General Systems Thinking (An Introduction to)

Your favorite books and links
jacob
Site Admin
Posts: 15907
Joined: Fri Jun 28, 2013 8:38 pm
Location: USA, Zone 5b, Koppen Dfa, Elev. 620ft, Walkscore 77
Contact:

Re: General Systems Thinking (An Introduction to)

Post by jacob »

In terms of problems like bryan's image recognition, this is a kind of type 1 error that's common in machine learning, especially the "blind" kind where an algorithm is thrown at the problem w/o considering/forcing any internal structure. Genetic programming is notorious for being able to fit almost any problem with a solution that's completely an utterly unique to the particular input sense while making no sense if the input data is varied only slightly. There was a crazy example, I think in Bostrom's Superintelligence book, about using genetic algorithms to optimize certain hardware circuits and ending up with a solution that turned out to exploit some unintended feedback process (radiowaves ... as some of the circuits of the PCB turned out to work as an antenna) from the testing equipment.

You also see this problem a lot on algorithmic trading and in normal human thinking.

One way to check it is (as in the video) to see if the solution outcome is robust to small perturbations in the input. If it is, that's a promising sign. A very general rule is that as long as one has the correct feedback process, the parameters are actually not that important (and work over a wider range)... whereas if one has the wrong structure, it's still possible to fine tune a system so that it works in special cases but becomes erratic just outside the training set.

An example of a correct structure would be Sclass braking system where the sensors are correctly hooked up to, say, the rear wheels, which, say, are the only ones that brake. An incorrect structure would be making the mistake of hooking a sensor up to a nonbraking wheel. Now, if the either of those are calibrated to a single test run (from 50mph to full stop in a straight line), they will likely both optimize well for this case. However, the wrongly hooked up system would not perform very well under other conditions.

In human proverbial (<- can one say that?): It's more important to do the right thing than to do things right.

bryan
Posts: 1061
Joined: Sat Nov 29, 2014 2:01 am
Location: mostly Bay Area

Re: General Systems Thinking (An Introduction to)

Post by bryan »

I hate exercises for the reader!

If 1st level is simple PID setting temperature (X) on temperature sensed(Y), is the 2nd level involving the derivatives, dX and dY?

jacob
Site Admin
Posts: 15907
Joined: Fri Jun 28, 2013 8:38 pm
Location: USA, Zone 5b, Koppen Dfa, Elev. 620ft, Walkscore 77
Contact:

Re: General Systems Thinking (An Introduction to)

Post by jacob »

@bryan - Nope ... any kind of installed algorithm is 1st level no matter how complicated. This is not an easy exercise btw.

bryan
Posts: 1061
Joined: Sat Nov 29, 2014 2:01 am
Location: mostly Bay Area

Re: General Systems Thinking (An Introduction to)

Post by bryan »

OK. so maybe you mean systems in systems, or X level removed from a system? For the later, the third level (I noticed you already gave a 2nd level) could be something like if a human in Norilsk loses ta's job, ta will end up lowering the thermostat since ta doesn't have the money for the bill?

edit: obviously if it is the later, X level removed, it explodes exponentially..

jacob
Site Admin
Posts: 15907
Joined: Fri Jun 28, 2013 8:38 pm
Location: USA, Zone 5b, Koppen Dfa, Elev. 620ft, Walkscore 77
Contact:

Re: General Systems Thinking (An Introduction to)

Post by jacob »

Y .. meaning system of systems of systems. A market economy (the whole structure!) is a good example of a level 3 system turning a person's furnace on and off. So what's a level 4 system for individual furnace operation? Or a level 5?

The size (possible structures of potential feedback mechanisms) explodes as a tetration, so much faster than exponential.

7Wannabe5
Posts: 9370
Joined: Fri Oct 18, 2013 9:03 am

Re: General Systems Thinking (An Introduction to)

Post by 7Wannabe5 »

bryan said: Another clear segment of emergence is AI/machine learning. For instance we have neural nets that can classify with quite high probability subject content of an image.
Right. One time I was trying out a new Plant ID app, but I didn't enable GPS, so it didn't work very well until I figured out that it didn't know where I was. I have played around with several of them, and I would not judge them to yet be accurate enough for important determination such as edibility.
Sclass said: I'm not so sure of the human analogy. I take it you're interested in AI? A friend I used to work with described adaptive feedback as a form of AI.
I think I am interested in adaptive feedback. Your self-tuning thermostat example is kind of on the money, but I would be part of the mechanism.

What I would like to be able to do is somehow combine the capabilities of systems simulation software with 3-D landscape design/Sketch-up/animation and the spreadsheet where I tracked my happiness level based on various lifestyle factors, and continued input of data gathered by me and other streamed sources such as weather forecast, stock market reports, fitbit, plant database or calendar.

