Hybrid societies: challenges and Perspectives in the Design of collective Behavior in self-organizing systems

Hybrid societies are self-organizing, collective systems, which are composed of different components, for example, natural and artiicial parts (bio-hybrid) or human beings inter-acting with and through technical systems (socio-technical). Many different disciplines investigate methods and systems closely related to the design of hybrid societies. A stronger collaboration between these disciplines could allow for re-use of methods and create signiicant synergies. We identify three main areas of challenges in the design of self-organizing hybrid societies. First, we identify the formalization challenge. There is an urgent need for a generic model that allows a description and comparison of collective hybrid societies. Second, we identify the system design challenge. Starting from the formal speciication of the system, we need to develop an integrated design process. Third, we identify the challenge of interdisciplinarity. Current research on self-organizing hybrid societies stretches over many different ields and hence requires the re-use and synthesis of methods at intersections between disciplines. We then conclude by presenting our perspective for future approaches with high potential in this area.

his paper originates from a small international workshop on "Methods for Self-Organizing Distributed Systems" that was held in Laubusch, Germany, during October 2015.We name several challenges and give our perspectives for the ield of hybrid societies [cf.Eiben (2014) and Prokopenko (2014)].In general, hybrid societies are made of diferent components instead of having a homogeneous identity.We call them "societies" because the components possess individual agency and interact persistently.Such societies can be comprised both natural and artiicial agents (Baxter and Sommerville, 2010;Halloy et al., 2013;Schmickl et al., 2013;Hamann et al., 2015) or diferent types of artiicial agents only (Dorigo et al., 2013).We focus on self-organizing collective hybrid societies that are characterized by multiple interactions of agents, positive and negative feedback processes, and luctuations (Ashby, 1947;Bonabeau et al., 1999;Camazine et al., 2001;Omicini and Viroli, 2011;Heylighen, 2016).Oten, these systems show collective behavior indicated by the emergence of global spatial and/or temporal patterns (Serugendo et al., 2006;Attanasi et al., 2014;Popkin, 2016).Furthermore, hybrid societies are describable on a microscopic level, the level of an individual agent, and a macroscopic level, the level of the whole society (Schelling, 1978;Alexander et al., 1987;Schillo et al., 2000;Hamann et al., 2014).We want to design and determine the artiicial part of these systems, although the artiicial subpopulation is in contact with a natural subpopulation in hybrid societies.
Typical examples of hybrid societies are investigated in the project ASSISI|bf (Schmickl et al., 2013) where robots closely interact either with groups of bees or ish.Such systems require diferent approaches than those developed for multi-agent systems because they are heterogeneous and while the robots are variably programmable the biological agents (bees and ish) have a determined behavior.he ASSISI|bf system heavily relies on social aspects because the robots need to learn the "social language" (Schmickl et al., 2013) of bees/ish to trigger desired behaviors.It is a hybrid system because the robot-animal interaction is not only in one way but also the animals determine the system's further development.Similarly, we have hybrid societies in socio-technical systems where human beings closely interact with technological artifacts (Baxter and Sommerville, 2010;Smirnov et al., 2014;D'Orsogna and Perc, 2015;Helbing et al., 2015).
We identify three common, primary challenges in the design of hybrid societies (see Figure 1).Each is discussed in detail, supplemented by additional secondary challenges, and we give our perspective on future approaches with high potential.

PriMArY cHALLeNGe A:
FOrMALiZAtiON OF HYBriD sOcieties he analysis of hybrid societies using tools of mathematics and computer science is essential to gain deep insights into the dynamics and prominent principles of hybrid systems.Besides allowing for predictions, the formal approach also guides one's thoughts when designing hybrid societies.he formalization of hybrid societies is the precondition to move from formal speciications to an integrated design process.

