Introducing Conviviality as a New Paradigm for Interactions among IT Objects Assaad Moawad and Vasileios Efthymiou and Patrice Caire and Grégory Nain and Yves Le Traon 1 Abstract. been introduced as a social science concept for multi-agent and ambi- The Internet of Things allows people and objects to seamlessly ent intelligent systems to highlight soft qualitative requirements like interact, crossing the bridge between real and virtual worlds. Newly user friendliness of systems [5]. created spaces are heterogeneous; social relations naturally extend In this paper, we extend conviviality as a new paradigm for IoT In- to smart objects. Conviviality has recently been introduced as a so- formation Technology (IT) objects in two ways. First, convivial rela- cial science concept for ambient intelligent systems to highlight soft tions among IT objects and human users allow the latter to fulfill their qualitative requirements like user friendliness of systems. Roughly, needs for social interactions, and second, convivial relations among more opportunities to work with other people increase the convivial- IT objects facilitate cooperation among participants. The aim is to ity. In this paper, we first propose the conviviality concept as a new enable knowledge sharing for the collective achievement of com- interaction paradigm for social exchanges between humans and In- mon objectives among entities which form various groups or coali- formation Technology (IT) objects, and extend it to IT objects among tions [3]. The challenge of capturing social relations among IT ob- themselves. Second, we introduce a hierarchy for IT objects social jects breaks down into the following research questions: 1) How to interactions, from low-level one-way interactions to high-level com- distinguish the different kinds of social interactions of IT objects? 2) plex interactions. Then, we propose a mapping of our hierarchy lev- How to map the social interactions among IT objects to convivial- els into dependence networks-based conviviality classes. In partic- ity classes? 3) How to measure the conviviality of an individual IT ular, low levels without cooperation among objects are mapped to object? and 4) How to use conviviality in the Internet of Things? lower conviviality classes, and high levels with complex cooperative Tools for conviviality are concerned in particular with dynamic IT objects are mapped to higher conviviality classes. Finally, we in- aspects of conviviality, such as the emergence of conviviality from troduce new conviviality measures for the Internet of Things, and an the sharing of properties or behaviors whereby each member’s per- iterative process to facilitate cooperation among IT objects, thereby ception is that their personal needs are taken care of [12]. In such the conviviality of the system. We use a smart home as a running dynamic circumstances, the conviviality of each participating mem- example. ber is a key criterion. In [4], conviviality measures were introduced by counting, for each pair of agents, the possible ways to cooperate, indicating degree 1 Introduction of choice or freedom to engage in coalitions. In this paper we build Two decades ago, Mark Weiser coined the term ubiquitous com- on these measures to define conviviality measures for each agent. puting. Ubiquitous computing “enhances computer use by mak- Our coalitional theory is based on dependence networks [6, 19], la- ing many computers available throughout the physical environment, beled directed graphs where the nodes are agents, and each labeled while making them effectively invisible to the user” [22]. edge represents that the former agent depends on the latter one to Today, microelectronic devices have become so small and inex- achieve some goal. Furthermore, in order to increase the conviviality pensive that they can be embedded in almost everything, making ev- of the system, we establish an iterative process through which the eryday objects “smart” [15]. The new paradigm of the Internet of least cooperative IT objects are identified, then, upgrades for these Things (IoT) has emerged. The basic idea behind it is the perva- objects are proposed to enhance their cooperations and increase their sive presence around us of a variety of smart objects which, “through inclusions into more coalitions. unique addressing schemes, are able to interact with each other and Our motivation lies in the vision that IT objects will be endowed cooperate with their neighbors to achieve common goals” [1]. with all the capabilities needed for a society of objects fully inte- Smart objects carry chunks of application logic. They sense, log, grated into human society. In [14] smart objects differ from simple and interpret what is happening to them and the world, they act on