Carsten Knudsen, Rasmus Feldberg and Steen Rasmussen- Information Dynamics of Self-Programmable Matter

8/3/2019 Carsten Knudsen, Rasmus Feldberg and Steen Rasmussen- Information Dynamics of Self-Programmable Matter

1/24

L&U~ -91-896

LA-UR--9l-896DE91 009954

TITLE INFORMATIONDYNAMICSOF SELF-PROGRAMMABLE~TTER

AUTHOR(S). CARSTENKNUDSEN,RASKLJSFELDBERGandSTEEN RASMUSSEN

SUBMITTEDTO Proceedings of Ct~mplex Dynamics and BiologicalEvolution held AuSust 6-10, 1990 in DenmarkDISCLAIMER

This report wus prepnrcd t !snn account or w,~rk sponaorcd by an agency of the United !htcsGovernment, Neithcrtlte [lnitd States Govelnmcnt nornnyagcncy thcrcof, nt]r any of theiremplcryccs, makes any wurranty , cxprexa or imp~kd, or acaumcs any !cgal Iiuhil ity or rcxponsi .bility for the accuracy, complctcncac, or uac(ulnatof any information, npparatus, product, orproccxx dixcloacd, or reprcacnts that i ts uas would not infringe privately owned r,ghts . Rc(or -encc here in to any .qrcc it ic commercial product , prmcm or acrvwc by trade name, trademark,manuracturcr, or crthcrwiac dots not ncceuaardy con, tt tu tc or Imply IW cndorxcmcnl, rccom-mcndation, or favoring hy the United Staten Govcrm, wnt or any sgency Ihercd The viewsand opinions of aulhorx eaprcaccd herein do m)t ncc,wamrily siatc or reflect those or the(Jnitcd S!atea Government or any agency thcrwf,

8yaccop!an~o o!thlsar l!cla th@publ ,snaf rUqn!zoalhaltho US C30wmmOnt WlalnSanOfWCWaWQ ?oyalty.ffMl,con3a lo Dubl!?hoff@oroducolhepbhgn~d form n! IIIIS Conlf,butfoo O? to allow otl-tors to do ao. Ior US Govornmonl owboaosl r! . Los Alamos Nal ,onal L8boralofv fequ.als Ihat lfw pubhahor IrjwItIly Ihla af!Kl@alwOfI( oo?fofmed u,, ~eflhoauaplcos ollngu S Dopaf!monlol Eno,gv

anrilos LosAlamos National LaboratoryLosAlamos,New Mexico 87545


2/24

IXFORMAmON DYNAM:3 OF SELF-PROGRAMMABLE MATTER

Carsten Knudsen,t R=mus Feldberg,t and Steen RasmussentCenter for Nonlinear Studies andPhysicsbboratory In and

Center for !kfodelling, Complex Systems Group,Nonlinear Dynamics and Theoretical DivisionIrreversible Thermodynamics MS B258Technical University of Denmark Los A.lamesNational Lakorato~DK-2800Lyngby Los Aiamos, New Mexico 87545Denmark USA

ABSTR4CFUsing the simple obsemation that programs are identical to data programs alter

data, and thus programs alter programs, we have constructed a self-programmingsystembased on a parallel von Neumann architecture. This system has the same fundamentalproperty as living systems have: the ability to evolve new properties, We demonstratehow this constmctive dynamical system is able to develop complex cooperativestntctures with adaptive responses to external perturbations. The experiments with thissystem are discussed with special emphasis on the relation between informationtheoretical measures (entropy and mutual information functions) and on the emergenceof complex functional properties. Decay and scaiing of long-rmge correlations arestudied by calculation of mutual information functions.

INTROfXJCIIONA fundamental feature of living organisms is the ability to compute, or p oces!,information. Information processing takes place over Awide scale of complexity,ranging

from the simple processes by which an enzyme recognizes a particular substratemolecule, to complicated feedback regulations contairdng many different levels ofinformation processing, to the extremely complex processes of the human brain,

An example of biological information processing in a fee+back loop is providedby one of the negativt feedback loops described by Sturis et al, (1991) in their modelof o$cillsto~ insulin release. The feedback control can here be divided into at leastfour different components: (i) an increased amount of Ulucosein the plasma stimulates


3/24

insulin production in the pancreu and secretion of insulin into the plasma; (ii) fromthe plasm% ir.wlin diffuses into the interstitial fluid; (iii) here insulin molecules a~tachto receptors on the surface of the cell; and (iv) insulin activated receptors enhance theuptake of glucose by the cells, which, of course, implies a decrease in glucose outsidethe cc]k.

This loop involves simple biochemical information processing such as therecognition of receptors by insulin molecules and the subsequent attachment of themolecules. In addition, there is a more complex information process involving actitetransport of glucose over the cell membrane facilitated by a cacade of confirmationalchanges in the cell membraues protein molecules.

The most complex kind of biological information processing is probably (heabstract and creative symbolic information processing in the human brain. Simpleaspects of these processes are subjects of numerous investigations, In particular avariety of models of artificial neural networks have recently been proposed (see forexample Palmer 1988, and Touic&ky 1990).

