The Minerva Multiagent System for Supporting Creativity in Museums Organization. Francesco Amigoni and Viola Schiaffonati. Dipartimento di Elettronica e ...
The Minerva Multiagent System for Supporting Creativity in Museums Organization Francesco Amigoni and Viola Schiaffonati Dipartimento di Elettronica e Informazione Politecnico di Milano Piazza Leonardo da Vinci 32, 20133 Milano, Italy Abstract The study of computational tools for supporting human creativity is a topic of increasing importance in current artificial intelligence research. In this paper, we present the Minerva multiagent system able to support the creative activity of museum organizers. Minerva is a computational tool that helps the human organizers in some tedious operations and enlarges their creativity space by favoring the fast development of virtual prototypes of possible museum organizations. Minerva has been experimentally tested in two different versions that are discussed in the paper.
1
Introduction
The study of computational tools for supporting human creativity is a topic of increasing importance in the current artificial intelligence research. These tools are oriented both to artistic creativity [Edmonds and Candy, 2002b] and to scientific creativity [Langley, 2002]. In this paper, we present a multiagent system, called Minerva, that supports the creative work of human museum organizers. In this scenario, the focus is on the user’s creativity and not on the system’s one. Minerva is a computational tool that simulates the organization of museums and helps the human organizers in two different ways. First, it performs tedious tasks, such as the verification of the compliance of a possible allocation of works of art with the current legislation; this enlarges the creativity space of the museum organizers by accelerating the production of virtual prototypes of possible museums organizations. Second, Minerva helps the users in the visit and fruition of (automatically organized) virtual museums. We explicitly stress that Minerva is not intended to be a commercial application, but rather an academic prototype in which different computational methods to support creativity can be tested. The academic nature of Minerva impacts on some design choices that keep the complexity of the system to a minimum by mainly sacrificing the usability of the user interface. Minerva has been experimentally implemented in two different versions to validate the flexibility and the generality of its architecture. The first version arranges the archeological finds of a collection and allocates them in historical
buildings; the second version arranges a collection of design objects and allocates them in imaginary virtual environments. This paper is organized as follows. The next section illustrates some typical scenarios of employment of Minerva and its technical implementation. Section 3 analyzes the role of Minerva in supporting human creativity. In Section 4 we collocate Minerva within the current research trends in machine creativity. Finally, Section 5 concludes the paper.
2
The Minerva Multiagent System
The main purpose of Minerva is the automatic organization of museums, given the works of art and the environments in which they must be displayed. By adopting the term ‘organization’ we mean both the preparation and the allocation of a museum or an art exhibition. The preparation is the arrangement of the works of art in conceptually relevant ordered groups. The allocation is the placement of these works of art within a given geometrical space preserving the arrangement imposed by the preparation process. In what follows, we describe how Minerva is used and some of its implementation details.
2.1
Scenarios of Use
The task of organizing a museum is usually thought as a typical human-exclusive activity, based on aesthetic and personal ineffable criteria. The final result of a museum organization is a global coherent setting that allows an effective visit of the museum. Among the variables to be taken into account in the organization there exist some “physical” ones: for instance, some works of art cannot be inserted in some rooms, some rooms can be scarcely lighted, some other rooms can be too hot or too cold to keep works of art in them. In addition, artistic criteria must be considered and play a central role in the final result, such as how to give emphasis to particularly important works of art or how to build connections among different works of art. In designing Minerva we have emphasized flexibility and generality with the aim of managing different kinds of works of art and different environments, as demonstrated by the implementations of two versions of the system. Both have the same architecture (illustrated in the next subsection), but different interfaces and contents. The first version of Minerva organizes the archeological finds of a collection in historical buildings (Minerva archeological); the second version of Minerva organizes a collection of design
objects in imaginary virtual environments (Minerva design). Both versions have Italian interfaces because they have been firstly designed for Italian users. A typical session of a user with each one of these versions is described in the following. Let us start from the Minerva archeological. The user interacts with an interface that guides him or her in the organization and in the visit. Minerva users can be roughly grouped in four different typologies, each one corresponding to a different pattern of behavior. The museum expert and the museum organizer users use Minerva for simulating potential organizations (controlled by the users themselves) of the works of art in some environments to study and evaluate them. The artistic expert users use Minerva by making requests on the works of art to be displayed, but who do not interfere in the organization of the museums. The inexperienced users make virtual visits of already organized museums. The technical users modify the system, for instance to update the database of the collections of works of art and that of the environments. For the purposes of this paper we concentrate just on museum organizers. They are the most representative users in illustrating the functioning of Minerva since they have the complete control over its parameters, especially at the light of the considerations of Section 3. Nevertheless, to explain some functionalities we will refer in some cases to the users belonging to the other typologies as well. In Fig. 1 the user has chosen to create a museum in which all the archeological finds stored in the database of Minerva archeological are displayed in the rooms of the Caserma Napoleonica (Napoleonic barracks), an historical building in the center of Milan (Italy). Other buildings could be chosen and several criteria