tracking objects such as RFIDs, in that they are autonomous physi- their own, interact with each other, and exchange information with cal/digital objects augmented with sensing, processing and network- human users. They know what “has happened to them in the past” ing capabilities. Here, we refer to both kinds of objects as IT objects. [15]. In this heterogeneous world, consisting of both human users The structure of this paper is the following: In Section 2, we pro- and objects, social relations naturally extend to objects. vide the background for our IT object interaction classification, we The concept of conviviality, defined by Illich as “individual free- then introduce our motivating example, in Section 3. We propose our dom realized in personal interdependence” [12], focuses on the co- mapping between IT objects interaction classes in Section 4, and the operative aspects of the interactions among humans. It has recently conviviality measures for individual IT objects in Section 5. We dis- cuss these measures in Section 6 and present some related work in 1 Interdisciplinary Center for Security, Reliability and Trust, Section 7, and conclude in Section 8. University of Luxembourg, email: firstname.lastname@uni.lu Workshop on AI Problems and Approaches for Intelligent Environments (AI@IE 2012) 3 2 IoT Evolution and conviviality issues 3 Running Example: Smart-House The Internet of Things relates to the interconnection of communica- In this section we present a scenario of a smart-house automation tion enabled-IT objects [9, 1]. IT objects from our everyday life are system, regulating the temperature of a room. The IT devices in this getting more communication abilities every day. TVs, phones or cars, scenario communicate, trying to figure out the cause of a heating are now able to share information and offer services to each other. problem. Such automation systems could be used to improve the New services taking advantage of these communication links and energy-efficiency of a house and also reduce the cost of living. In shared data are emerging from these new abilities of IT devices. But similar ways, smart-home or smart-city automation systems could the way toward seamlessly interacting devices and smart services is achieve a better quality of life, improved public services, ambient still long. assisted living, or simply entertainment. The miniaturization of hardware material for computation made it In our example, illustrated in Figure 1, we use five types of IoT possible to introduce programs in electronic devices. That is how objects that can communicate with each other; a refrigerator and its autonomous regulation devices made it possible to automate sev- log, a heating system, door sensors and a phone. To accomplish their eral household tasks and duties (e.g.: in Heating Ventilating and Air goal and find the source of the heating problem, the devices exchange Conditioning (HVAC) systems). This automation of basic tasks re- information, query their logs and perform reasoning. sulted in an increase of the comfort and security for users. These The heating system is responsible for keeping the room in a spec- autonomous regulation systems were the foundation of Internet of ified temperature at all times. However, in the last several minutes Things. it has not been able to reach this temperature. Therefore it informs Along with the democratization of computers, the Internet and the refrigerator that it has problems heating the room (step 1). Like communication technologies, autonomous regulation systems got en- the heating system, the refrigerator is responsible for keeping its in- riched with customization capabilities and sometimes remote ac- terior in a specified temperature. In other words, they have similar cesses [18]. Now, users can specify the behavior of such configurable tasks. Hence, if the refrigerator has encountered a similar problem devices to enhance their own comfort and usage. Many devices have and solved it in the past, then there is a possibility that this solution been equipped with bi-directional communications links, for read- could also work for the current problem of the heating system. Con- ing and writing their configurations. With simple user interfaces, non sequently, every time the heating system encounters a problem that it electronic-specialists are now allowed to configure and/or remotely cannot solve, it “consults” the refrigerator. use their devices. The refrigerator receives and processes this transmission. It dis- The availability of Internet everywhere and at any time, opens the covers from its log (step 2) that the last time it had a problem reach- door to remote accesses to IT devices, being at the office or at home. ing a specified temperature, this was because its door was