Collective Adaptive Systems: Challenges Beyond Evolvability

Collecti v e Adapti ve Systems: Challenges Be yond Ev o lv ability 1 Serge Kernbach, 2 Thomas Schmickl, 3 Jon Ti mmis 1 Institute of Parallel and Distributed Systems, Un iv ersity of Stuttgart, Germany , serg e.k ernbac h@ipvs.uni-stuttgart.de 2 Artificial Life Lab of the Department for Zoology , Univ ersity of Graz, Austria, thomas.schmic kl@uni-gra z.at 3 Department of Electronics and Department of Computer Science, Univ ersity of Y ork, UK, jtimmis@cs.york.ac.uk Abstract — 1 This position paper ov erviews sev eral challenges of collectiv e adaptiv e systems, which are beyond the resear ch objectiv es of current top-projects in ICT , and especially in FET , initiatives. The attention is paid not only to ch all en ges and new re search t opics, but also to thei r impact and potential breakthroughs in i nform ation and communication technologies. I . I N T RO D U C T I O N Collectiv e systems ar e on e of th e most largest classes in na- ture (Bonabeau et al. , 1999) and technics (K ernbac h, 2008): insect colonies, b ird flocks, fish shoals, an imal h erds, hu man crowd, c ars o n streets, computers in intern et, cellula r ph ones, molecules in bio-syntectic sy stems and m any other examples. All these domain are ”collective systems”: social, networked, swarm, collaborative, co lloidal, nano and others, howev er all of them ind icate the same essential property: elements pr ovide ”more” func tionality when they are causally coupled . The value of th is ”more” depend s o n technolog y: many com- munities w orld-wide , a nd especially in Eu rope, are working on various aspects of co llectiv e systems. Th ese com munities includes adaptive an d bio -inspired sy stems, evolv able and reconfigu rable hardware, biologica l and bio -syntectic systems, software-intensive an d distributed systems as well as various branch es o f networks and network-b ased ap proach es. Joining effort o f se veral such commu nities, as e.g. collective and adaptive, allows fo rmulating comm on problems, solutions and challenges as well as essentially increase impact in the field of inform ation and com munication technolog ies. Currently , technolog ical progre ss enables a new kind of real-world systems: self-replicating (like bacteria) (Martel et al. , 200 9), huge number (nano- and micro- areas) (Balzani et al. , 2003), with a high developmental plasticity (bio-ch emical an d micro- modular systems) (Jaramillo, 200 9), with m any self- (healing, maintaining , programming , developing) and so on. Development in networked and software-in tensiv e systems (W irsing et al. , 2008), like mob ile telephony , demonstra ted a great achievement in terms of flexibility and scalability . One of fun damental r esearch questions in these systems is re lated to ca pabilities of their adap tability and self-determin ing beh avior in real ha rd-to-work environments, like u nderwater, in space or in ha zardous situa tions. T hese 1 Appeared in the workshop “Fundamental s of Collecti v e Adapti ve Sys- tems”, European Commission, 3-4 No ve mber , 2009, Brussels. en vironm ents are o f hu ge practical relev ance and possess essential mar ket potential, espe cially fo r oil-fr ee econo mics and tomorr ow’ s g reen tec hnolog ies. State of th e art in th e research o f collecti ve adap tiv e sys- tems (Kernbach, 2011) is very large: it includes bio-in spired and self-organizing branches, e v olutionar y and adapti ve- control strategies, different software an d hard ware approaches. Approx imating a research pr ogress in different top- projects in EU, the point of adap ti ve systems is focu ses on makin g co l- lectiv e systems cogn itiv e, co operative, ev olvable (bo th fitness- and concep t-driven), s elf-organiz ing, with a hig h dev elopme n- tal plasticity as well as exploring their em erging properties. Howe ver , expandin g this research mainstream further, after today’ s start-o f-the-ar t, we are basically m issing a few princ i- pal cr oss-domain elements, which are not related to a specific