Purpose of Formalization
From our experience in work with collective hybrid societies, we have the strong belief that our ield of research requires a tremendous efort to develop a generic model.Hence, a grand challenge of the design of collective behavior in hybrid societies is to develop an appropriate generic formalization.A truly generic formal model would overcome the diversity of methods and models in the ield.
If not completely generic, we would at least require a methodology that allows to model a large range of diferent collective hybrid societies.he purpose of a generic model is to understand the desired system and to gain deep insights.Formalization is necessary to achieve a good understanding of a system's inner dynamics and, if possible, to predict its outcome.With the optimal model, we could predict future behaviors and efects of hybrid societies.Such a model would permit to analyze a wide variety of collective systems, enable rigorous mathematical comparisons, and help to understand potential problems in system design before realization in simulation, and hardware was achieved.

requirements and Actions for a Formalization
he formalization approach should be generic and applicable in many domains sharing essential system features.he development of such modeling techniques requires, however, to unify methods, concepts, and deinitions from many diferent ields.It requires a high degree of integration, knowledge about each of these domains, and a high convertibility of the model.First steps toward a uniied methodology have been made, for example, in the ields of socio-technical systems (Baxter and Sommerville, 2010;Jones et al., 2013;Schöttl and Lindemann, 2015) and swarm robotics (Lerman et al., 2005;Brambilla et al., 2013).Models originating from natural sciences are limited in representing typical abilities of agents and also modeling the emergence of self-organizing artifacts is challenging (see Sec. 2.6).A generic framework relecting domain-speciic characteristics while accurately capturing the evolution and dynamics of collective behavior, both on the micro-and macroscopic level, needs to be established.
he total amount of modeling and investigation techniques for homogeneous and heterogeneous collective systems is huge and spans ields such as collective animal behavior, statistical physics, network theory, control theory, opinion dynamics, and diverse subields of computer science.In order to give a little, incomplete overview, we cite only a few of these, see Table 1.Despite the strict column-wise presentation of methods, there exist already approaches that combine several methods from diferent ields, such as the combination of game theory with networks (Perc and Szolnoki, 2010), percolation and networks (Piraveenan et al., 2013a), and hybrid systems with temporal networks (Boerkoel and Durfee, 2013).Furthermore, the ield of evolutionary game theory investigates hybrid societies, especially the interaction of agents also with reference to collective behavior and self-organization (Perc and Szolnoki, 2010;Perc and Grigolini, 2013).However, the developed models oten abstract away proximate mechanisms, that is, the behavioral rules that generate the spatio-temporal dynamics of collective systems (André, 2014).Partially due to the extreme diversity of methods, it is diicult to compare hybrid societies or their models.A generic, formal modeling approach of collective hybrid societies would help to overcome that problem.

secondary challenge: system complexity
Another challenge is the complexity of hybrid societies due to self-organization that contains by deinition a multitude of locally interacting agents.Local interactions between agents create dynamic environments, which are complex to model. he agents operate locally but can trigger emergent global patterns; we have diferent types of agents, and they oten live in dynamic environments, which are challenging to model.For example, a diiculty speciic to self-organization is to link the model that describes the global behavior of the system to the model that describes the behavior of the individuals.Deining the so-called micro-macro link is a fundamental issue in both directions (Schelling, 1978;Hamann and Wörn, 2008).Macro-to-micro means that a certain global behavior is required; however, the respective individual behaviors are unknown.Micro-to-macro is the challenge of predicting the macro-behavior for a given micro-behavior.Particular internal states of these agents may be essential, e.g., the internal energy levels are crucial especially in lying agents (e.g., quadrocopters) or forest ecosystems (Zamuda and Brest, 2013). he formal approach has to address these internal states and model their dynamics.Local and global correlations between these internal states add another challenge.
In summary, we have the dynamics of the internal states and local interactions of individual agents on the one side and the overall dynamics of the global system on the other side.he challenge is to ind the link between these two sides, which is key to understand and formalize hybrid societies.

