Tesla installation, despite its impressive speed, was not particularly cheap. But if battery costs keep declining, in 5 years it could still be lightning quick to build another installation like this and cheaper than gas/diesel peakers. If that happens then storage-backed renewables become the very cheapest electricity source across large swaths of the Earth, threatening huge chunks of coal and natural gas demand. There may be a parallel risk to oil demand if Tesla and other auto makers execute well on their EV plans currently in the pipeline. I think Tesla faces some significant risks from the SolarCity acquisition, and execution risks on the Model 3, but I sure can’t fault the ambition."


#5

That is a very interesting comment and observation. I definitely think it would make sense for TM as well, exactly because of the complexity. Being able to build it bit by bit, taking out and adding in, improving parts and so forth all in parallel seems crucial to any kind of success with any kind of project like this.


#6

Yeah. This is why I envision building small complete “Machines” (modules) before slowly understanding how to integrate them into larger more general machines.

Also the first thing to prove out is if these modules can be profitable somehow on their own. If so, we’re in business, so to speak.


#7

That’s true. It’s also significantly more difficult to make money with the things you build than to just build them :stuck_out_tongue: haha

It would be quite incredible if it was possible to generate some reasonable stream of revenue working on this kind of project, though. I’d feel inclined to do it full time when I’m done studying if there was a potential to support myself doing it


#8

Yeah. I’ve seen so many neat projects that are only ever neat projects. Building some machine is the easy part really - building some machine that is better than the alternatives is really hard. But if you do that, you can exploit your advantage for profit. And then you can keep going.


#9

Do you know of any good open source robotic arms? In all my research, I’ve yet to find anything that actually looked “good”. The new Dobot m1 looks pretty nice, but it’s not open source.


#10

Not in particular, but I have a friend who really loves arms who operates this site:

https://robotarm.org/

I see some recent posts so make sure to check that out.


#11

The newest arm he posted there is described as open source, but in reality, only the tool head mounting design is open source, and some of the software used. This seems to be the typical route that robot developers go - they make it “open source” by only open sourcing those parts that people need to extend while not letting people see what the actual robot looks like.

It’s a real shame


#12

Good points about the complexity and countering it with the module system. The solar vs hydro comparison is very interesting. On a similar note, I’ve always had so mixed feelings about mega projects like ITER and the limits of human collaboration (considering the funding as a part of project). But now I see a the module system as a viable solution, at least to a certain point.

So for me the TM crystallizes as a potential company functioning within the means of current economic world (at least for now). It is producing technical solutions for automation. The grand future plan however is to transform the human society model to one where everyone’s basic needs can supplied thanks to freely available technology, instead of just accumulating massive and unnecessive wealth for its shareholders.

Do you also see it in a similar way?
Bonus thought-provoking question - what could be the entry product for such TM?


#13

I think before processes get handed over entirely to machines, I feel there’s a lot of work we can do to simplify things on the user interface side. Graph databases and flow based visual programming mixed with code for specific low level components and unifying stuff together in a manner similar to inferno - shared virtual machines running common code and abstracting networks into the file system so the whole thing is like one big os with different resources in folders - stuff like that, not necessarily specifically that - might be good starting places.
I feel like a lot of options get buried / overlooked / forgotten. I think people dislike visual programming in part because it doesn’t feel like coding, even if it can be a lot easier to make sense of the arrangement of elements in a uml-type flowchart than to stare at a wall of code or a folder structure.