Complexity science starting to rev up

J

Thread Starter

Jim Pinto

Automation List :

Chaos Theory, also known as complexity science, is the study of how order inevitably emerges from chaos and how all systems - natural and created - innately seem to crave and develop efficiency and organization.

Dick Morley, technology guru, programmable logic controller pioneer, and coauthor of The Technology Machine, has been preaching the applications of Chaos Theory for more than a decade.

Now, Jim Rutt, Internet pioneer quit his job to dedicate himself to the applications of complexity science to computers so machines and their applications can become more intelligent.

Your challenge is to find applications for genetic algorithms, neural nets, and complexity science in industrial measurement and controls instrumentation.

Read the latest Pinto's Point (27 Feb. 02) on the ISA website
- Complexity Science starting to rev up :
"http://www.isa.org/journals/intech/brief/1,1161,1671,00.html":http://www.isa.org/journals/intech/brief/1,1161,1671,00.html

Cheers:
jim
----------/
Jim Pinto
email : [email protected]
web: www.JimPinto.com
San Diego, CA., USA
----------/
 
JimPinto:
> Your challenge is to find applications for genetic algorithms, neural
> nets, and complexity science in industrial measurement and controls
> instrumentation.

This is backwards.

The proper challenge is to find solutions to problems, not to go around attempting to apply (poorly-characterised) technologies to everything in sight.

And the behaviour of all of the above *is* poorly characterised: both in obtaining the solution and in the solution itself, it is difficult to evaluate how good they are (reliability, maintainability, etc).

Jiri
--
Jiri Baum <[email protected]> http://www.csse.monash.edu.au/~jirib
MAT LinuxPLC project --- http://mat.sf.net --- Machine Automation Tools
 
JimPinto had said:
> Your challenge is to find applications for genetic algorithms, neural
> nets, and complexity science in industrial measurement and controls
> instrumentation.

Jiri Baum <[email protected]> commented :
>This is backwards.

Jim Pinto responds :
No, it is not backwards.
Instead of solving control problems using the same, old, standard methods, it is time to develop new solutions using self-organization and other complexity science methods.

See my article from Dick Morley's Chaos Conference, April '98 - The Advent of Self-organizing Industrial Controls - The end of centralized, deterministic control systems :
"http://www.jimpinto.com/writings/selforg.html":http://www.jimpinto.com/writings/selforg.html

Jiri Baum continues:
>The proper challenge is to find solutions to problems, not to go
>around attempting to apply (poorly-characterized) technologies to
>everything in sight.

The control engineer who is used to using deterministic controls which are "well characterized" will simply play out the same old scenarios. It is clear that deterministic control systems are simply not robust enough, and break down at higher I/O point-counts.

Complex adaptive systems (CAS) have significant advantages, which can be applied to many current applications.

Jiri Baum :
>And the behaviour of all of the above *is* poorly characterised:
>both in obtaining the solution and in the solution itself, it is
>difficult to evaluate how good they are (reliability, maintainability,
>etc).

Yes, you are right. But, who will apply and "characterize"? Leaders demonstrate superiority of new technology, and then followers
follow....

Cheers::
jim
----------/
Jim Pinto
email : [email protected]
web: www.JimPinto.com
San Diego, CA., USA
----------/
 
> JimPinto had said:
> > Your challenge is to find applications for genetic algorithms, neural
> > nets, and complexity science in industrial measurement and controls
> > instrumentation.

> Jiri Baum <[email protected]> commented :
> >This is backwards.

> Jim Pinto responds :
> No, it is not backwards.
> Instead of solving control problems using the same, old, standard
> methods, it is time to develop new solutions using self-organization and
> other complexity science methods.

That makes sense, of course. The original phrasing was rather different, though - "Your challenge is to find applications for ...". Even to an open-source proponent and PhD student like me, that sounded too much like a solution looking for a problem.

> See my article from Dick Morley's Chaos Conference, April '98 -
> The Advent of Self-organizing Industrial Controls -
> The end of centralized, deterministic control systems :
> http://www.jimpinto.com/writings/selforg.html

Programming systems by specifying constraints rather than desired behaviour is probably a good idea, but it has little to do with chaos, complexity, neural networks or genetic algorithms. Indeed, it sounds quite good to me.
There are a few problems I can see, but once you solve those it'll likely become a very nice way of programming machines.

- I'm not aware of any language or representation that'd have a good handle on talking about time.

- It is not necessarily how people think. Some cases are sufficiently algorithmic to begin with that it's easier to specify what the machine should do rather than what it shouldn't.

- Relatively little experience with the systems, debugging of these programs, etc, outside academia (logic programming, planning).

Once you solve these problems, you can plug the result into something like mercury, which will compile it from the declarative, constraint formulation into an efficient executable. The techniques are reasonably standard (and
deterministic, I might add).

> Jiri Baum continues:
> >The proper challenge is to find solutions to problems, not to go around
> >attempting to apply (poorly-characterized) technologies to everything in
> >sight.

> The control engineer who is used to using deterministic controls which
> are "well characterized" will simply play out the same old scenarios. It
> is clear that deterministic control systems are simply not robust enough,
> and break down at higher I/O point-counts.

It is not clear that self-organizing algorithms are robust enough, or that they scale to higher I/O point-counts. In fact, I'm pretty sure that many of them don't - no-one knows how to train them beyond what are in reality very simple tasks.

Perhaps rather the control engineers need better tools, and/or better training. If the state of the art is ladder, augmented by a hand-drawn SFC,
it is not terribly surprising that it breaks down on even medium-sized problems. But the solution isn't to throw away sound engineering.

> Complex adaptive systems (CAS) have significant advantages, which can be
> applied to many current applications.

Such as?

> Jiri Baum :
> >And the behaviour of all of the above *is* poorly characterised: both in
> >obtaining the solution and in the solution itself, it is difficult to
> >evaluate how good they are (reliability, maintainability, etc).

> Yes, you are right. But, who will apply and "characterize"?

Field application of untested technology is usually a bad idea. This is the job of R&D departments, universities and other centres of research.

However, the declarative programming idea is promising; if you can make it practical, it sounds like it'll be very good.

Jiri
--
Jiri Baum <[email protected]> http://www.csse.monash.edu.au/~jirib
MAT LinuxPLC project --- http://mat.sf.net --- Machine Automation Tools
 
Top