technolog y , see Fig. 1. The main qu estions are: Collective Systems New Paradigms for Collective Adaptive Systems Purposeful artificial self-organization Evolutionary self-organization Evolvability Driven forces behind evolution Complexity of „natural chemistry“ Artificial sociality Emergence of Self- Controllability of long-term autonomy Developmental plasticity Developmental drift Stability , scalability , reliability , optimization Long-term homeostasis, self-identification local rules state of the art challenges beyond evovlability concept, fitness technology computation Adaptive Systems + Fig. 1. Sketc h of seve ral challeng es be yon d the state-of-the-art of curr ent t op r esear c h pr ojects ar ound collective adaptive systems. • What is beyond adaptability , ev olve-ability and em er- gence of behavior? • What are the d riving forces of long -term developmental processes? • Are lon g-term developmental processes still controllable? Is ev olutionary self-organization still contro llable? • Is th ere a ny developmental dr ift du e to emergence of artificial sociality and self-reco gnition? • Are there ar tificial stru ctural elements, wh ich ar e ”ab- solutely plastic” in the developmental sense, such as biological amino acids? • Is a “n atural chemistry” ( =high co mplexity of ev olu- tionary processes) importan t for ada ptability and self- development? • Is there an “artificial chemistry” th at has the ab ility to to adapt software in-situ ? • Does artificial homeostasis and rules of eco logical sur- viv al lead to self-identificatio n and to emergence of self-? I. Controllability of long -term self-developmental pro- cesses. The issues of a long-term controllability of au- tonomo us artificial systems is extremely impo rtant. Ar ti- ficial adaptive systems with a high degree of plasticity (Levi & Kernbach, 2010) demonstrate a developmental d rift. There are many reasons for this, like lon g-term developmental indepen dency and a utonom ous b ehavior , emergen ce of arti- ficial sociality , mechan isms o f e volutionary self-organization (which are also a huge challeng e) and so on. Su ch sy stems a re very flexible and ad aptive, b ut they also massi vely inc rease own degrees of f reedom. New ch allenges in th is ar ea are re- lated to a long-term controllability and predictability of ”self-”, principles of m aking plastic pu rposefu l systems, p redictability of a structural d ev elopmen t and g oal-orien ted self- developing self-organization . These cha llenges h av e a great im pact o n a human c ommun ity in gene ral (the ”terminator” scenario ) as well as in dif ferent areas o f em bodied e volution, like synthetic biology or ev olvable/reconfigerab le systems and network s. II. Complexity o f ”natural chemistry”. Coupled with natural chemistry is the development o f novel artificial chemistries th at hav e the ability to re-write may be the op- erating system, or contro l system in which it is em bodied . In biological systems, th e chemical machin ery of an organ ism is data and pro cessor simultaneously , thus provid ing very complex interac tion n etwork but also powerful comp utational computatio n. The und erstanding of su ch networks is very importan t as well as the genera tion of building rules, that allo w to build such systems in an engineering way . System dynam ics and evolutionary appro aches from the field of Artificial Life could to dev elop such systems more easily . Basically , dev elop- ing such ch emistries is non -trivial: many su ch chemistries at the m oment suffer from scaling issues, syntactic and semantic problem s and gener al flexibility . The in clusion of such a chemistry r aises its own problems, and indeed would affect to the ev o lvability and stability of the system, a s what is being ev olved is also in con stant flux. III. Artificial socia lity . No t only chemical networks become complex q uite q uickly . Also in teraction networks which ar ise in social systems can easily get so complex that the cause-and- effect chains are hidden by the overwhelming netw ork of