open. After One can cite media centers, alarm or heating systems, or video cam- its door was closed, the refrigerator could function properly, so this era for example. However, the configuration of such communication- was a confirmed solution to this problem. Therefore, the refrigerator enabled devices can sometimes turn into a nightmare for the unini- searches for a signal from the door sensors and receives that one of tiated. As a consequence, protocols have been set up to allow auto- the house doors is open (step 3). The heating system is informed by matic device recognition and connections. Also, zero-configuration the refrigerator that the problem comes from an open door (step 4). devices[20] that are able to self-configure and get ready for use are Finally, the heating system stops functioning, until the problem is more and more present in the IT environment. resolved (step 5). It also informs the phone that there is a heating The paradigm of Cloud tend now to get rid of the precise location problem and the recommended action is to close the door (step 6). to access a device, a service or a content. Resources can be accessed at any time, from anywhere and in several ways, with no idea about the precise location of this resource. Today, Things (i.e. IT Objects) are able to communicate, are re- motely accessible and are available from anywhere at any time [21]. But the services offered by these devices do not adapt or evolve with the presence of other services from other IT Objects. The next gen- eration of Internet connected Things should be able to autonomously collaborate, adapt their behavior and services offered, according to their capabilities and to surrounding objects’ capabilities and needs. They will participate at a time, in a community of devices by provid- ing a new service, and integrate later another community as a backup for an already existing service. Some classification or measures have to be developed to catego- rize these interactions among Things from a simple data provision to a collaborative decision making capability. As a social interaction measure, the conviviality can be applied to interactions between IT objects, and with humans, and provide a first set of tools for the next Figure 1. IT objects cooperate to solve a heating problem. generation of smart devices. They could then be able to make more accurate decisions when adapting to their surroundings and evolving. This is a typical example in Ambient Intelligence, where devices They could be able to choose the community of devices to connect to with different interaction capabilities have to cooperate. We now for- by maximizing the benefit for both the community and themselves. malize the levels of this interaction, that we call social interaction of They could even be able to improve their social involvement by ac- IT objects, by using the notion of conviviality. quiring new skills or taking charge of some duties. 4 Workshop on AI Problems and Approaches for Intelligent Environments (AI@IE 2012) 4 IT Objects Classification and Mapping We use the term social interaction for IT objects, in a way similar to the human social interaction, as their ability to communicate with In this section we discuss how IT objects can have a social interaction other IT objects and exchange information. and how these interactions can be classified. For this classification, In Table 1, we illustrate the different levels of social interaction we use the notion of conviviality and Dependence Networks. that an IT object can have. Level 0 IT objects are those who can only receive information from other IT objects. The phone, in our scenario Definition 4.1 (Dependence networks) . A dependence network belongs to Level 0, as it only receives alerts from other devices. (DN) is a tuple �A, G, dep, ≥� where: A is a set of agents, G is a Level 1 is about the objects that only share their information with set of goals, dep : A × A → 2G is a function that relates with each other objects. The door sensors are of Level 1, because they can only pair of agents, the sets of goals on which the first agent depends on transmit the state of the doors to other devices. the second, and ≥: A → 2G × 2G is for each agent a total pre-order Level 2 IT objects are programmed explicitly to interact with spe- on sets of goals occurring in his dependencies: G1 >(a) G2 . cific objects. The refrigerator’s log is of Level 2, as it interacts only with the refrigerator. The heating system is on the same level. Level Moreover, a Dependence Network can be represented by a di- 2 is the current maximum social interaction level of IT objects. rected graph, where the agents are the nodes of the graph, and the de- Finally, Level 3 IT objects have the potential to interact with any pendencies form the