Categories of Artificial Societies Paul Davidsson Department of Software Engineering and Computer Science Blekinge Institute of Technology, 372 25 Ronneby, Sweden [email protected]

Abstract. We investigate the concept of artificial societies and identify a number of separate classes of such societies. These are compared in terms of openness, flexibility, stability, and trustfulness. The two most obvious types of artificial societies are the open societies, where there are no restrictions for joining the society, and the closed societies, where it is impossible for an “external agent” to join the society. We argue that whereas open societies supports openness and flexibility, closed societies support stability and trustfulness. In many situations, however, there is a need for societies that support all these aspects, e.g., in systems characterized as information ecosystems. We therefore suggest two classes of societies that better balance the trade-off between these aspects. The first class is the semi-open societies, where any agent can join the society given that it follows some well-specified restrictions (or, at least, promises to do so), and second is the semi-closed societies, where anyone may have an agent but where the agents are of a predefined type.

1 Introduction A collection of software entities interacting with each other for some purpose, possibly in accordance with common norms and rules, may be regarded as an artificial society. This use of the term “society” is analogous to human and ecological societies. The role of a society is to allow the members of the society to coexist in a shared environment and pursue their respective goals in the presence of others. As a software entity typically acts on the behalf of a person or an institution, i.e., its owner, we will here refer to these entities as agents. This use of the term “agent” is somewhat more general than is common. However, since the principles we will discuss are general, covering all kinds of (semi-)autonomous software processes, there is no reason for limiting the discussion to "proper" software agents. There are a number of other notions used to refer to organizational structures of software agents, e.g., groups, teams, coalitions, and institutions (cf. for instance [2, 4, 9]). Since a society may contain any number of institutions, coalitions, teams, groups, and individual agents, the concept of society belongs to a higher organizational level than these structures. Also, whereas a society is neutral with respect to co-operation and competition, coalitions are formed with the explicit intention of

co-operation. Similarly, a team is a group in which the agents have a common goal. The difference between a group or a team and an institution is that an institution has a legal standing distinct from that of individual agents. Artikis and Pitt [1] have provided a formal characterization of an agent society that includes the following entities: - a set of agents, - a set of constraints on the society, - a communication language, - a set of roles that the agents can play, - a set of states of affairs that hold at each time at the society, and - a set of owners (of the agents). They describe the set of constraints as “constraints on the agent communication, on the agent behaviour that results from the social roles they occupy, and on the agent behaviour in general.” Another way describing the set of constraints is that they constitute the norms and rules that the agents in the society are supposed to abide. When appropriate, we will refer to the above list of entities when discussing different types of societies. In addition, we will here regard yet another entity, namely the owner of the society. By this we mean, the person or organization that have the power to decide which agents may enter the society, which roles they are allowed to occupy, what communication language should be used, and the set of constraints on the society. Note, however, that all societies may not have an owner and that there is a difference between the owner and the designer of an agent or a society [6]. Depending on its purpose, an artificial society needs to support the following properties to different degrees: - openness, i.e., the possibilities for agents to join the society - flexibility, i.e., the degree to which agents are restricted in their behavior by the society - stability, i.e., predictability of the consequences of actions, and - trustfulness, i.e., the extent to which the owners of the agents trust the society (which may be based on e.g. mechanisms for enforcing ethical behavior between agents). These properties are not completely independent, e.g., the trustfulness of a society is typically dependent of its stability. An emerging view of future distributed software systems is that of information ecosystems. These are populated by infohabitants, i.e., (semi-)autonomous software entities typically acting on the behalf of humans. According this vision, which is inspired by biotic ecosystems, the information ecosystem should be able to adapt to changing conditions, easily scale up or down, and have an openness and universality. (This vision is shared by the Future and Emerging Technologies initiative “Universal Information Ecosystem” within the Information Society Technologies program of the European Commission (see http://www.cordis.lu/ist/fetuie.htm).)

Thus, information ecosystems correspond to artificial societies with a high degree of openness, flexibility, stability, and trustfulness. In this paper, which builds upon earlier discussions [3], we will categorize artificial societies based on their degree of openness. We start by describing two basic types of artificial societies, the open and closed societies, and discuss their strengths and weaknesses. Based on this discussion we suggest two other types of artificial societies, semi-open and semi-closed, which are able to balance the strengths and weaknesses of open and closed societies. Finally, we discuss the trade-off between openness, flexibility, stability and trustfulness.