to prune the collection of works of art (according to their type, age, culture, material, and place of origin) could be applied. The user can tune the various parameters and criteria driving the processes of preparation and allocation of the selected works of art in the selected environment (Fig. 2). The parameters of the preparation process determine the order in which the works of art are displayed (according to their shape, production technique, type, age, culture, material, and place of origin) and the conceptual criterion (historical or artistic) by which the works of art are classified in consistent and coherent groups. The parameters of the allocation process control all the aspects of the placement of the works of art in the environment, including the maximum and minimum occupation of the rooms and of the shrines, the order (clockwise or anticlockwise) in which the works of art are placed in the rooms, and the algorithm by which the taxonomic tree produced by the preparator agent (see next subsection) is visited. Some of these parameters (like shape, production technique, . . . ) are relative to the nature of the works of art to be organized; those presented here are relative in particular to archeological finds. When the preparation and the allocation activities have been performed, the user can visit the automatically organized museum in a VRML virtual reality setting, as shown in Fig. 3. All the works of art are virtually reproduced according to their three-dimensional models obtained from the real objects or from their representations (e.g., pictures). In the virtual museum, the user can follow a list of viewpoints and access the detailed information (including a better VRML
Figure 3: The VRML virtual visit of the automatically organized archeological museum
model and a number of textual descriptions at various levels of detail) associated to the works of art by clicking on the works of art themselves (Fig. 4). The information can also be accessed by a rudimentary natural language interface. The second version of Minerva is devoted to the organization of a museum in which a collection of design objects that have been awarded with the Compasso d’Oro (the most prestigious Italian design award) are displayed in a totally imaginary virtual environment, shaped as a spiral whose dimension depends on the number of selected design objects. The user performs the same operations as in the previous case. Specifically, the user tunes some parameters that select the design objects to be inserted in the museum (the spiral imaginary environment is selected by default since it is the only imaginary environment we have considered so far) and that constrain the preparation and allocation processes (Fig. 5). In this case, the parameters are the following: the years of awarding; the object degree of innovation in shape, function, material, manufacturing process, structure, and communication; the area of use (house, work, communication, entertainment, and movement). The prepared and allocated museum is displayed (with SHOCKWAVE technology) to the user (Fig. 6). As in the previous case, the user can navigate in the automatically organized museum and can request more information about the displayed design objects (Fig. 7).
2.2
Technical Implementation
Minerva is implemented as a multiagent system [Weiss, 1999] composed of autonomous computational entities, called agents [Russell and Norvig, 2002]. Museum organizations are the result of the cooperation activity of the agents of Minerva. As the previous subsection shows, the architecture of the system is designed to manage any kind of works of art, from archeological finds to design objects. A preliminary description of the system can be found in [Amigoni et al., 2001]. The main distinction among the entities composing Minerva is between agents and components. The agents pro-
Figure 1: The pruning criteria and the environment (the Caserma Napoleonica building) in which to organize a collection of archeological finds
vide fundamental “intelligent” and autonomous functions, whereas the components provide peripheral “service” functions. The user interface is managed by the components that exploit the functions of the agents to provide a performance to the user. The agents and the components of Minerva have been designed to be modular and independent in their activation. As a consequence, the structure of Minerva dynamically changes according to the activities of the system. The architecture of Minerva is shown in Fig. 8. This is a static representation of the composition of Minerva, in which only the main communication flows are shown, even if all the agents and the components can communicate together in an ad-hoc internal message format. The agents and the components are coded in JAVA; the agents employ JESS [Friedman-Hill, 2003] to perform their inferential activities. In the following we describe the components and the agents of Minerva. The supervisor component “recruits” and activates the agents and the components that are needed to carry out a given activity. For example, during the organization of a museum, the preparator and allocator agents are the only active agents; instead during the visit of an organized museum, only the navigator and commentator agents are active. The user interface component generates the pages displayed to the user via a HTML browser and manages the in-
teraction with him or her according to what discussed in the previous subsection. The user interface component translates the geometrical description of the allocation of a collection in an environment, which is produced in the Minerva internal format, to the format displayed to the user (VRML [Tittel et al., 1996] or SHOCKWAVE [Lengel, 2003]). More precisely, the user interface component reads (from the works of art database) the three-dimensional models of the works of art and inserts them in the three-dimensional model of the environment (read from the environments database) according to the geometrical result of the allocation process. Then, the VRML or SHOCKWAVE model is displayed to the user who can navigate inside it. We underline that the modular architecture of Minerva allows to create different versions of Minerva by substantially maintaining the kernel (the four agents described below) and some of the components and by radically changing only the user interface component, as demonstrated by the two versions illustrated in the previous subsection. The natural language processing component translates the requests of the users from (a subset of) Italian to the Minerva internal message format. The DB management component translates the queries between the internal message format and SQL. The databases of Minerva are the works of art database and the environments database.