directed edges. For example, Figure 2 illustrates other object, in order to achieve a goal. In our scenario, the refrig- the graph that represents the Dependence Network derived from our erator is of Level 3, since it is not explicitly programmed to interact motivating example of Section 3. Note that dependencies are poten- with a specific set of devices. tial, i.e, not all of them are actualized in our scenario. It should be also We ignore IT objects operating only autonomously. The heating clarified that DNs are not equivalent to data flow networks; the latter system could also work autonomously and try to keep a stable tem- model information exchange, not dependencies. The heating system perature in the room. However, its social interaction led to an im- h depends on the door d to be close, in order to function properly, proved, a more efficient functionality. but h does not have the capability to interact with d. The social interaction level of IT objects can be associated with the conviviality of a network, in which they participate. To present d this association in Table 1, we first suggest four possible DNs, each of them including at least one node of the specific level and then l r p analyze the maximum conviviality in such a DN. The maximum conviviality class of a network that includes a Level h 0 or 1 object is N , since this node cannot be a part of a cycle. Level 0 objects have no incoming edges in a DN and Level 1 objects have Figure 2. The DN of our example. h is the heating system, r is the no outgoing edges. For networks that include an object of Level 2, refrigerator, l is the refrigerator’s log, d is the door sensor and p is the phone. the conviviality cannot be better than AP e, since such an IT object is not able to interact with every other device. Hence the graph of DN cannot be a clique. The maximum conviviality of a network that Conviviality has been introduced as a social science concept for includes an IT object of Level 3 is P . Furthermore, P conviviality is multi-agent systems to highlight soft qualitative requirements like achieved only if every node of the DN is a Level 3 IT object. W con- user friendliness of systems [4]. The idea of conviviality is based viviality can exist if all nodes are Level 1. The maximum conviviality on the notion of interdependency; Cycles denote the smallest graph of a graph with only Level 0 and 1 objects is AW e. topology expressing interdependence, and are considered as atomic Interaction 0 1 2 3 relations conveying conviviality. When referring to cycles, we are implicitly signifying simple cycles (as defined in [8]), without repe- tition, with order and discarding self-loops. In [4], conviviality is classified as presented in Figure 3, through Data a ranking of the DNs. Briefly, (W ) is the worst class of conviviality because all agents are isolated. On the opposite side, (P ) achieves perfect conviviality because the corresponding graph is a clique. For the in-between classes, (AW e) class has some dependencies but no DN cycles, (N ) class has at least one isolated node and one cycle and in max conv N N AP e P (AP e) class, all the agents are participating in at least one cycle. CONVIVIALITY Table 1. Social interaction level of IT objects and maximum conviviality if at least one such object appears in a DN. W AWe N APe P In this section, we have introduced a novel approach to classify- a b a b a b a b a b ing the social interaction of IT objects. We have mapped the social interaction level of an object with the maximum conviviality that can d c d c d c d c c d be achieved if this object is included in a DN. To do this mapping, we have established correspondences between DNs and IT objects interaction level. This way, we can have a maximum conviviality es- Figure 3. Conviviality Classes. timation, just by knowing the social interaction level of IT objects that participate in our system. However, it is sometimes necessary to get a more accurate measure to improve the conviviality of a system. Workshop on AI Problems and Approaches for Intelligent Environments (AI@IE 2012) 5 5 Conviviality Measures where Len(c) is the length of the cycle c and ω is the maximum num- ber the sum in the numerator can get, over a dependence network of Social network analysis has been providing many measures to re- the same size but with all possible dependencies (a clique). More- flect social interactions among agents [16]. However none of these over, ω is related to the Ω measured in Section 5.2 by the formula: considers cycles as basis. Our measurements meet the following re- ω = Ω/|A| because of the symmetry between all agents in a clique. quirements and assumptions. This measurement per agent is also a rational number bounded in [0,1]. An agent participating in no cycle at