Thursday, February 10, 2011

Chemical Finite State Machines... CFSMs ?

(Editorial note: This post started on Jan 18, 2011 and remains largely unedited flow of consciousness rambling. A few notes are added to sort of make this worth posting. I did say sort of)

Dear MIT, Stanford, UC-Berkeley, CIT, UI-Urbana, GIT, UM, Cornell, CMU, UT Austin and others, (gah, missed out Caltech)

Understand that I have no funding, have never been published, am not considered even a hobbyist scientist, and probably won't be able to write this up well enough to get your attention. I DO have the ambition to try though. That probably means you'll regret reading further... you've been warned. (bajeebus, that's what the justice of the peace said to me!)
(Yes, we are aware that these ramblings are incriminating)

What I know about Finite State Machines and their many derivatives can be described as 'shady' at best. I do know enough to understand that those who understand them and the math involved are most likely from another galaxy. Oh, the basics are easy enough, but once  you get past that first page... phew! 0 to brain pain in less than 4 paragraphs.

If you consider the bulk of the rest of my posts (no I don't expect that actually happened) it's not seriously deranged to believe that I might ponder FSM mechanisms quite often. In view of 'human intelligence' I have pondered the smallest measure of intelligence and how to assemble that small measure into larger systems. I don't think it possible to do so without thinking of FSMs - even if you have no idea what one is. Given all that, a dichotomy seems to exist between human wetware and the FSM mechanisms we actually have access to. Human wetware is made of carbon and literally swimming in chemicals. Every cell of it is subject to a number of inputs which are not generally considered TMK. I am beginning to believe that we need to study carbon based FSM systems or model them more closely. 

(Nailed itttt !!! )

Yes, yes, I'll get to the point in a minute.

(Yes, there may be no point... sorry about that)

Now, there are FSM style systems which might attempt to function in the way that I'm thinking, but I do not believe that they will perform in the same ways. I'm talking about the difference between using a prius to test a race track and a McClaren F1 to test it. Design and material differences limit the efficacy of testing the manufactured 'wetware' like machines. That would mean that we need to simulate some test of such a system. This requires carbon based wetware to be inserted into the test system in some way. Looking around us we see several examples of such systems which we might be able to use... if we have enough hardware, coordination, and ambition to use them for such observational experimentation. To name a few:

People driving cars on a highway
People in a large city who are walking
People in a sport stadium
People in a shopping mall

The Key component is people. No, not Soilent Green. People have the wetware we need. If we can capture their behaviors in FSM type situations where 'intelligence' is not the major driving force behind their behavior, we should be able to draw generalized understandings of the myriad inputs and states of carbon based FSM sytems. In fact, the math should be similar - I think, but we introduce myriad and seemingly random input variations to each switch in the FSM. The idea being that there is carbon based chemically reactive wetware running the switch and controlling the state based on inputs which are not necessarily fixed or known. We can introduce energy levels to the machine at a cellular granularity (more or less)!!! think about it.

Ok, it's not perfect, wildly uncontrolled, and we can't measure all the inputs to any (never mind all) the switches in the machine but we CAN observe the effects the machine creates in response to generally understood inputs. Yes, at this point I agree with you. I probably have no idea what I'm talking about.

The point or thought here is to observe the 'information' moving through the machine rather than each element of the machine itself. It is possible that by trying to measure the exact level of the noise we lose the data. I mean that in the nicest of way, of course. Ok, yeah, I'm losing it too. The information in this system is the overall goal or purpose of the system under test. Search for 'how to study traffic patterns' and you'll find pages that talk about improving throughput ... yet none of them address the myriad (I like that word) inputs to the system that affect its efficacy, and only rely on bandwidth studies without really appreciating the data being put through as it were. If this is sounding like binary feng shui, you might be right. Using some kind of feng shui on traffic throughput might address the myriad inputs that are typically ignored. Where is the study of disruptive billboard ads on traffic patterns.

Wow, that was lots of rambling. So what can that mean to my other thoughts? Ah yes... back to intelligence. Random inputs to an understood system will produce random and not understood results. Despite that, when a generalized system can be categorized, we should be able to formulate understandings of information flow and more importantly, information usage, even in environments of randomized and not understood inputs. Gestalt is a good word here.

Example of how this works: how do you decide which of the 240 channels on your cable system to watch when you have come in from the cold and smell something that reminds you of your grandmother's cookies? How much influence does 'cold' have on the system? How much influence does the smell of a familiar event have on the system? What weight do they place on meta rules in the system? These are the problems which have to be addressed in order to understand what intelligence is and what it is NOT. Yes, an autonomous intelligent system is subject to this kind of non-random randomness. Every input of information is weighed against stored meta rules and other information. So when it's cold and you smell your grandmother's cookies, what channel would you watch?

No comments: