Lifelike Systems Need Some Kind of a Skin: First Thoughts on Cyberskin Capabilities Kirstie L. Bellman1 , and Sven Tomforde2 1 Topcy House Consulting, Thousand Oaks, USA 2 Christian-Albrechts-Universität zu Kiel, Intelligent Systems, Kiel, Germany st@informatik.uni.kiel.de Abstract biological inspiration (Section 2) and map this onto a first concept for a ’Cyberskin’ in ICT systems (Section 3). The Robustness is one of the primary goals of self-adaptive and paper closes with an outlook on a possible path towards the self-organizing systems, including appropriate reactions to unanticipated conditions, to disturbances, and even to attacks. development of such a ’Cyberskin’ (Section 4). In this paper, we draw from biological systems to introduce a ’Cyberskin’ as a concept for introducing some potentially Biological Inspiration new strategies for robustness in complex systems. We briefly summarize characteristics of the skin of animals and humans ICT systems are approaching the complexity of biological and map this into desirable computational capabilities. We systems in their massively large number of parallel compo- then discuss some beginning ideas for implementing some of the technical functionality to support such a Cyberskin. nents and subsystems, concurrent activities and goals, and very diverse hardware and software entities. One of the tri- umphs in computing has been message passing and the use Introduction of explicit models and knowledge in many forms. With a Openness and adaptivity are key enablers for next- few basic protocols this has permitted enormous benefits generation information and communication technology in communicating among diverse systems. However, al- (ICT). Driven by trends such as intelligent systems (e.g., though biological systems use message passing they also Calma et al. (2017)), cyber-physical systems (e.g., Rajku- make use of other different ways of being architected and mar et al. (2010)), and Internet of Things (e.g., Atzori et al. integrated. These architectures are characterized by having (2010)), we face constellations of typically distributed au- multiple layers and multiple tightly integrated subsystems tonomous subsystems that integrate dynamically into an that allow a tremendous amount of implicit information to overall system constellation and self-adapt their behavior. be conveyed among a widely distributed and massively par- One of the goals of self-* (e.g. self-organization, self- in- allel system. Unlike the hairs in the ear or the visual retina, tegration, or self-healing) ICT systems is that these adap- sensory systems such as the skin or the architecture of joint, tations should not be limited to a pre-defined operational muscle and tendons are connected and integrated with both context or a fixed repertoire of behaviors at design time. In implicit and explicit ‘signals’ and actions. In addition to the addition, we emphasize that self-* mechanisms are also crit- continual reliance on networks, messages and explicit sig- ically needed to make technical systems resistant against ex- nals, we want to explore what qualities this implicit struc- ternal or internal disturbances (see Tomforde et al. (2018)). turing gives to computational systems. We suggest that there Such self-* systems may not perform better than conven- may be new characteristics in terms of response speed, adap- tional systems (although depending on the novelty of the tations, robustness, and protection possible. dynamic environment they may) but they return faster to a As background to the discussion on skin, let us look at a corridor of acceptable performance in the presence of dis- few of the different principles in biological systems and their turbances. The ultimate goal of self-* systems is to be both implications for computational systems. Bellman and Wal- more creatively adaptive and resilient against disturbances ter (1984) suggest that a good image for biological systems and attacks from the outside. We call this property ’robust- may be that of a community rather than a single agent. That ness’ (Müller-Schloer and Tomforde (2017)). is, the system is composed of elements that are in themselves In this paper, we propose a set of mechanisms which we sophisticated living elements. As Micheva and Smith (2007) call a “Cyberskin” that will consolidate and augment exist- state “One synapse, by itself, is more like a micro- proces- ing self-* capabilities. Inspired by biology, such a standard- sor - with both memory–storage and information- processing ized enclosure will allow for even higher robustness and re- elements – than a mere on/off switch. In fact, one synapse silience when acting in open environments. We discuss the may contain on order