all would have 0 convivi- 5.1 Assumptions and Requirements ality, and all agents in a fully-connected dependence network would have a conviviality of 1. In this work, the cycles identified in a dependence network are con- Finally, the conviviality measurement for the whole dependence sidered as coalitions. These coalitions are used to evaluate convivial- network defined in Section 5.2 can be deduced by calculating the ity for the network and for each agent. average conviviality of all agents in the dependence network: In our second assumption, we consider the conviviality of a depen- � dence network or a specific agent to be evaluated in a bounded do- (convDN (a)) main, i.e., over a [min; max] interval. This allows reading the values Conv(DN ) = a∈A (3) obtained by any evaluation method. |A| In terms of requirements, the first requirement for our conviviality measures concerns the size of coalitions. It is captured by the state- 5.4 Computation ment that larger coalitions are more convivial than smaller ones. Our second requirement concerns the number of coalitions. It is We apply our computation on the dependence network of the running captured by the statement that the more coalitions in the dependence example illustrated in Figure 2. In this example, the set of all cycles network, the higher the conviviality of DN would be (all else being is C = {(h, r), (r, l)} equal). Similarly, the more coalitions an agent is participating at, the The pairs participating in one cycle are (h, r), (r, h), (l, r), (r, l) higher its conviviality measure would be. This requirement is mo- and there are no pairs participating in more than one cycle, thus the tivated by the fact that a large number of coalitions indicates more conviviality of the dependence network, according to Equation 1 is interactions among agents, which is positive in term of conviviality Conv(DN ) = 4/Ω with Ω = 980 calculated over a clique of 5 according to our definition based on interdependence. nodes. Now, to calculate the conviviality of each agent, we need to list the cycles containing that agent and applying Equation 2. We get: 5.2 Conviviality of a dependence network CDN (h) = {(h, r)}, conv(h) = 1/ω CDN (r) = {(h, r), (r, l)}, conv(r) = 2/ω The conviviality of a dependence network DN is defined in [4] as CDN (l) = {(r, l)}, conv(l) = 1/ω � CDN (p) = {∅}, conv(p) = 0/ω = 0 coal(a, b) CDN (d) = {∅}, conv(d) = 0/ω = 0 a,b∈A,a�=b Conv(DN ) = (1) Where ω = Ω/5 = 196. Ω Note that, by taking the average of the convivialities of all the where coal(a, b) for any distinct a, b ∈ A is the number of cycles agents, we get avg = (1/ω + 2/ω + 1/ω + 0 + 0)/5 = 4/(5ω) = that contain both a and b in DN and Ω is the maximum the sum 4/Ω = Conv(DN ) as stated in Equation 3. Figure 4 shows our in the numerator can get, over a dependence network of the same set computation and the IT level of the objects of DN . of goals and the same number of agents but with all dependencies (fully-connected graph). Ͳ   This way, the conviviality measurement of a dependence network Level  1   ͳ ʹ which is a rational number in [0,1], can be used to compare different ͳͻ͸   ͳͻ͸   Ͳ   dependence networks, with 0 being the conviviality of a dependence network having no cycles at all (class W , class AW e) and 1 the conviviality of a fully-connected dependence network (class P ). Level  2   Level  3   Level  0   However, this measurement just reflects the conviviality of the whole dependence network and does not allow to compare, inside the Level  2   ͳ   ସ same dependence network, the conviviality of two different agents. Conviviality  of  DN:   ଽ଼଴         ͳͻ͸ 5.3 Conviviality of an agent Figure 4. IT levels and conviviality measurements for the agents of DN . In this work, we extend the conviviality measures of a dependence network DN , by defining the conviviality of each agent inside DN . First, Let CDN (a) be the set of all cycles in DN that contains the agent a. As a conclusion, these measurements provide a way to compare We define the conviviality of an agent a ∈ A as agents to each other according to their social interactions and there- fore they can be used to find potential improvements in the depen- � dence network. For instance, in this example, we can deduce that (Len(c) − 1) c∈CDN (a) agents d and p are the least convivial and can be seen as bottlenecks convDN (a) = (2) for the conviviality of DN . ω 6 Workshop on AI Problems and Approaches for Intelligent Environments (AI@IE 2012) 6 Using Conviviality in IoT 6.3 Conviviality as an Incentive 6.1 Iterative Process Conviviality can be used in agent theory to satisfy requirements on user-friendly systems and ensure that considerations such as the us- Improvement of the conviviality of a system is an iterative process. ability of a system