The theoretical foundation of information processing in man-made machines canbc described in terms of computation theory (Hopcroft and Unman 1979), In computa-tion ;.,:oty, a number of different formalisms exist, of which the Turing machine (TYf)for historical re~orts is the mcst well. known. The Turing machine has been examinedthoroughly by mathematicians and computer scientists because it is believed to be ableto perform the most general type of computation, universal computation, Thisconjecture is known u the Church.Tunng thesis (Hopcroft and Unman 1979),

Besides Turing machines, several other systems have been shown to supportuniversal computation, including cellular automat% the A-calculus, Post systems, ~hehard billiard ball computer, general recursive functions, clwifier syst~ms, partialdifferential equations, von Neumann machines, and even ~ maps of the unit ~quarconto itself. Ithas been shown that each of these formalisms is equivalent, since any oneof them can simulate any other,

The information processing found in biological systems seems to be different innature from that of a Turing machine. In facg none of the above mentionedcomputational paradigms capture the full spectrum of bi~rnolecular informationprocessing.A fundamental property of computation {nbimolecular systemsarises fromtl,eir ability to alter or program themselves. Self.progrmt.rnirtgoccurs at all times andlength scales in bimolecular systems,&though the above mentioned computational$ystcrnsin principlo can program themselves, this capacity hm never been studied orused. It is knowmthat any of the universal systems have the foilowing properties: (1)Ihe ability to store information, (11)the ability to communicate information, and (III)the ability to perform non-trivial information processing,

The$e abilities arc also found in the living cell, aithough they we more difficultto cl~si$, Howrver, using the same scheme to discuss elements of biomoiccularcomputatio~ we obtain:


4/24

(T)Storage and memory abilities: (1) singlemolecules, e,g,DNA and proteins, and(2) assembly/disassemblyofsuprarnolecular structures, e.g. the cytoskcleton and the ceilmembrane.

(11) Signal abilities: (1) diffusion (passive transport of materials, energy, andinformation) occurs everywhere in the cell, (2) active transport (non-specific(convection) and specific transport of materials, energy, and information) convectionoccurs in the cytoplasm and specific transport occurs for example over the cellmembrane and along the microtubules, (3) confirmational changes (transfer of energyand information), e.g. of dynein and kinesin in relation to cilia mobility, and (4)electromagnetic irradiation (transfer of energy and information), e.g. photochemicalprocesses in chlorophyll.

(HI) Transformation abilities: (1) chemical reactions - often using signalmoleculestu reactants to produce new signal molecules as products or using sigtwk which act 3Scatalysts or triggers, and (2) transcription of DNA to RNA and translation of RNA intoprotein molecules which fold up and act as constructive and regulatory units in the cell,

From this scheme it should be obvious that most fundamental bimolecularprocesses can be interpreted in terms of computation, These bimolecular processesare all coupled through a very complex network of functional interactions about whichwe oniy know certain details and in which the overall bauplan is still a mystery, Thecell continuouslyprograms and re.prograrns itself, and in multicellular organisms thisself-programming alsb occurs at the organism level (recall the discussion of thefeedback loop controlling insulin relew).

Livhg systems can through a re.programrning of some of their parts alterfunctional properties which are of vital importance for survival. Viewed over Icngertime scales this seif-prograrnming ability is also used to create new properties which areincorporated through the selection process of evolution. Since any comput~t;onaluniversal system in principle, is able to program itself, wc shall modifyone of them rothat we can study self-programming as a phenor non in a much simpler arid moretractable system We have chosen to modify the parailel von Neumann architecture,The modified von Neumann machine (MVNM) is easy to program since most moderndigital computers are based on the von Neumann principle, and since the autonomousdynarnia of such a system even at its lowest level (one single instruction) has 4 clearcomputational interpretation, We shall in the following focmson the emergence of newfunctional properties in MVNMSwhich most clearly reflect the evolutionary aspect ofbiocomputing.

SELF-PROGRAMMABLE MATTERWe can in genera] terms define self.prcgrnmmable matter as a dynamical system

of functional interacting elements, or compositions of elements, which through


5/24

..

.. autonomous dynamics can develop new compositions of functionally active elerrlen:s.Such systems are characterized by an ability to constmct novel elements wi!hinthemselves. Thereby chemistry by definition becomes a particular kind of self-prograrrtmab]e matter. The physical properties (e.g. shape and charge) of the chemic~lspecies define the possible interactions with other molecules and thereby theirfunctional propenies. Chemical systems create new properties through recombinationof molecules via chemical bonds. New combinations between existing molecules andcombinations of new molecules with other molecules then define new functionalproperties, This defines a constructive or seif-programming loop given by:n lolecules + p}l~s ical propen ies + j i.mctional propenies + interartiorts - new molecu les

Documents

Carsten Knudsen, Rasmus Feldberg and Steen Rasmussen- Information Dynamics of Self-Programmable Matter