2 Open and Closed Agent Societies Two basic types of artificial societies are open societies, where there are no restrictions for joining the society, and closed societies, where it is impossible for an “external agent” to join the society. 2.1 Open Societies In principle, it is possible for anyone to contribute one or more agents to an open society without restrictions. An agent joins the society simply by starting to interact with some of the agents of the society. If we characterize an open society with respect to the four desired properties listed in the introduction, it supports openness and flexibility very well, but it is very difficult to make such a society stable and trustful. For instance, it is not possible to control the set of constraints or monitor whether the agents abide these. In fact, it is not possible to determine the set of agents in any effective way. Within an open society, the only structure is typically just a generally accepted communication language and a limited set of roles. Of course, it is possible to consider artificial societies that do not even have this, which we may call anarchic societies. The most obvious example of an open artificial society is the World Wide Web (WWW), where the set of members of the society consists of the set of WWWbrowser processes together with the set of WWW-server processes that are connected to the Internet. HTTP (Hypertext Transfer Protocol) is the communication language. The number of roles is limited to clients, i.e., the browsers, and servers. Finally, the set of owners is either the owners of the machines on which the browser and server processes are run, or the persons/institutions that started the browser and server processes. The openness of the society is obvious in this case; anyone with an Internet connection is allowed to start a browser process or a server process and join the artificial society defined by the WWW without any restrictions.

2.2 Closed Societies Closed agent societies are typically those where a Multi-Agent System (MAS) approach is adopted by a team of software developers to implement a complex software system. The MAS is designed to solve a set of problems, specified by the society owner. The solving of these problems is then distributed between the agents of the MAS. It is not possible for an “external agent” to join the society. Zambonelli et al. [10] refer to this type of systems as “distributed problem solving systems” and describe them as “systems in which the component agents are explicitly designed to co-operatively achieve a given goal.” They argue that the set of agents is known a priori, and all agents are supposed to be benevolent to each other and, therefore, they can trust one another during interactions. In open systems, on the other hand, agents are “not necessarily co-designed to share a common goal” and cannot be assumed to be benevolent. The concept of closed agent society corresponds to the large majority of existing MAS. An advantage of closed societies is that it is possible to precisely engineer the society, e.g., specify exactly which agents interact, and why etc. Consequently, closed societies provide strong support for stability and trustfulness. However, they are able to provide very little openness and flexibility. Just as we identified anarchic societies as extreme type of open society, it is possible to identify and extreme type of closed societies, namely, the fixed societies. In a fixed society, all agents are created in the initialization phase, whereas in a close society, the owner/designer of the society may create new members of the society during execution time. Our conclusion is that neither open nor closed societies to a sufficient degree support all of the desired features, i.e., flexibility, openness, stability, and trustfulness, that are required for the implementation of large class of software systems, e.g., information ecosystems. We will now investigate two types of artificial societies that are more suitable for this type of systems, namely “semi-open” and “semiclosed” societies to which anybody may contribute an agent, but where entrance to the society is restricted and behavior may be monitored by some kind of instit ution: - in semi-open societies, the agents may be implemented and run locally - in semi-closed societies, the agents are implemented and run on remote servers.

3 Semi-open Artificial Societies In what we will call semi-open artificial societies, there is an institution that function as a gate-keeper. Agents wanting to join the society contacts the institution to whom it promises to follow the set of constraints of the society. The institution then makes an assessment whether the agent is trustworthy and eligible and decides whether to let it join the society or not. (See Figure 1.) It is, of course, possible to differentiate between classes of trustworthiness so that agents that are considered more trustworthy are given access to more services etc than agents considered less trustworthy.

1

2 Figure 1: Logical view of an agent entering a semi-open society. Agents may be arbitrarily distributed in the physical space.

A type of institution that implements this functionality to some extent is the portal concept as used in SOLACE [5]. In addition to keeping track of the entities and services of the society, it can be used to “ensure that requirements such as security, integrity, and privacy of information related to the services associated with a particular portal is effectively taken care of.” In fact, there are already a number of distributed information systems that resemble semi-open societies. For example, consider peer-to-peer systems [8], such as the Internet-based Napster service (cf. www.napster.com) which let users share their achieves of mp3 files containing music. Each user must use a particular type of Napster software. In order to get access to other users' files, a Napster software process needs to connect to a central server, which then may let the process join the society. If the process succeed to join the society, it will be able to interact with other users’ Napster software processes, downloading and uploading mp3 files. Thus, anyone may potentially contribute an agent (or more) to the society, but before it joins the society it is registered at the central server. We argue that semi-open societies only slightly limits the openness compared to completely open societies, but have a much larger potential for providing stability and trustfulness. For instance, it is possible to monitor which agents are currently in the society. This also makes the boundary of the society explicit.

