Development of Integrated Metamedia Environment

based on Agent Modeling

1. What are metamedia?

Metamedia are intelligent media that mediate other media. Metamedia depict data contents and interact with other media or the human end users to represent a multimedia application. In order to do so, an agent approach is taken: each metamedium is a combination of media data and agent program. To communicate with other media, metamedia architecture utilizes CORBA-compliant object request broker. However, a higher level or agent level of standardization is necessary to ensure the interactivity of metamedia agents in a heterogeneous networked environment.

2. Why metamedia?

In recent multimedia applications, distributed media data are integrated by a central console or a "browser." For example:

        +-+     +-+
        |M|---->|B|     +-+
        +-+     |r| MM  |U|
        +-+     |o|====>|s|
        |M|---->|w|     |e|     M:       Media data
        +-+     |s|     |r|     Browser: Netscape, Mosaic etc.
        +-+     |e|     +-+     MM:      Multimedia representation
        |M|---->|r|      |      User:    Human end user
        +-+     +-+

WWW browsers, such as Netscape and Mosaic, gather media data from various URLs to represent a multimedia information to the human end user. The problem of this schema is that communication is asymmetric. Users can browse the multimedia data easily and efficiently. But publishing the user data is quite difficult. You must either write an HTML document that HTTP conveys to other browsers in the same asymmetric manner, or use CGI that capability is so limited that you can just send some character data, not multimedia data.

The following picture illustrates a possible metamedia environment:

   news source ---  newspaper --- user
   medium (1)  \ /  medium (1)\   medium (1)\
                X              \             \  * encounter of media *
   news source /_\  newspaper   \  user       \ virtual mall
   medium (2)  \ /  medium (2)\--\ medium (2)---medium
                X              \              / |||
   news source / \  newspaper   \  user      /  ||+- shop medium (1)
   medium (3)  ---  medium (3)---\ medium (3)/  ||+--shop medium (2)
                                                +--- shop medium (3)
         * freedom of choice *

Each medium is created equal. A medium is an agent that represents the data. Communication among the agents is symmetric. With this integrated metamedia environment, several advantages over existing legacy multimedia are obvious. One is an intelligent freedom of choice that agent assists human end user according to the human's preferences. For example, you can customize your own newspaper medium that automatically connects to news source media and compiles an article. The other possibility is an encounter of media. Your virtual agent, or avatar, may dive into a virtual mall where people talk to each other, purchase goods, or even open their stores. Other applications expected include: virtual city, learning assistance, media publishing, agent conference and so on.

3. Agent model

Metamedia architecture defines three main components: Agent Description Language, Language Mapping and Agent Platform.

  [Agent Platform] 
    intra and inter agent space protocols
    agent protocol for content agent, media agent and user agent
    --------------------------------------- [Language Mapping]
  [Agent Description Language]
    GUI extension and object service extension
    network extension   
    scheme kernel
    media extension
    media data

The Agent Description Language is based on Scheme, a Lisp dialect. Unlike huge Common Lisp, Scheme is a compact yet powerful language. Unique features to Scheme such as continuation and closure, along with reflective nature of Lisp are quite appropriate for describing the reflective media or metamedia. The media extension to the Scheme kernel enables ADL to handle media data as a primitive element of the language. GUI extension and object service extension are interfaces to the human end users and other agents via CORBA compliant object request broker, respectively. The network extension under GUI and object service extensions handles low level communication protocols.

The Agent Platform is a collection of agents communicating each other. The basic agent types are content agent, media agent and user agent. There are two types of communication: one is an intra agent space communication and the other is an inter agent space communication. The agent space is a logical domain or thread where execution of the agents is synchronous. Physically an agent space may be a single computer or computers connected to a high bandwidth, reliable LAN. On the other hand, the inter agent space communication is asynchronous. It might be an unreliable and slow WAN environment.

The Language Mapping is an interface layer between the Scheme-based ADL and the agent platform.

4. Standardization

The following is a list of protocols that are being developed under the metamedia project. A generalization and standardization of the protocols are needed.

content agent (1) acquisition of knowledge
content agent (2) description of knowledge
content agent (3) application of knowledge
content agent (4) acquisition of communication
content agent (5) description of communication
content agent (6) application of communication
media agent (1) acquisition of knowledge
media agent (2) description of knowledge
media agent (3) application of knowledge
media agent (4) acquisition of communication
media agent (5) description of communication
media agent (6) application of communication
user agent (1) acquisition of knowledge
user agent (2) description of knowledge
user agent (3) application of knowledge
user agent (4) acquisition of communication
user agent (5) description of communication
user agent (6) application of communication
intra agent space cooperation protocol
inter agent space cooperation protocol

May 22, 1996

Hiroki Suguri (suguri@comtec.co.jp)

Comtec Co., Ltd.