Figure 2: The criteria and parameters of the preparation (top) and allocation (bottom) processes for an archeological museum
Figure 5: The criteria and parameters of the organization for a design museum
Figure 4: The window for accessing the detailed information about a stele
MINERVA SYSTEM AGENTS ALLOCATOR
NAVIGATOR
PREPARATOR
COMMENTATOR
SUPERVISOR
COMPONENTS DB MANAGEMENT
USER INTERFACE
NATURAL LANGUAGE PROCESSING
HTML - JSP - APPLET DB
HTTP SERVER
USER
Figure 6: The automatically organized design museum displayed to the user
Figure 8: The architecture of Minerva Let us consider now the agents that constitute the kernel of Minerva. Given a collection of works of art and an environment (selected by the user), two agents operate to find, respectively, a preparation and an allocation. The first agent, called preparator agent, automatically determines, on the basis of the criteria selected by the user, the conceptual arrangement in which the works of art of the collection will be displayed. More precisely, the preparator agent uses the JESS engine, namely inferential techniques, to classify the works of art of the selected collection into groups, each one composed of homogeneous objects. The preparator agent exploits the expertise inserted in its knowledge base about museum criteria concerning the possible ways in which the pieces of a collection can be classified and grouped. This expertise has been acquired from human experts in museum organizations. The criteria the preparator agent adopts to create groups of homogeneous objects are based on both cultural (such as age, type, place of origin, and culture of origin) and physical (such as size, weight, shape, and material) features of the objects. The preparation criteria are expressed as rules in JESS; for example the following rule pairs two works of art that are similar given the selected preparation features:
Figure 7: The window for accessing the detailed information about the Detector glasses
(defrule Pair-Objects (selected-criteria ?head $?tail) (object (name ?name1)) (object (name ?name2)) (and (neq ?name1 ?name2) (eq (extract-feature ?name1 ?head) (extract-feature ?name2 ?head))) (not (exists (pair ?name2 ?name1))) => (assert (pair ?name1 ?name2)))
The archeological finds of the previous subsection can be grouped, for example, according to a chronological criterion, based on the age of the pieces, or to a type criterion, based on the type of the pieces (e.g., stele, reliquary, statue, and so on).
MUSEUM
Age: I century BC
Culture: Greek
Type: statue
1
Culture: Roman
Type: vase
Place of origin: Crotone
Type: ring
Place of origin: Siracusa
Place of origin: Crotone
Type: candelabra 8
Place of origin: Girgenti
Representation: mythological
Representation: agrestic
Representation: mythological
Representation: funeral
2
3
4
5
6
7
Figure 9: The taxonomic tree of some archeological finds from I century BC Figure 10: The grid model used by the allocator agent These criteria can also be combined together to obtain, for example, a first division based on the subject, a second subdivision based on the type, and then a further subdivision based on the age. We have implemented two sets of preparation criteria: one for the archeological finds and one for the design objects. The central feature in both cases is that the user can tune the application of these criteria by acting on some parameters. The preparator agent stores the obtained groups of works of art in a taxonomic tree (Fig. 9). Each leaf of the tree represents a group of works of art that are conceptually homogeneous according to the selected criteria. The groups are ordered according to their importance. The importance is determined as a function of the works of art in the group and of the selected criteria. In conclusion, the preparator agent performs firstly a classification of the works of art in groups and then an ordering of these groups. The ordered groups produced by the preparator agent and the environment selected by the user constitute the input for the allocator agent. The allocator agent automatically finds, within a given environment, the best geometrical allocation for the works of art of the groups in the taxonomic tree, preserving the ordering on groups imposed by the preparator agent. Also the allocator agent exploits the knowledge inserted in its knowledge base about museum criteria concerning where to place the works of art in a given environment. These criteria consider, among others, the minimum and maximum occupation of the space in a room, the minimum and maximum occupation of the perimeter of a room, and the distances governing the collocation of the shrines according to the collocation of the doors, windows, and other shrines. The allocation criteria are expressed as rules in JESS; for example the following rule determines if a group of works of art can be allocated in a room: (defrule Allocate-A-Group-in-A-Room (min-Perimeter-Occupation ?mip) (max-Perimeter-Occupation ?map) (min-Area-Occupation ?mia) (max-Area-Occupation ?maa) (group ?group ?area-g ?perimeter-g) (room ?room ?area-r ?perimeter-r) (and (>= ?area-g (* ?area-r ?mia))
(>= ?perimeter-g (* ?perimeter-r ?mip)) (