of 1000 molecular scale switches”. In Copyright c 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). our nervous system, it is estimated that there are 100 billion from response and provide time for adaptive processes and neurons, each with 7000 synapses. And that is only one type later reasoning processes. It is one of the major methods for of cell within the human body. Hence, the scale and hetero- ’buying time’ for adaptive rather than fixed responses. geneity of components, subsystems and processes fits very Skin is the largest organ in your body and compared to much into the idea of today’s Interwoven systems (see Bell- the other sensory systems, it does not get enough attention. man et al. (2014) and Tomforde et al. (2014)), which also Although only a subset of the skin’s functions will immedi- have enormous diversity; systems with their own functions ately resonate with our computational goals, we are listing and goals connected in a variety of ways and yet also inter- many of skin’s diverse functions in order to convey all the dependent and requiring coordination. processes and activities that go beyond being a barrier. Rather than being focused on a single goal or function, One of the major functions of skin is protection; it is an biological systems have very different ’efficiencies’ and op- anatomical barrier from pathogens and damage between the timization criteria that include adaptability, robustness and internal and external environment. As such, it is key to heat resiliency as well as performance, see Kantert et al. (2017). regulation and the control of evaporation. Dilated blood ves- Also, they have an enormous variety of subsystems and pro- sels increase perfusion and heat loss, while constricted ves- cesses occurring simultaneously at different levels. For ex- sels greatly reduce cutaneous blood flow and conserve heat. ample, mammalian brains consist of different regions with In the control of evaporation, it ensures that the skin pro- distinct architectures, characteristics, and capabilities. This vides a relatively dry and semi-impermeable barrier to fluid means at any given moment, there may be very different ac- loss. Skin is also a key part of the adaptive immune system tivities and processes and goals for those processes. (Langerhans cells) and also is one of the body’s storage and Biological systems are always multi-goaled (Bellman and synthesis sites; it acts as a storage center for lipids and wa- Walter (1984)), and they quickly drop goals as needed, take ter, as well as a means of synthesis of Vitamin D by action advantage of situations to adopt new goals, merge goals, and of UV on certain parts of the skin. take a long term view in optimizing their situation versus Given the functions above, it is not surprising to find that any one task. Such long term optimization includes placing it plays a major role in excretion and absorption. It is good oneself in a better position for later behavior and, hence, to keep these functions in mind as we consider ideas for con- might actually result in a temporarily lower goal accom- trolling what goes into systems (desirable entities e.g., data, plishment for a given task. Biological systems make use of code, sensors etc.) and what is released from computational ‘substitutability’ (equivalent classes of actions) and compro- systems (e.g., garbage, suspicious code or data). The skin mises among activities and goals. Like computer networks is an important site of transport in many organisms: oxy- that may have resource contentions, multiple users and con- gen, nutrients, or water. In humans, the cells comprising the flicting goals among those users, biological networks have a outermost 0.25-0.40 mm of the skin are supplied by oxygen. magnitude more diverse types of processes, goals, activities The role of the skin in sensation is of immediate inter- and subsequently, contentions. To handle the needed coor- est for our computational systems. Skin contains a variety dination, biology has developed many more styles of inter- of nerve endings that react to heat and cold, touch, pres- action and influence that includes both explicit and implicit sure, vibration, and tissue injury. The skin includes sen- networks, architectures, and responses to activities. In com- sors and processing for touch (several types including pres- plex systems, one needs both the speed coming from rapid sure and vibration,) temperature, proprioception (body po- communication and signaling, as well as the structure that sition, position of muscles, bones and joints), and nocicep- allows the system to take advantage of implicit connections tion (pain/tissue injury). The system reacts to diverse stim- that can quickly