get the same importance as the functionality. First, we identify the less participating agents in the network. Then, In this section we discuss how conviviality measures can be used we try to involve them in more coalitions, which will increase their by agents as an incentive for cooperation, using a game-theoretic conviviality and consequently the conviviality of the system. If this framework [17]. IT objects with a social interaction level of 3, as solution is not applicable, then we upgrade these agents, when pos- defined in Section 4, are not programmed explicitly to interact with sible, to increase their participation. The overall conviviality of the specific objects. Depending on their needs, they have the ability to system can thus be improved by iterating these steps. cooperate with any other object that will help them, or that needs to be helped. In order to decide on the form of cooperation, Level 3 6.2 Computation Examples objects have first to find out from which coalition they will benefit more, or, to which coalition they can contribute more. In the previous scenario, agents d and p are the least convivial and The conviviality measures, introduced in Section 5, can be used to cannot do better because of their IT interaction levels of 0 and 1. We calculate the payoff of each agent participating in a coalition. Thus, suggest as an alternate scenario S � to upgrade them by other IT ob- agents have a formal way to calculate the gain that their participation jects (d� and p� ) having an IT interaction level of 2. But this is not in a coalition infers and therefore, decide which coalition to join. enough. If the upgraded objects do not have the possibility to partici- A co-operative game is determined by a set Ag of agents wherein pate in more coalitions, the measurements will still remain the same, each subset of Ag is called a coalition, and a characteristic func- as the number of coalitions is unchanged. In the alternate scenario tion V , assigning each coalition its maximum gain, the expected to- S � , the smartphone p� (level 2) can have a more important role than tal income of the coalition (the so-called coalition value). The payoff just being a display device: it has a very good computation power distribution, P , assigns each agent its utility out of the value of the and the ability to connect to the Internet to get updates and some coalition it is member of in a given coalition structure [13]. In other information for example on potential solutions to a problem in the words, P is the gain of the agent and V is the gain of the coalition. smart home context. In particular, the refrigerator and the heating The main idea behind these notions is to find out if the agents system can potentially depend on its computation and connectivity have an incentive to form coalitions. If the agents are not motivated capabilities. Figure 5 illustrates the dependence Network with the to form coalitions, or if they find another, better coalition for them, conviviality computation for scenario S � . then the current coalition is at risk; it is unstable. If the payoff of Ͳ   the agents is greater when they are in a specific coalition, than what Level  2   they would gain otherwise, then that makes this coalition stable. We      ͛   ͳ   ͹   ͸   propose conviviality measures as a way to quantify this gain. ͳͻ͸ ͳͻ͸ ͳͻ͸ In our example, let’s consider that the Level 3 refrigerator r is   not yet a part of a network and it is trying to decide which of the Level  3   Level  2   networks, Figure 4 or Figure 5, to join. Then r can calculate what Level  2   its conviviality would be if it joined each of these networks. It finds out that in the network of Figure 4 its conviviality would be 2/196, Level  2     ଶ଴ ͸ whereas in the one of Figure 5 its conviviality would be 7/196. Conviviality  of  DN:           ଽ଼଴ ͳͻ͸ Therefore, it decides to join the second network. Figure 5. The alternate scenario S � with the new conviviality computation. 6.4 Computational Challenges In our vision of the IoT, each IT object has the ability to act au- Comparing to Figure 4, we can deduce that conviviality of the re- tonomously, in the sense of decision making. This means that IT ob- frigerator, heating system and the phone has improved. On the other jects can perform computations before joining a coalition, like the hand, conviviality of the door and the log remain unchanged. Glob- refrigerator in the previous paragraph. This is different from what ally, the conviviality of the whole dependence network in S � has im- usually exists today; a centralized system that makes all the computa- proved 5 times (4/980 to 20/980). Note that, the more dependencies tions. In today’s systems, devices are usually programmed to interact we add to DN, the faster the