shortcomings in the state of the Art
he vast number of methods of hybrid societies comes with individual shortcomings.We discuss only a few that may serve as representative examples.he methods of formal speciication from the ield of sotware engineering [e.g., see Hoare (1978) and Jackson (2006)] are challenged by the number of interacting entities and their local interactions because the size of state space grows with the size of a collective (Brambilla et al., 2014).When this is coupled with the complexity of the dynamic environments that we typically expect these agents to exist in, we rapidly ind ourselves in need of novel techniques to model and explain the dynamics of our systems.Concise mathematical descriptions of systems, such as methods from chemistry (van Kampen, 1992), are typically incapable to model complex agent-to-agent interactions, especially in the case where spatiality plays a central role (Ohkubo et al., 2008).Computational models oten require rather strong abstractions for the sake of run-time eiciency.Agent-based models typically require an increased number of parameters with increasing system complexity which challenges their signiicance (Mayer et al., 2010).

Our Perspective and Approaches
Engineered hybrid societies are complex, and therefore it is diicult to develop de novo novel mathematical formalisms.
A common option is to use frameworks that were developed for natural systems to formalize artiicial systems when they share key features.In general, two aspects are formalized: (1) the behavioral mechanisms themselves (at microscopic or macroscopic level) and ( 2) the process that leads to these mechanisms (e.g., evolution in natural systems, machine learning in artiicial systems).
Chemistry and statistical physics provide formal, mechanistic descriptions of hybrid systems.hey are the disciplines that inspired, for example within swarm robotics, the most commonly used modeling frameworks (Brambilla et al., 2013), such as the master equation approach from chemistry (Martinoli et al., 1999) and use of Fokker-Planck and Langevin equations from statistical physics (Hamann and Wörn, 2008).However, the main challenge consists in going beyond the typical assumptions of these approaches that are intrinsic for large numbers of components ["Avogadro-large, " cf.Beni (2005)] and lack capabilities to model cognition and communication.Hence, collaborations with physicists and theoretical chemists could help to extend these models, to account for smaller system sizes, and to model cognition, and to explicit communication.
Less attention has been paid to the formalization of processes leading to self-organization as done in theoretical evolutionary biology and machine learning.In the irst case, evolutionary game theory (Nowak, 2006) with ininite (e.g., diferential equations) and inite (e.g., birth-death processes) populations provides promising approaches but is limited to the evolution of inite discrete strategies, rather than continuous behavioral traits.Reinforcement learning is a framework suited for single-agent systems (Kaelbling et al., 1996) and in some cases collective systems (Wolpert and Tumer, 1999).In multi-agent settings, machine learning struggles with the combinatorial explosion of possibilities, which is usually approached with sophisticated methods that reduce the search space (Matarić, 1997).To the best of our knowledge, machine learning techniques have never been extended to hybrid societies.

PriMArY cHALLeNGe B: sYsteM DesiGN OF HYBriD sOcieties
Even if we assume that we have a formal speciication of our hybrid society already, then the actual system design is still a big challenge.We would like to deine an integrated process that implements the step from a speciication of a self-organizing collective system to the actual real-world system and its deployment in the ield.In addition, we have to consider typical requirements for engineered systems, such as safety, reliability, and stability.Also note that we consciously take an engineering perspective on hybrid societies, hence assuming that such self-organizing collective systems can actually be designed.his hypothesis is in line with assumptions made in standard approaches, such as swarm robotics (Martinoli, 1999;Brambilla et al., 2013).However, one can also take the perspective that self-organizing systems can at most be guided but not fully determined (Prokopenko, 2009).

requirements and Actions for system Design
Moving from a speciication of a hybrid society to a veriied implementation on actual hardware remains diicult.Dealing with issues such as time, non-determinism, and scale presents signiicant challenges to formal methods.Hybrid societies can be designed with a smaller efort for pre-speciied environments but for real-world implementations quality characteristics have to be determined (Mahendra Rajah et al., 2005;Levi and Kernbach, 2010;Brambilla et al., 2013).Formal methods help to develop tools that ensure system properties, a level of safety, and guaranteed safe sotware from speciication to implementation.
he design for reliability and stability needs to be addressed before we are able to deploy many hybrid societies in the real world.he stochasticity and the autonomy present in such systems make assuring reliability a diicult task.herefore, developing such systems needs to provide evaluation tools that allows for measuring those aspects in a representative way.