coordinate and inform a wide variety of dis- uli using different receptors: thermo-receptors, nociceptors, tributed elements. Although there are many such examples mechano-receptors and chemo-receptors. The skin is also in biological systems, in this paper we focus on the proper- important in aesthetics and communication; it is used by ties of skin which has both explicit signaling and transmis- conspecifics to assess mood, physical state and attractive- sion of information and implicit and indirect methods. ness. Both these internal state and external environmental One of the key stabilizing principles of biological net- changes are rapidly available across the system because of works is the use of active ’stabilizing’ processes that include its intertwined and diverse architectures. different strategies for ’buffering.’ By buffering, we mean Furthermore, these numerous types of receptors are not mechanisms and architectures that separate input from out- evenly distributed across the skin. The density and variety put and isolate the impacts of abnormal/foreign elements of receptors matter. This density may relate to priorities or to (including the detection of abnormal behavior as suggested coordination with other sensory systems or to the specializa- in Gruhl et al. (2015)). This includes nature’s version of tions of the system to its ecosystem. For example, sensing firewalls (e.g., skin, membranes) which are permeable, con- injury is an important role of skin; there are 200 pain recep- trollable and, in some cases, sentient. Buffers in living sys- tors for every square centimeter of the human body. Cat- tems allow the separation of immediate reaction to stimuli fish have chemo-receptors all along their body (a catfish is like a giant tongue) Atema (1980) while humans have 6-12 has a major role in self-monitoring; it informs the system of million olfactory receptors in the human nose alone (blood- internal changes (for example, proprioception and changes hounds 4 billion) , see Bullock (1982). Electric eels have such as hydration) and by nature to its external and internal a sensory system (electrolocation) that combines some as- interfaces, helps integrate internal and external states. pects of skins and vision; the sensing of electric fields can Lastly, the ’elasticity’ of skin refers to the property of be used for object detection, communication, defensive and some solid materials that return to their original shape and offensive behavior such as prey stunning and warding off size after the forces deforming them have been removed. predators Shier et al. (2004). This variety may give us new This has been a key adaptive capability in biological sys- ideas for how to prioritize and fuse incoming information tems that grow and develop, but it also allows immediate and how to protect key areas. One thing we would like from flexibility in movements. It is also clear that the deforma- a Cyberskin is quick detection of the locality, the extent, and tion (stretch and bend of skins) is signaling changes of states the type of attack. Skin can immediately locate site of con- such as a full stomach or the extent of voluntary and invol- tact (within different just ’noticeable differences’): untary movements (e.g., breathing) and triggering additional actions. One of the things we would like from both rapid lo- Some Thoughts on a ’Cyberskin’ cality and elasticity is perhaps a measure of how long the Over the last twenty years, there has been increasing work interaction/attack has been ongoing. In biological skin, such on haptic interfaces and artificial skin. Robotics and sensors information is conveyed by the type of sensory event and applications such as telemedicine and manufacturing have by how deeply down through skin layers the interaction has resulted in an interest in artificial touch receptors and skin- gone. Many sensed events become pain (e.g., an alert to dan- like sensor meshes (for example Lumelsky et al. (2001)). ger) the farther down in the layers they go. Biological pain Our purpose here is different. Instead of building a skin per (extreme pressure, heat, cold, cuts) could have new corollar- se, we examine the implications of skin’s properties and con- ies in computational systems such as the location of a loss in sider how to adopt them into computational systems. There connectivity or a drop in data rate. are several properties of the skin that are of particular in- terest to us. First, skin is inside as well as on the external A ’Cyberskin’ for Technical Systems interface to the world; the importance to us is that poten- In this section, we introduce a way of going beyond biology tially the same methods and architectures may be used for and discuss