conviviality increases, exponentially, to only with the central computer and get these computational results, reach the 980/980 for a fully-connected DN because of the combina- or request an available service. torial nature of the measurements. The problem with our Level 3 objects is that smaller IT objects Finally, having the maximum conviviality is not always the best usually have a limited capability of processing. The computational option for an IT system, because it might have other requirements complexity of our conviviality measurements is prohibiting such and other constraints like security, privacy, efficiency, power man- small devices to perform this calculation, especially as the number agement requirements, costs, etc. For instance, in a secured location, of agents in the network increases. This also limits the potential size a smartphone might not be allowed to connect to the Internet for se- of coalitions that can be created, since for larger coalitions it is harder curity reasons. In a camping context, a smartphone might not be able to compute the conviviality in a reasonable time. to do a lot of computations due to power saving measures. A good One possible solution is to revise these measurements and make trade-off between conviviality and other requirements is the key to them computationally easier for such devices. The new measure- have a better system. Smart IT objects should be capable to adapt to ments should also meet the requirements introduced in [4]. Another different situations and contexts, selecting between different trade- approach would be to consider new definitions of conviviality. offs accordingly in order to optimize their utility. Workshop on AI Problems and Approaches for Intelligent Environments (AI@IE 2012) 7 7 Related Works example, when adapting to their surroundings, and while evolving. Furthermore, we will focus on the capability for smart devices and Many measures exist in graph theory domain that can be used to IT objects to choose the community of devices and objects they may reflect the “social importance” of a node and the “structural impor- connect to. This choice may be guided by maximizing both their own tance” of a graph [10, 11, 16]. Some of the most relevant measures benefit as well as their communities’, i.e., the coalitions they belong for a node are: clustering coefficient of the node which is the ratio to. Finally, we plan to enable devices and objects with the possibility of existing links connecting the node’s neighbors to each other, to to improve their social involvement through new skill sets acquisition the maximum possible number of such links, closeness centrality of and the adoption of new goals. the node which is the reciprocal of the sum of distances to all other nodes in the graph. For a graph, we have the clustering coefficient of the graph which is the average of the clustering coefficients of all the REFERENCES nodes. However, these measurements do not take into consideration [1] L. Atzori, A. Iera, and G. Morabito, ‘The internet of things: A survey’, cycles in the graph. Our conviviality measures are based on the num- Computer Networks, 54(15), 2787–2805, (2010). [2] Guido Boella, Luigi Sauro, and Leendert van der Torre, ‘Algorithms for ber and size of cycles in the graph which reflect interdependence. finding coalitions exploiting a new reciprocity condition’, Logic Jour- In [7], Castelfranchi et al. use dependence networks to represent nal of the IGPL, 17(3), 273–297, (2009). trust among agents. They build a socio-cognitive model of trust and [3] P. Caire, S. Villata, G. Boella, and L. van der Torre, ‘Conviviality masks present measurements for the degree of trust and trust capital. in multiagent systems’, in 7th International Joint Conference on Au- The i-dud property [2] is a reciprocity property, saying that “an tonomous Agents and Multiagent Systems (AAMAS 2008), Estoril, Por- tugal, May 12–16, 2008, Volume 3, pp. 1265–1268, (2008). agent sees to a goal of another agent only if this enables it to obtain, [4] Patrice Caire, Baptiste Alcade, Leendert van der Torre, and Chat- directly or indirectly, the satisfaction of one of its own goals”. This is trakul Sombattheera, ‘Conviviality measures’, in 10th International also a desired approach for our goal-directed agents, similar to what Joint Conference on Autonomous Agents and Multiagent Systems (AA- we refer to as interdependency, or conviviality. MAS 2011), Taipei, Taiwan, May 2–6, 2011, (2011). [5] Patrice Caire and Leendert van der Torre, ‘Convivial ambient technolo- The notion, issues, and challenges of dynamic coalition formation gies: Requirements, ontology and design’, The Computer Journal, 3, (DCF) among rational software agents are introduced in [13]. For the (2009). formation of a