secondary challenge: stochasticity, Uncertainty, Unpredictability
Most real-world environments show a high degree of stochasticity, which makes it challenging to deploy hybrid societies in realworld applications.We need methodologies to deal with known uncertainties but also to deal with unforeseen uncertainties.For collective behaviors, we are missing a general model that could be used to verify the system against the expected behaviors.In addition, there might be even unpredictable behaviors [cf., emergent behavior Matarić (1993) and Bedau (2002)] that prevent us from assuring that the system never leaves the set of safe states.

secondary challenge: Dynamic environments, run-time Decisions, and Open systems
Related to the above complex problems, we also face the challenge of dynamic environments that require non-trivial run-time decisions of our system.Run-time decisions and coupling the collective hybrid society with other systems at run-time require new methodologies.Especially systems with high requirements for robustness operating in dynamic environments have to be able to appropriately self-adapt their behaviors and organization structure (e.g., topology).he required time for non-productive reorganization and adaptation processes should be minimal.
If we allow dynamic changes of the system size, that is, we have an open system, then we need to tackle the challenge of scalability at runtime as well.his adds additional uncertainties introduced by added or removed system components.hese changes need to be balanced by the system at run-time to establish a stable and robust system behavior.We oten face diiculties when attempting to make guarantees about the behaviors of our systems and in the scenarios when existing techniques can be used they oten model a ixed number of agents, making our proofs meaningless as the size of our collective changes dynamically.

secondary challenge: Design of Feedbacks for self-Organization and User Feedback
Natural collective systems exhibit diferent features that are remarkable, such as lexibility, adaptability, and robustness.To achieve these through self-organization, they resort to positive and negative feedback mechanisms, the ability to amplify and weaken local individual decisions.he careful design of appropriate feedback processes requires special attention and sophisticated design methods.Besides behavioral feedbacks, collective systems also rely on certain network topologies and network properties, such as power-law degree distributions (scale-free networks), that increase the system's robustness to the loss of connections (Albert et al., 2000;Crucitti et al., 2003;Piraveenan et al., 2013b).
Another feature is that of scale-free correlations (Cavagna et al., 2010), which is the ability of collective systems to inluence far-away neighbors independently of the system size, by still resorting to local interactions only.Besides research on modulating positive feedback (Valentini et al., 2014), the negative feedback and scale-free correlations have received little attention yet and are challenging.
A notable quality of deployed systems is user behavior feeding back steadily into the system.his inevitably entails risks such as collusion, free-riding, or other exploitative and destabilizing actions. he additional challenges, for example in terms of robustness and reliability, therefore need to be considered and irmly rooted in the system design.

Our Perspective and Approaches
Once deployed in the ield, bugs are likely to appear in ways unforeseen by the formalization process.his limitation of the formalization task is termed reality gap in robotics and has been studied in recent years.Solutions range from the restriction of the search space (Koos et al., 2013;Cully et al., 2015) to the design of behaviors during the deployment of the system (Watson et al., 2002;Bredeche and Montanier, 2010). he design of a hybridsociety system could beneit from these approaches.
In order to allow our system to adapt to changes in its dynamic environment, it requires a suicient degree of freedom enabling it to self-optimize and to show reliable behavior.We need to allow for methods of self-repair (Ismail and Timmis, 2010) and self-sustainability (Bredeche and Montanier, 2010), which adds even more complexity to the system and increases the challenge of system design.Incorporating the capability for autonomous reasoning (Anshakov and Gergely, 2010) certainly improves the system but at irst it increases its complexity.