how the system would actually construct an on- different sizes of systems and for system-of-systems. Also, going skin due to learning and experience. Biological skin there are many layers and types of skin in a given biologi- does this a little especially in terms of size (as one gets big- cal system. The external skin has a large number of layered ger), in terms of injury, and of course, the complex set of different sensor networks and activities. The overlapping adaptive sensory mechanisms. However, the ideas here in- and diverse architectures of these nets allows for simultane- clude intelligent processing and learning with some of the ous monitoring of multiple conditions of interest and layers aspects of skin to incorporate learning. reached have implicit information (e.g., injury or danger). The design process of traditional technical systems is One of the most important features of skin is that it is se- based on analysis of anticipated and predictable events, fol- lectively permeable, allowing for many different implemen- lowed by the development of countermeasures and con- tation strategies on what can pass through – both from the tingent responses (Tomforde and Müller-Schloer (2014)). system and into the system. Skin is also sentient in terms However, especially in dynamic and open environments, of ’having sense, perception’ – locality, nearness, size, dam- there will always be unpredictable events. This needs to be age, and so on. This is a key property in terms of having a covered by a concept called “spontaneous closure” Müller- distributed system capable of almost immediate processing Schloer and Tomforde (2017), see Figure 1). Exception han- where it is being interacted with and with what. The multi- dlers and diagnosis systems are examples of implementing tude of sensors noted above monitor for different things and spontaneous closures in technical systems. In a watchdog can be used to determine size, type of interaction, and site of timer, a timer is reset and restarted at regular intervals. If first responses to foreign objects. Given that, it is interesting the timer is not reset, because the monitored system is in a is to imagine new relevant measurands for non-biological malfunction, a reset is performed to a defined initial state. entities. That is, the current sensors in biological skin de- Research in self-adaptive and self-organizing (SASO) pend upon our animal evolution in many physical states systems has presented a wide variety of concepts for estab- (e.g., cold, hot) and threats (e.g., pushed, pained, pressured). lishing and persisting such spontaneous closure. Most of the These may still be of importance for cyber-physical systems machine learning mechanisms used in SASO systems are as many failures are caused by temperature and dampness. dedicated to finding appropriate reactions to previously un- However, what properties should we sense within a compu- known or less certain conditions D’Angelo et al. (2019) – tational environment? Noisiness? Density of activity? In ad- in this sense, learning means that recurring reactions of the dition to the external world, skin within biological systems spontaneous closure become part of the designed core. that need to be answered before an interaction is established. We observe that – at least for now – all the aspects men- tioned above are mostly considered in an isolated manner. The notion of a Cyberskin is especially important as a means to better integrate these efforts into a unified whole. Also the basic purpose of the Cyberskin is not restricted to the outer shell – it should also connect and protect the inner components or subsystems. Several efforts in SASO sys- Figure 1: A ’spontaneous closure’ to handle exceptions tem engineering have proposed machine learning solutions to handle Aspect 1 and Aspect 2. However, Aspect 3 has mostly been considered in an attack-specific manner (i.e., Based on the experience of machine learning applications, trying to identify attacks based on abnormal behavior) and we distinguish three major classes of effects that need to utilizes authentication and authorization mechanisms using be handled by spontaneous closure: a) Unanticipated con- standard encryption schemes. We propose to augment this ditions, b) disturbances as a result of malfunctions, and c) (i.e., to extend the spontaneous closure) with more sophisti- attacks. Unanticipated conditions reflect the fact that only a cated and lifelike mechanisms. Especially, we would like to fraction of environmental and internal conditions can be an- replace the current isolated handling by a localized, efficient ticipated at design time. Disturbances as result of malfunc- and approximate computing metaphor