4 Semi-closed Agent Societies In what we will refer to as semi-closed societies, external agents are not allowed to enter. However, they have the possibility to initiate a new agent in the society which will act on the behalf of the external agent. This is done by contacting a kind of institution representing the society and ask for the creation of an agent (of one of a number of predefined types). The institution then creates an agent of this type,

which enters the society with the goal of achieving the goals defined by the external agent. (See Figure 2.)

2 1

Figure 2: An agent initiating a representative in a semi-closed society.

All agents are run on the same (set of) servers. Typically, the agents are implemented and the servers are managed by a third party, i.e., the owner of the society. As the possible behaviors of all agents are known, it is easier to control the activities in the society. The following example of a semi-closed society is based on the activity of searching for and booking “last-minute” holiday travel tickets. The price of this type of tickets may change on a daily basis and are determined by current supply and demand. For instance, if the whether is nice at the place from where the travel departures, prices on travel tickets to Mediterranean beach resorts typically drop as the departure date approaches. Today, customers find and buy tickets manually by browsing newspaper ads and WWW-pages, as well as phoning and visiting local travel agents. Most people regard this as a time-consuming and boring activity and would probably be happy to let software agents do the job. The society is based on an existing prototype implementation. 1 Customers specify their preferences (departure date, destination, max. price etc.) through either a WAP or WWW interface. An agent is then initiated on a service portal (at a remote computer) with the goal of finding a ticket satisfying the customer preferences. It continually searches a number of databases until it reaches its goal (or is terminated by the customer). When a ticket is found, the agent either books it directly, or sends an SMS to the customer asking for confirmation. To book the ticket, the customer agent contacts an agent representing the travel company (the info in the database 1

The system was developed during the first five months of 2000 by a group of 14 students doing their final exam project for their B.Sc. theses at Blekinge Institute of Technology, under the supervision of the author. It was a joint project with HP, who donated servers, workstations, and E-speak licences, and Aspiro, a Swedish company in the area of Internet-based services, who contributed knowledge about development of Internet-based services and reuse components. E-speak is a platform for service management developed by HP and was used for agent administration and communication.

contains the address to the travel agent). When the customer agent receives a confirmation, it immediately sends an SMS message to the customer about this and then terminates. The description above corresponds to the current version of the system. However, it is easy to imagine possible future extensions, e.g.: - letting the customer agent and the travel company agent negotiate about the price, - auctioning off the remaining tickets when there is only a specified number of hours or days left until departure, - letting the travel company agents autonomously decide the ticket price based on available information about the current situation (e.g., supply, demand, and number of days before departure). Semi-closed societies provide almost the same degree of openness as semi-open societies but are less flexible. On the other hand, they have a larger potential for implementing stability and trustfulness. An interesting property of the semi-closed societies is that they seem to indicate the limit of how open a society could be where the owner of the society may still control the overall architecture of the society. To have control over the architecture is a prerequisite for applying many of the ideas on how to achieve multi-agent coordination (cf. Lesser [7]). Moreover, this type of society poses interesting questions regarding ownership: Who is actually responsible for the actions of the agents?

5 Conclusions We have described four different categories of artificial societies that balance the trade-off between openness, flexibility, stability, and trustfulness differently. Based on the analysis of completely open and completely closed societies, which revealed that open societies support openness and flexibility but not stability and trustfulness and that the opposite is true for closed societies, we suggested two other categories, namely semi-open and semi-closed societies. We argue that these types of society let us have the best from both worlds. Whereas semi-open societies are more flexible than semi-closed societies, they have lower potential to achieve stability and trustfulness. A summary comparison between the different types of societies (ordered in degree of openness) is provided in Table 1. The balancing of trade-off provided by the semi-open and semi-closed societies is necessary for implementing systems that can be characterized as information ecosystems. In this type of systems there is a strong need for mechanisms for “enforcing” ethical behavior between agents in order to provide trustful systems to end-users. In completely open societies such mechanisms probably need to be very complex (if they exist at all), which means that the potential for achieving trustful systems is very low. In completely closed systems, on the other hand, the potential for achieving trustfulness is great, but the price you have to pay, by making it im-

possible for new agents/owners to enter the society, is too big in applications where openness is desired.