side- effects an d indirect causations. Thus studying such systems, and in parallel, developing the tools need ed to un derstand these systems is an important go al. On the one han d, mo dern technical networks are still n ot ”autonomous” and ”scal- able” en ough to be s atisfying, thus inv estigating compar ably complex interac tion networks in nature (e.g., social insect colonies) will provide new mechanisms and novel insights, that will h elp to u nderstand the emerging complexity in modern- world systems. In parallel, creating such systems fr om scratch (artificial evolution) or for m ”building blocks” o f co mplexity is also a very pr omising approach , as long as the pro ducts of these ”function ality gen erators” are really investigated and analyzed. Wit hout un derstanding the ”wh y?” and the ”how?” in the ev olu tionary pathway , these approach es ar e providing just snapshots and no gen eralizable insigh ts. IV . Emergence and controllability of Self-. Different computatio nal proc esses, lead ing to a global optimizatio n, scalability and reliability of co llectiv e systems, create a home- ostatic regulation. Homeostasis, as we ll as ar tificial h ormon al regulation, are imp ortant and ch allenging m echanisms in co l- lectiv e adaptive systems. Mor eover , conditions of ecolo gical surviving, imp osed on such systems, lead to a discrimin ation between ”self” and ”non-self” as well as to emergen ce of different self-p henom ena (den oted as ”self-”): self- replication, self-development, self- recovering and other self-. There are se veral great challen ges, like understand ing these m echanisms or long -term predic tability (see above) which have a large impact on the areas of artificial intelligence and intelligent systems as well as create a ne w paradigm for adaptiv e and self- developmental systems. An addition al c hallenge is to be able to “en gineer emergen ce” (Stepney et a l. , 2006). W e envisage systems that are highly ev olvable, will adapt themselves over long per iods of time, and p resent em ergent p roperties: tod ays engineer ing appr oaches simp ly can not address such a chal- lenge. W ays o f contr olling emergence in systems, such that we know at least what they wont do is essential to constructing systems that m ight be used in a daily and nor mal en vironme nt. The mentione d challeng es are only a few very importan t ones in the area of collective adaptive systems. They a re related to a large community in Europe and ha ve a great impact in tomorrow’ s ICT field. R E F E R E N C E S Balzani, V ., V etturi, M., & Cr edi, A. 2003. Molecular Devices and Mach ines. A J ourne y into Nanoworld . W ei nhaim, German y: W iley-VCH. Bonabeau, E., Dorigo, M., & Theraulaz, G. 1999. Swarm intelli genc e: fr om natural to artificial systems . New Y ork: Oxford Univ ersity Press. Jaramillo, Alfonso. 2009. Model-Based Design in Synthetic Biology . Chapman & Hall/CRC. Ke rnbach, Serge. 2008. Structur al Self-org anization in Multi-Agents and Multi-Robotic Systems . Logos V erlag, Berlin. Ke rnbach, Serge (ed). 2011. Handboo k of Collective Robotics: Fundamentals and Challenges . P an Stanford Publishing. Lev i, Paul, & Kernbach , Serge (eds). 2010. Symbiotic Multi- Robot Orga nisms: Reliability , Adaptability , Evolution . Springer V erlag. Martel, Sylvain, Mohamma di, Mahmood, F elfoul, Ouajdi, Lu, Zhao, & Poup onneau, Pierre. 2 009. F lagellated M agnetotactic Bacteria as Controlled MRI-trackable Propulsion and S teering Systems for Medical Nanorobots Operating in the Human Microv ascu- lature. Int. J. Rob . Res. , 28 (4), 571–582. Stepney , Susan, Polack, Fiona, & Turner , Heather . 2006. Engineer - ing Emergen ce. P ag es 89–97 of: ICECCS 2006: 11th IEEE International Confer ence on Engineering of Complex Computer Systems, Stanford , CA, USA, Augu st 2006 . IEEE. W irsing, Martin, Ban ˆ atre, Jean-Pierre, H ¨ olzl, Matthias M., & Rauschmayer , Axel (eds). 2008. Softwar e-Intensive Systems and New Computing P aradigms - Challenges and V isions . Lecture Notes i n Computer Science, vol. 5380. Springer .