that we call Cyber- tions include expected events such as wear and aging, as well skin. There are approaches that go in this direction – such as unanticipated effects that cause failures and malfunctions as the idea of a ’membrane’ introduced in Diaconescu et al. of individual components and entire systems. The system (2016). This and related work certainly provide good start- should continue to maintain its functionality despite these ing points for developing more flexible and powerful spon- failures. Lastly, attacks are likely because an autonomous taneous closures and finally a strong and efficient Cyber- system operates in a shared environment where interaction skin. The basis for the implementation of this Cyberskin are with known and unknown other systems takes place. This paradigms such as the MAPE-k cycle (Kephart and Chess not only opens up potentials for continuous optimization but (2003)) or the Observer/Controller tandem (Tomforde et al. also a multitude of heterogeneous attack vectors. (2011), Tomforde et al. (2016)). Living systems are characterized by a powerful spon- taneous closure partly realized by the skin. In particu- Conclusion lar, we can observe the following technical implications as The trends in ICT such as openness, heterogeneity of inter- three aspects of Cyberskin. As noted above, skin can de- action partners, (hidden) mutual influences, and faster re- tect and classify novel conditions extremely fast. Techni- sponse times especially in human-machine interaction re- cally, this means in Aspect 1 of Cyberskin we want to apply quire a more ’life-like’ system design. We described our novelty detection mechanisms (see Pimentel et al. (2014); vision of a ’Cyberskin’ that better integrates already exist- Gruhl et al. (2020) for details) in combination with other ing technology (e.g., self-healing techniques, learning ca- autonomous learning capabilities (see, e.g., Krupitzer et al. pabilities) and contains novel safety and security means to (2015); Sick et al. (2018)) to improve the reaction to these dynamically realize a context-aware interaction protocol. stimuli over time. Next natural skin recovers quickly from It is also important that as a community we consider new injuries (within limits determined by the system) and raises types of non-message-passing architectures and how we can alerts through prioritized alert signals such as pain or pres- imitate the other manners of interactions found in biologi- sure. Aspect 2 of Cyberskin technically then refers to self- cal systems. The information conveyed by the deformation healing techniques, provisioning of backup or alternative of the skin, the immediate identification of locality, and the components, and alarm systems that are coupled to imme- interplay of different types of receptors are not conveyed by diate counter measures. Aspect 3 refers to the ability of networks alone in biology and we need creative concepts for natural skin to assess contact with novel objects and oper- how we could architect and model that. Cyberskin requires ational environments and immediately map this sensory in- new thinking about the meaning of what is monitored for formation to approximate reactions. Besides standard en- and the implementation of new measurands and methods. cryption and authentication or computational trust mecha- Future work is on implementing aspects of ’Cyberskin’ nisms (e.g., Kantert et al. (2015)), we think that skin may and seeing how generalizable those implementations can be- provide new inspiration for an interactive and context-based come. Looking across the commonalities among animal contact establishment protocol. This protocol would be one skins gives us the hope that there will key principles in im- way to allow for some of the skin’s most subtle, rapid, and plementing Cyberskin, such as permeability (intelligent gat- widespread capabilities in technical systems. This may be ing of traffic across membranes) or the criticality of having part of the means to process customized security questions different dedicated receptors. References Kephart, J. and Chess, D. (2003). The Vision of Autonomic Com- Atema, J. (1980). Chemical senses, chemical signals and feeding puting. IEEE Computer, 36(1):41–50. behavior in fishes. Fish Behaviour and its Use in the Capture Krupitzer, C., Roth, F. M., VanSyckel, S., Schiele, G., and Becker, and Culture of Fishes, pages 57–101. C. (2015). A survey on engineering approaches for self- adaptive systems. Pervasive and Mobile Computing, 17:184– Atzori, L., Iera, A., and Morabito, G. (2010). The internet of 206. things: A survey. Computer networks, 54(15):2787–2805. Lumelsky, V. J., Shur, M. S., and Wagner, S. (2001). Sensitive