Table 1. A comparison between the different types of societies. Fixed Agents

Fixed at design time Constraints Fixed Communication Fixed

Closed

Semi-closed

Semi-open

Open

Anarchic

Known at design time Fixed Fixed

Known at run time Fixed Fixed

Known at run time Fixed Fixed

Cannot be known Not fixed Not fixed

Usually fixed Cannot be monitored Can be known yes

Cannot be known Not fixed Usually fixed Usually fixed Cannot be monitored Cannot be known no

language Roles

Fixed

Fixed

Fixed

State

Can be monitored Fixed at design time yes

Can be monitored Fixed at design time yes

Can be monitored Can be known yes

Agent owners Society owner

Not fixed Cannot be monitored Cannot be known no

5.1 Future Work In this work we focussed on four properties of artificial societies that we believed were the most important. However, there are a number other properties that may be relevant in some domains. Examples of such properties are: fairness, i.e., the degree to which the society members are treated equally, performance, i.e., how efficient is the society (measured in e.g., time or number of messages needed to perform a certain task), and "controllability", i.e., the support for the owner of the society to control the activities in the society, e.g., by punishment. Future work will take also take these and other relevant aspects into account. An important part of the future work is the definition of appropriate metrics for quantifying the properties of artificial societies and developing methods for measuring them. Also, methods for determining how well different types of artificial societies balance the trade-off between these aspects need to be developed. Such methods could be based on simulation experiments and/or theoretical analyses. Finally, both theoretical and practical aspects of institutions need further investigation, answering questions such as: What functionality is possible to implement and what is not? Acknowledgements The author acknowledges the valuable contribution from the colleagues in the ALFEBIITE project (IST-1999-29003) and the members of the Societies of Computation research group at Blekinge Institute of Technology.

References 1. A. Artikis and J. Pitt. A Formal Model of Open Agent Societies, In Proceedings of the Fifth International Conference on Autonomous Agents, 2001. 2. K.M. Carley and L. Gasser. Computational Organization Theory. In G. Weiss (editor), Multiagent Systems, MIT Press, 1999. 3. P. Davidsson. Emergent Societies of Information Agents, In Cooperative Information Agents IV, Springer Verlag LNCS series, Vol. 1860, 2000. 4. J. Ferber and O. Gutknecht. A Meta-model for the Analysis and Design of Organizations in Multi-Agent Systems. In Proceedings of the Third International Conference on MultiAgent Systems, IEEE Computer Society, 1998. 5. R. Gustavsson and M. Fredriksson. Coordination and Control of Computational Ecosystems: A Vision of the Future. In: A. Omicini, M. Klusch, F. Zambonelli, and R. Tolksdorf, editors, Coordination of Internet Agents: Models, Technologies, and Applications. Springer Verlag, 2001. 6. S.J. Johansson and J. Kummeneje. A Preference-Driven Approach to Agent Systems. To appear in the Proceedings of the Second International Conference on Intelligent Agent Technologies, 2001. 7. V.R. Lesser. Reflections on the Nature of Multi-Agent Coordination and Its Implications for an Agent Architecture. Autonomous Agents and Multi-Agent Systems, Vol. 1:89-111, Kluwer, 1998. 8. A. Oram (editor). Peer-to-Peer: Harnessing the Power of Disruptive Technologies, O’Reilly, 2001. 9. M.P. Singh, A.S. Rao, and M.P. Georgeff. Formal Models in DAI, In G. Weiss (editor), Multiagent Systems, MIT Press, 1999. 10.F. Zambonelli, N.R. Jennings, and M. Wooldridge. Organizational Abstractions for the Analysis and Design of Multi-Agent Systems. In P. Ciancarini and M. Wooldridge, editors, Agent-Oriented Software Engineering. Springer Verlag LNCS series, Vol. 1957, 2001.

Categories of Artificial Societies

browser processes together with the set of WWW-server processes that are con- nected to ... Closed agent societies are typically those where a Multi-Agent System (MAS) ap- ... In order to get access to other users' files, a Napster software pro-.

54KB Sizes 3 Downloads 219 Views

Recommend Documents

Categories and Haskell - GitHub
This is often summarized as a side-effect free function. More generally ... The composition g ◦ f is only defined on arrows f and g if the domain of g is equal to the codomain of f. ...... http://files.meetup.com/3866232/foldListProduct.pdf ... Pag

ePermit Beneficial Use Categories
Mar 26, 2013 - Instream Flow-only State of Wyo can apply. LAK. Maintain Natural Lake Level (Phase II Award). MUN_SW. Municipal-- Surface water. NAT.