PriMArY cHALLeNGe c: iNterDisciPLiNAritY iN HYBriD sOcietY reseArcH
As the reliance on knowledge gained from other scientiic disciplines grows, so too does the need for researchers from all ields to be prepared to learn from the insights and techniques of others.he investigated problems are becoming too complex to stay within the scope of a single discipline, and hence, interdisciplinary research is becoming more popular (Helbing et al., 2015).Hybrid societies are an inherently interdisciplinary problem domain, thus the inclusion of indings from various disciplines is essential for their structural and algorithmic design [e.g., combination of results from plant science, robotics, and architecture (Hamann et al., 2015)].Interdisciplinarity is crucial to produce a valid model of a system observed in another discipline, or to take inspiration from another discipline in the design of systems.From an engineering perspective, being inspired by biology, chemistry, and sociology is becoming common place.However, engaging in a meaningful way with another discipline can be challenging and oten, not fruitful in part because an approach remains rather supericial where an extra efort with additional overhead would have been required.

requirements and Actions for interdisciplinarity
Engineering has much to ofer to the life sciences, but beneits of engagement must be bi-lateral, so that all disciplines beneit from the collaboration.In particular, the contribution of computer science should go beyond that of a mere service to life sciences but instead establish a bidirectional interaction that also scientiically enriches computer science.For example in the context of bio-hybrid societies, modeling and simulation can be an efective vehicle for collaborations between computer scientists (e.g., multi-agent simulations) and biologists (e.g., behavioral models), with computational models being useful to help understand challenges in behavioral biology, yet providing a formal background and inspiration to the creation of an artiicial system, for example based on behavioral models of animals (Schmickl and Hamann, 2011) or growth models of plants (Zamuda and Brest, 2013;Hamann et al., 2015).We should try to get inspiration from biology and sociology while lending our skill sets to the understanding of other ields.However, interdisciplinary research in hybrid societies has proven to be challenging.

secondary challenge: common Language
Despite our best will to ensure interdisciplinarity, it remains diicult to achieve in practice.hese diiculties stem from the disparity in vocabulary, the diferent methodologies used, and a general lack of understanding of the way of thinking and the tools available on each side.Time is needed to develop an interdisciplinary collaboration.A common language needs to be developed so that deep and meaningful collaborations are possible.

secondary challenge: integration of Methods
Once a simple mutual understanding of the available methods and present problems is obtained, it is tempting to merely transfer a method from one ield to the other and to directly apply it to a particular problem.However, mastering the complex problems at hand and lastingly improving these systems goes beyond applying existing results but requires true interdisciplinary collaboration.Providing a broad set of insightful tools, only highly integrated research on novel systems leads to a meaningful design method for hybrid societies.Prime examples of successful integration of methods are the integration of robots and ish (Marras and Poriri, 2012) and the automatic analysis of social networks in honeybees (Wario et al., 2015).Again, establishing such a deep understanding of the other ield requires time.

secondary challenge: interdisciplinarity in a Mono-Disciplinary World
Despite the success of interdisciplinary research and a lot of hype and lip service in favor of interdisciplinarity, realities still look diferent.Many institutions and traditions in research are still forming tiny mono-disciplinary worlds.Hence, there is a challenge for individual researchers to fulill their own discipline's requirements in terms of measures of success.

Our Perspective and Approaches
A probably obvious solution is to enable the human factor and to form small, strongly linked teams that work interdisciplinarily.In addition, interdisciplinary researchers should receive an elaborate training for the ield they are collaborating with.hen the methods that are used to design solutions for diferent problems should transgress disciplinary bounds, in order to allow re-use of methods across ields of research.
Similarly to the situation when travelers have to adapt to local customs, all involved parties need to compromise. he common vocabulary needs to be found and the various perspectives and the diferent knowledge need to be understood.Only then one can start to discover where and how both sides can beneit from each other or how they can join forces to design novel methods for hybrid societies.

cONcLUsiON
We have identiied three primary challenges of designing hybrid societies: formalization, system design, and interdisciplinarity.All of them require a lot of attention and a major efort to be overcome.However, a generic formalization approach and eicient interdisciplinary collaborations shall create synergies and enable us to re-use methods at intersections between disciplines.An appropriate system design approach would enable us to quickly deploy safe, reliable, and stable systems in hardware.

tABLe 1 |
Diversity of methods for the formalization of hybrid societies.