Other interesting "science" related news...

[Judith Rosen <***>]

Another science newsletter that I subscribe to is from NewScientist.com and they've got some interesting stories going, as well. One struck me as the perfect story to analyze from a Rosennean perspective because it appears that the technology being developed is already going in that direction-- with an anticipatory approach, to boot! 

A long time ago, when artificial intelligence enthusiasts turned away from using biological models of the brain for developing AI technology, my father said; "That's a mistake. They'll never get there that way." He said they needed to investigate the capabilities of "neural networks" because of the many options these information processing networks are able to generate. (They're loaded with all sorts of "hidden" entailment.) Such processing networks are much more resilient and far better at multi-tasking, he said. They also should be more likely to be capable of the kind of "function change" ability that organisms are so good at; a trait that is partly what drives evolutionary processes. I'm not sure why the digital aficionados chose to scratch the biological model for intelligence if they were already using it, but it smacks of a misguided attempt to "be more scientific." (If we buy in to the machine metaphor of physics, and agree that all systems are just like machines, then the logic would probably be: Why use a biological model for a machine technology?) In any case, it appears that enough time has passed, with enough frustration building up, that AI has finally turned back to learning lessons from biology. It may be a machine technology, but it's a capability of living organisms they want to create in their machines (...and Descartes had it backwards to begin with).

The first paragraph:

<x-tad-smaller>Smart fire detector could slash false alarms
</x-tad-smaller>	 14:10 25 October 2005
	 	 NewScientist.com news service
	 	Kurt Kleiner
<x-tad-smaller>A fire detector that can tell the difference between burning toast and a burning building could save money, annoyance, and possibly even lives, by cutting down on false alarms.

German company Siemens will start selling the detector in the UK to commercial users by January 2006, and the technology could eventually make its way into homes, says the firm's fire safety manager, Andrew Morgan.

The detector uses four sensors and a neural network to determine if the smoke and heat it's detecting are from a fire or are just part of the normal room environment.

In the UK more than half of the 872,000 fire call-outs in 2004 were bogus, and 285,000 of those false alarms were due to fire detectors. Responding to false alarms costs money, and in the home annoying false alarms encourage people to disable their alarms.</x-tad-smaller>

The last bit has a little anticipatory flavor to it, as well as a "counter-intuitive" relational consequence of not implementing this new technology. Siemens is producing a technology to avoid problems in the future of the sort they have found analyzing the past and among those consequences is that people whose smoke alarm goes off every time they cook something are going to kill the smoke alarm. The reason I describe that as a counter-intuitive response is because the whole idea of a smoke alarm is an anticipatory one, created by human beings (arguably the most anticipatory organism on the planet). Yet, I'm one of those people who would actually smash a smoke detector that false alarmed more than twice in a short space of time. Why? Because I don't trust anything that stupid! Besides, there's something to be said for quality of life, vs quantity. But mainly, it would be because I would conclude (being an anticipatory system) that the alarm would continue to go off, based on it's past record (I would predict the future behavior of the system based on a model I'd generated of its behavior pattern). So the end result would be that one anticipatory approach completely undoes the intentions of the other and the smoke alarm is no longer functional.

<x-tad-smaller>Built-in intelligence

</x-tad-smaller><x-tad-smaller>Most home alarms are designed to go off when smoke in the air exceeds a certain concentration. But as most people have discovered, the detectors do not distinguish between smoke from frying bacon, steam from a hot shower, or fumes from a smouldering mattress.

Some commercial systems are more sophisticated, feeding data from a number of different sensors to a central computer and letting the computer decide whether the readings indicate a fire.

The Siemens detector is different, Morgan says, because it builds artificial intelligence into each individual detector, using custom-designed integrated circuits.</x-tad-smaller>

<x-tad-smaller>Neural network

</x-tad-smaller><x-tad-smaller>The detector contains two thermal sensors and two optical sensors. The thermal sensors monitor temperature and its rate of change. The optical sensors monitor the size and colour of the smoke, with one sensor optimised for thick black smoke. Thick black smoke is, paradoxically, relatively hard for an optical sensor to detect, since the sensor works by reflecting light off of the smoke.

These four sensors input their data into a neural network – a computer circuit that approximates the basic structure of biological brains. Neural nets are good at considering a number of inputs and recognizing patterns in them.</x-tad-smaller>

It's interesting to me that the definition of intelligence here appears to be quite lenient. While I have defined it for myself as "the ability to think and learn" I would say that this device doesn't really qualify. What do you think?

The current commercial smoke detector system, borne of an attempt to solve the same problem of too many false alarms, is the typical reductionist/mechanistic way to "create a more complex system" when the word complex is synonymous with "complicated" or "intricate". They've done it by adding more machines, most of which all have the same capability. Now, however, they are each linked to a new, central machine with its own, separate capability. Not only is this getting way more expensive, but there are now a collection of links in linear chains all leading to a single processor on which all of the satellite units completely depend), ALL of which has the potential to malfunction in numerous ways, leading to false alarms or no alarms. I sure hope that at least some of the central processor's capability is spent assessing the fitness of itself and of the network, with an alarm if it detects any hint of malfunction, but even so, there are very limited options for rerouting information with this kind of centralized authority. This situation contains a list of  ingredients for what Robert Rosen termed "a fatal infinite regress". No matter what we could try to do to safeguard against the weaknesses of the organization, it's still just adding more of the same, which increases the chances of things going wrong. There's no way to fix the problems without doing a complete reorganization of the system, itself-- a fact that is consistent with RR's definition of the word "complexity."

The Siemen's approach, in contrast, uses an internalized network of information processing, which is programmed to respond to a particular set of  patterns in the readings. The only way to recognize patterns from individual readings is by analyzing the relations between the actual readings and comparing the pattern of relations to preset "models" of significant patterns. Even more intriguing, is the information in the next paragraph:<x-tad-smaller>

When they are installed, the detectors are programmed with a set of typical parameters for the kind of room they are in, so that they know what patterns of temperature, smoke colour, and particle size to consider normal. For instance, a detector in a workshop would know to ignore dust and machinery exhaust, and one in a bathroom would ignore steam, but both would react to unusual smoke or temperature readings that could signal a fire.</x-tad-smaller>

What fascinates me about that is the recognition, which went into designing this system, of the importance of context. They didn't create a "one-size-fits-all" kind of pattern-sorter; they created open options for tailoring each system to the environment it would be "inhabiting". I think that's cool! I have a lot of questions, though. Like: How did they arrive at their patterns for the templates? Is it possible to reset or re-task the system after it has been installed? What if they have bugs in the programming, independent of any actual problems in the patterns, the tolerances, the electrical wiring, etc?

Even so, this is an intriguing application of two of my father's suggestions for how to put biological information and examples to good use. I wonder how much the smoke detectors cost, though....

Judith

Web address: http://www.rosen-enterprises.com
BioTheory: An electronic journal of general science based on the Relational (Rosennean) Complexity Paradigm