The first time I designed a garden using 3-D software, almost 20 years ago, I was rather amazed because my planting actually ended up looking very much like it was predicted to look by the program. However, in retrospect, I realize that this is because there was a certain level of collaboration, covert level of understanding of how to proceed with such a project, going on between me and the program. For instance, the program and I were both assuming that I would clear the ground of weeds and provide adequate soil. Also, the database of 3-D models of plants was limited to varieties with well-known characteristics. So, given these rather strict parameters, the program was able to project a quite accurate image of the appearance of the garden in given season and/or after several years of growth had occurred.

OTOH, when I kept a spreadsheet where I kept track of my daily happiness level and how I spent my time, I was rather surprised by some of the resulting correlations. For instance, I determined that I enjoyed travel/novel activity even more than I had expected. However, I have not been able to do a very good job of feedback looping these results, or developing measures a bit more complex than self-reported level of daily happiness.

Anyways, I am making this sound very serious, but it is really my idea of a fun toy. For instance, wouldn't it be cool if I could inform my system model that I added 3 American Cranberry shrubs to my garden this morning, and then there would be a 1.2% probability that if I requested animation of Future Me at 7 AM on 9/5/2025, I would see that I was putting American Cranberry Jelly on my toast for breakfast, but then if I failed a blood glucose test on 1/26/2020, and the system learned that I had stuck to my resolve to give up Jelly for 3 months, the likelihood of 9/5/2025 animation showing me eating any sort of jelly on toast would fall to .25%.

7Wannabe5
Posts: 9370
Joined: Fri Oct 18, 2013 9:03 am

Re: General Systems Thinking (An Introduction to)

Post by 7Wannabe5 »

@jacob: I was going to say that Level 3 would be the human's set of increase-warmth-comfort strategies, maybe conscious and unconscious. For instance, human chooses to develop habit of wearing thick sweater and socks rather than adjusting thermostat, or the human's lifestyle design system notices that whenever the human works out on a regular basis, has a cuddly girlfriend or dog, and remembers to water the houseplants, his thermostat behavior at 65 degrees is the same as his behavior would otherwise be at 69 degrees.

bryan
Posts: 1061
Joined: Sat Nov 29, 2014 2:01 am
Location: mostly Bay Area

Re: General Systems Thinking (An Introduction to)

Post by bryan »

> tetration

Never used the term. Fair enough. Granted, it is syntactic sugar on exponentiation which is syntactic sugar on multiplication which is syntactic sugar on addition.. today I shall coin bryantion in which it is "n copies of a combined by tetration"

Are you sure it is tetration and not bryantion? :D

Regardless, it all comes back to discrete numbers if we freeze things (no lower level states changing? could be thought of in different ways, I think. Time being an obvious choice to freeze.). I mention this just because it's interesting to think about and keep in mind (also noting that the universe is more vast going down the size scale than it is going up it). So yeah.. emergence all the way down.

User avatar
Sclass
Posts: 2791
Joined: Tue Jul 10, 2012 5:15 pm
Location: Orange County, CA

Re: General Systems Thinking (An Introduction to)

Post by Sclass »

bryan wrote:
Tue May 09, 2017 3:53 pm
> the compiled code cannot necessarily be modified on the fly

It all depends on the underlying system (such a general term..). If the lower level hardware/software does not enforce things like memory read/write/execute exclusivity, you can compile a program that can modify itself. In practice this means concepts like JIT/dynamic compilation/translation or more commonly interpretors. These are very often built into programs; but even if they are not you may be able to find ways to manipulate a program input/state as if they are (Sclass said as much here).

Very interesting. As said I'm not a real "computer" programmer. I have a pretty myopic view of computer organization. I treat my program memory as sacred ground in my applications. Thanks for the links, I'll have something interesting to read after dinner tonight. :)

The last time I played with a program that could change program memory after compilation I got an uncomfortable visit from a system administrator and my management. :lol:

User avatar
Sclass
Posts: 2791
Joined: Tue Jul 10, 2012 5:15 pm
Location: Orange County, CA

Re: General Systems Thinking (An Introduction to)

Post by Sclass »

7Wannabe5 wrote:
Tue May 09, 2017 4:50 pm
, I realize that this is because there was a certain level of collaboration, covert level of understanding of how to proceed with such a project, going on between me and the program. For instance, the program and I were both assuming that I would clear the ground of weeds and provide adequate soil. Also, the database of 3-D models of plants was limited to varieties with well-known characteristics.
Reminds me of Jerome Lanier's criticism of the invention of midi messing up how music sounds today.

7Wannabe5
Posts: 9370
Joined: Fri Oct 18, 2013 9:03 am

Re: General Systems Thinking (An Introduction to)

Post by 7Wannabe5 »

Sclass wrote:Reminds me of Jerome Lanier's criticism of the invention of midi messing up how music sounds today.
Exactly. I wonder how a robot gardener system would go about recreating Vita Sackville West's white garden? Maybe I would just instruct the system that Aesthetic = maximize (Refined, Natural) and see what happens.

Post Reply