dynamic coalition, the coalitions are represented by a [6] C. Castelfranchi, ‘The micro-macro constitution of power’, Protosoci- coalition leader, who continuously attempts to improve the value of ology, 18, 208–269, (2003). its coalition, by building re-configurations and suggesting them to its [7] Cristiano Castelfranchi, Rino Falcone, and Francesca Marzo, ‘Being trusted in a social network: Trust as relational capital’, in iTrust, pp. coalition members. This is different to what we vision for the future 19–32, (2006). IoT, as discussed in 6.4, where there is no coalition leader. However, [8] Thomas H. Cormen, Charles E. Leiserson, Ronald L. Rivest, and Clif- this approach could solve the computational issues that are discussed ford Stein, Introduction to Algorithms, The MIT Press, 2nd edn., 2001. in the same section. [9] The Internet of Things - 20th Tyrrhenian Workshop on Digital Commu- nications, eds., Daniel Giusto, Antonio Iera, Giacomo Morabito, and Luigi Atzori, Springer, 2010. [10] D. Gómez, ‘Centrality and power in social networks: a game theoretic 8 Conclusion approach’, Mathematical Social Sciences, 46(1), 27–54, (August 2003). [11] Robert A. Hanneman and Mark Riddle, Introduction to social network In this paper, we extend the social concept of conviviality as a new methods, 2005. paradigm for IoT IT objects in two ways. First, convivial relations [12] Ivan Illich, Tools for Conviviality, Marion Boyars Publishers, London, among IT objects and human users support the latter in fulfilling August 1974. their needs for social interactions, and second, conviviality among [13] Matthias Klusch and Andreas Gerber, ‘Dynamic coalition formation among rational agents’, IEEE Intelligent Systems, 17, 42–47, (2002). IT objects facilitates their cooperation. [14] G. Kortuem, F. Kawsar, D. Fitton, and V. Sundramoorthy, ‘Smart ob- We first introduce a hierarchy for IT objects social interactions, jects as building blocks for the internet of things’, Internet Computing, from low-level one-way interactions to high-level complex interac- IEEE, 14(1), 44–51, (2010). tions. Second, we propose a mapping of our hierarchy levels into [15] Friedemann Mattern, ‘From smart devices to smart everyday objects (extended abstract)’, in Proceedings of sOc’2003 (Smart Objects Con- dependence networks-based conviviality classes. In particular, low ference), pp. 15–16, Grenoble, France, (may 2003). levels without cooperation among objects, are mapped to lower con- [16] Alan Mislove, Massimiliano Marcon, Krishna P. Gummadi, Peter Dr- viviality classes, and high levels with complex cooperative IT ob- uschel, and Bobby Bhattacharjee, ‘Measurement and analysis of online jects are mapped to higher conviviality classes. Third, we define new social networks’, in Proceedings of the 7th ACM SIGCOMM confer- measures, since conviviality measures introduced in [4] are over the ence on Internet measurement, IMC ’07, pp. 29–42, New York, NY, USA, (2007). whole network, and do not differentiate among objects. [17] Martin J. Osborne and Ariel Rubinstein, A Course in Game Theory, The Fourth, in order to increase the conviviality of the system, we es- MIT Press, July 1994. tablish an iterative process through which the least cooperative IT ob- [18] Timothy I Salsbury, ‘A survey of control technologies in the building jects are identified, then, upgrades for the identified objects are pro- automation industry timothy i. salsbury’, Proc IFAC World Congress, (2005). posed to allow more cooperations among them, by increasing their [19] Jaime Simão Sichman and Rosaria Conte, ‘Multi-agent dependence by inclusions into a greater number of coalitions. The process iterates to dependence graphs’, in Procs. of The First Int. Joint Conference on satisfy the system requirements, in which the tradeoffs among poten- Autonomous Agents & Multiagent Systems, AAMAS 2002, pp. 483–490. tially conflicting requirements have been set, for example between ACM, (2002). conviviality, efficiency, privacy and security. [20] Daniel Steinberg and Stuart Cheshire, Zero Configuration Networking: The Definitive Guide, O’Reilly Media, Inc., 2005. In future works, we plan to define the requirements needed for [21] International Telecommunication Union, ‘The internet of things’, ITU communications and negotiations among level three objects. We also Internet Reports 7, International Telecommunication Union, (2005). want to provide a first set of tools for the next generation of smart [22] Mark Weiser, ‘Some computer science issues in ubiquitous computing’, devices and IT objects. More specifically, we plan to endow such Commun. ACM, 36(7), 74–84, (1993). objects with the capability of making more accurate decisions, for 8 Workshop on AI Problems and Approaches for Intelligent Environments (AI@IE 2012)