Derived Categories
D ⊆ D(A) denote the full subcategory corresponding to K. Let q (resp. qB) denote the localization functor K → D (resp. K(B) → D(B)). (1) A right derived functor of F : K → K(B) is a triangulated functor of triangulated categories. RF : D −â

registered societies Accounts
Apr 12, 2016 - may not always be in the best interest of the District and/or its schools to accept such funds. In ... April 2016 .... Social Host Liability Guideline.

Short Selling China - Societies
Feb 12, 2014 - China Inc. is changing. – All about “who” you know. – Foreign .... Chinese 'Apple' and 'IKEA' stores. – Shopping strips with Starbocks, King ...

Derived categories of resolutions of cyclic quotient ...
Abstract. For a cyclic group G acting on a smooth variety X with only one character occurring in the G-equivariant decomposition of the normal bundle of the fixed point locus, we study the derived categories of the orbifold [X/G] and the blow-up reso

human societies
of forces that translate into the observable behavior of com- mon substances. .... A very heterogeneous fluid may mix under one of two conditions: 1) by ...

Short Selling China - CFA societies
Feb 12, 2014 - All about “who” you know. – Foreign ... Recognized social financing comprises 194% of 2013 GDP .... Source: China Media Project. Sources: 1) ...

Categories, stereotypes, and the linguistic perception of ...
examine the linguistic perception of sexuality in its wider social context, and, as ... There is a popular belief that speech is a reliable marker of an individual's sexuality, ... speaking in more formal contexts are more likely to be perceived as f

On Some Categories of Involutive Centered Residuated ...
for the involution. We call eiRL the full subcategory of ciRL determined by ... a center c which satisfies the “Leibniz condition”(see Definition 7.2). We denote this ...

Derived categories and the genus of space curves
Abstract. We generalize a classical result about the genus of curves in projective space by Gruson and Peskine to principally polarized abelian threefolds of Picard rank one. The proof is based on wall-crossing techniques for ideal sheaves of curves

Simulating Computational Societies
and implementors, this can be used to model, analyse and explain the .... model and the software developed to facilitate experimentation with agent so-.

LOCALISATION AND QUOTIENT CATEGORIES ...
The part of the story that we are concerned ...... commutativity axioms upto isomorphism, see [ML98, Ch.VII§.1 and §.2] for the definition and basic properties.

On the evolution of coarse categories
If, however, there are substantial costs to categorization such as a reduction in decision making .... individual could learn (subject to errors) action profiles from others .... example provides a numerical illustration of the intuition behind. Resu

Estimating the Aspect Layout of Object Categories
as robotics, autonomous navigation and manipulation. In ..... ergy values as opposed to the energy values themselves. From the point of view of ..... mance of our algorithm with [27], we bin our viewpoint .... unsupervised scale-invariant learning. I

Infants attempting to learn the phonetic categories of ...
Statistical learning, cross-constraints and the acquisition of speech categories: a computational approach. Joseph Toscano. Bob McMurray [email protected] [email protected]. Dept. of Psychology. Dept. of Psychology. University of Iowa. Un

Identification of Rare Categories Using Extreme Learning Machine
are useful in many fields such as Medical diagnostics, Credit card fraud detections etc. There are many methods are use to find the rare classes, they are ...

Cerebral Performance Categories Scale CPC Scale - FDA
CPC 1. Good cerebral performance: conscious, alert, able to work, might have mild ... Coma or vegetative state: any degree of coma without the presence of all ...

Antisocial Punishment Across Societies - CiteSeerX
Oct 9, 2007 - A 360,. 2475 (2002). 42. B. Steinberger, Phys. Earth Planet. Int. 164, 2. (2007). 43. ..... views about acceptable behaviors and the devia-.

Towards long-term visual learning of object categories in ... - CiteSeerX
50. 100. 150. 200. 250. 300. 350. 400. Fig. 3. Histogram of hue color component in the image of Fig. 2 .... The use of the negative exponential has the effect that the larger the difference in each of the compared ... As illustration, Figs. 6 and 7 .

Donaldson-Thomas theory for categories of homological dimension ...
Dec 30, 2015 - all known realizations, i.e. mixed Hodge modules, perverse sheaves and ...... bi-exact. These data have to satisfy properties similar to the one of ...

Towards long-term visual learning of object categories in ... - CiteSeerX
learning, one-class learning, cognitive, lists of color ranges. 1 Introduction ... Word meanings for seven object classes ..... As illustration, Figs. 6 and 7 show the ...

Identification of Rare Categories Using Extreme Learning Machine
are useful in many fields such as Medical diagnostics, Credit card fraud detections etc. There are ... Here the extreme learning machine is use for classification.ELM is used .... Generative Classifiers: A Comparison of Logistic Regression and.