skin. Bellman, K. L., Tomforde, S., and Würtz, R. P. (2014). In- IEEE Sensors Journal, 1(1):41–51. terwoven systems: Self-improving systems integration. In Eighth IEEE International Conference on Self-Adaptive and Micheva, K. D. and Smith, S. J. (2007). Array tomography: a new Self-Organizing Systems Workshops, SASOW 2014, London, tool for imaging the molecular architecture and ultrastructure United Kingdom, September 8-12, 2014, pages 123–127. of neural circuits. Neuron, 55(1):25–36. Bellman, K. L. and Walter, D. O. (1984). Biological processing. American Journal of Physiology-Regulatory, Integrative and Müller-Schloer, C. and Tomforde, S. (2017). Organic Computing – Comparative Physiology, 246(6):R860–R867. Techncial Systems for Survival in the Real World. Autonomic Systems. Birkhäuser Verlag. ISBN: 978-3-319-68476-5. Bullock, T. H. (1982). Electroreception. Annual review of neuro- science, 5(1):121–170. Pimentel, M. A., Clifton, D. A., Clifton, L., and Tarassenko, L. (2014). A review of novelty detection. Signal Processing, Calma, A., Kottke, D., Sick, B., and Tomforde, S. (2017). Learning 99:215–249. to learn: Dynamic runtime exploitation of various knowledge sources and machine learning paradigms. In 2017 IEEE 2nd International Workshops on Foundations and Applications of Rajkumar, R., Lee, I., Sha, L., and Stankovic, J. (2010). Cyber- Self* Systems (FAS* W), pages 109–116. IEEE. physical systems: the next computing revolution. In Design Automation Conference, pages 731–736. IEEE. D’Angelo, M., Gerasimou, S., Ghahremani, S., Grohmann, J., Nunes, I., Pournaras, E., and Tomforde, S. (2019). On learn- Shier, D., Butler, J., and Lewis, R. (2004). Hole’s human anatomy ing in collective self-adaptive systems: state of practice and & physiology (lee, wt, lee, je, park, sh, lee. k, l., park, ka, & a 3d framework. In 2019 IEEE/ACM 14th International hong, sg). Seoul: Jung Dam Publishing Co.(Original work Symposium on Software Engineering for Adaptive and Self- published 2002). Managing Systems (SEAMS), pages 13–24. IEEE. Sick, B., Oeste-Reiß, S., Schmidt, A., Tomforde, S., and Zweig, Diaconescu, A., Frey, S., Müller-Schloer, C., Pitt, J., and Tom- K. A. (2018). Collaborative interactive learning. Informatik forde, S. (2016). Goal-oriented Holonics for Complex Sys- Spektrum, 41(1):52–55. tem (Self-)Integration: Concepts and Case Studies. In Proc. of the 10th IEEE Int. Conf. on Self-Adaptive and Self- Tomforde, S., Hähner, J., and Sick, B. (2014). Interwoven Systems. Organising Systems, pages 100–109. IEEE. Informatik-Spektrum, 37(5):483–487. Aktuelles Schlagwort. Gruhl, C., Sick, B., Wacker, A., Tomforde, S., and Hähner, J. (2015). A building block for awareness in technical systems: Tomforde, S., Kantert, J., Müller-Schloer, C., Bödelt, S., and Sick, Online novelty detection and reaction with an application in B. (2018). Comparing the Effects of Disturbances in Self- intrusion detection. In IEEE 7th Inte. Conf. on Awareness Adaptive Systems – A Generalised Approach for the Quan- Science and Technology, iCAST 2015, pages 194–200. tification of Robustness. Springer Transactions on Computa- tional Intelligence XXVIII, 28:193 – 220. Gruhl, C., Tomforde, S., and Sick, B. (2020). Novelty detection in continuously changing environments. Future Generation Tomforde, S. and Müller-Schloer, C. (2014). Incremental de- Computing Systems, Special Issue on ’Self-Improving Sys- sign of adaptive systems. J. Ambient Intell. Smart Environ., tems Integration’. 6(2):179–198. Kantert, J., Edenhofer, S., Tomforde, S., and Müller-Schloer, C. (2015). Representation of trust and reputation in self- Tomforde, S., Prothmann, H., Branke, J., Hähner, J., Mnif, M., managed computing systems. In 13th IEEE International Müller-Schloer, C., Richter, U., and Schmeck, H. (2011). Ob- Conference on Dependable, Autonomic and Secure Comput- servation and Control of Organic Systems. In Müller-Schloer, ing, DASC 2015, pages 1827–1834. C., Schmeck, H., and Ungerer, T., editors, Organic Comput- ing - A Paradigm Shift for Complex Systems, Autonomic Sys- Kantert, J., Tomforde, S., Müller-Schloer, C., Edenhofer, S., and tems, pages 325 – 338. Birkhäuser Verlag. Sick, B. (2017). Quantitative robustness - A generalised approach to compare the impact of disturbances in self- Tomforde, S., Rudolph, S., Bellman, K. L., and Würtz, R. P. (2016). organising systems. In Proceedings of the 9th International An organic computing perspective on self-improving system Conference on Agents and Artificial Intelligence, ICAART interweaving at runtime. In 2016 IEEE International Con- 2017, Volume 1, Porto, Portugal, February 24-26, 2017, ference on Autonomic Computing, ICAC 2016, Wuerzburg, pages 39–50. Germany, July 17-22, 2016, pages 276–284.