FIPA 96/04/17
Source: L. Chiariglione fipa_rationale.htm




Agent has become a fashionable technoword of the late nineties. Still, there is no universally agreed definition of an "agent" beyond those found in dictionaries.

A strawman definition of agent used in this document is: an entity that resides in environments where it interprets "sensor" data that reflect events in the environment and executes "motor" commands that produce effects in the environment. An agent can be purely software or hardware. In the latter case a considerable amount of software is needed to make the hardware an agent.

Building upon this definition some attributes of agents can be derived, not all of which need to be actually present in an agent:

autonomy agents can operate without the direct intervention of humans or others
social ability agents can interact with other agents and/or humans
reactivity agents perceive their environment and respond in a timely fashion to changes that occur in it
pro-activeness agents can exhibit goal-directed behaviour by taking the initiative
mobility agents can move to other environments
temporal continuity agents are continuously running processes, not "one-shot" computations that terminate
adaptivity agents automatically adapt to changes in their environment

By combining some of these attributes different types of agent can be identified:

Autonomous agents agents that inhabit some complex, dynamic environments, sense and act autonomously in this environment and by doing so realise a set of goals or tasks.
Entertainment agents interactive, simulated worlds providing entertainment to a user
Information agents agents that have access to potentially many information sources and are able to collate and manipulate information obtained from these sources to answer queries posed by users and/or agents
Intelligent agents agents that carry out some set of operations on behalf of a user or another program with some degree of independence.
Interface agents agents that provide assistance to users dealing with another agent

Several other attributes can be found in the literature in combination with agents: cognitive, commerce, search, situated, software, synthetic, user and so on, but no attempt will be made here to define their meaning.

The lack of definition of the word agent and the presence of so many attributes associated with it suggest that there exist many instances of agents because agents are a common technology that is utilised by a number of diverse fields.

This happens frequently in most instances of information technology, that are not treated for what they are, i.e. generic technologies to be customised by different application fields. The usual approach of an application field is to use some information technology tools to solve its own problems. This results in a waste of efforts at the time the technology is developed and, more worrisome, inhibits interoperability of applications because independent developments usually lead to incompatible technical solutions. A case in point is provided by security, a key technology for anything digital, which is independently being specified by the entertainment, telecommunications and financial transaction fields, in many cases by individual players within each of these communities.

Agent technologies are the last example of this approach. Thanks to the efforts made by academia and other research institutes considerable progress has been made and some products that claim to (and sometimes do) use agent technologies have begun to appear. It is therefore already possible to attempt the integration of several of the key technologies and provide specifications for use by the industry. Availability of such specifications will have three main benefits:

  1. technological because of economies of scale that will accelerate the deployment of products and applications;
  2. enabling because there are several instances of products and applications that become possible only when common standards exist;
  3. synergetic because different applications will benefit from a higher level of interoperability.

While the benefits of uniform specification and standardisation do not escape the attention of application and product developers, there is understandable reluctance on the part of many to commit resources in what are perceived as processes whose timing of result availability are unknown because they are bureaucratic, passive, driven by an extreme search of consensus and often poorly managed. Under such circumstances it makes little surprise that companies resort to agreements between individual business players. These are inevitably antagonised by other players out of fear that an advantage will be gained by a competitor because of the adoption of a particular technology that is unfamiliar or simply owned by others or because the technology that needs to be chosen is based on specific business models not shared by other players.

In recent years the response has been the establishment of industry consortia that have promised, and in many cases delivered on the promise, a faster development of specifications with a fair process. The case being considered here, however, is more complex because what needs to be specified is

Two examples of mechanisms that were successfully applied to produce generic specifications similar to those needed for agent technologies will be considered.

  1. In the ISO/IEC Moving Picture Experts Group ( MPEG) standards for audio-visual coding were collaboratively developed by completely different industries all interested in a common generic technology. In spite of participating industries having different sets of requirements, these were all accommodated by developing and standardising technical "tools" that satisfied those requirements. The tools that were needed to support the most widely supported requirements were introduced in the basic set (Main Profile), the tools needed for less supported profiles were moved to other sets (Higher Profiles).
    The result of the huge effort made to produce the MPEG-1 and MPEG-2 standards has been the timely and volume availability of this key technology where development costs were shared by all industries while each industry could customise the generic technology for its own use. Interoperability because of a common coding algorithm was the added advantage from the effort.
  2. In the Digital Audio-Visual Council ( DAVIC) more than 200 companies from 25 countries have produced complete end-to-end specifications of broadcasting and interactive digital audio-visual applications in less than two years. Applications being manifold and sometimes not well defined, DAVIC considered the requirements of a representative set of applications and identified commonalities of requirements across applications. Then the technical "tools" that would support the requested requirements were developed and specified. Tools were then grouped in Profiles. The "System Integration" Technical Committee constantly monitored the consistency of the major systems that could be built using the tools so specified.
  3. The result of the huge and intense effort to produce the DAVIC 1.0 specification has been the reusability of different key technologies in different application domains with benefits from greater volumes for component technologies and accrued interoperability across applications.

In conclusion MPEG was successful in providing a single technology for a multiplicity of industries and DAVIC was successful in integrating largely existing technologies in a consistent whole. The MPEG case seldom happens in standards bodies and industry consortia because they tend to be organised along traditional business demarcation lines. The DAVIC case seldom happens in standards bodies and industry consortia because work there is usually assigned to small working groups that operate on rather small technical fields. Technical coordination is usually missing because committees sitting above the technical level deal mostly with bureaucratic matters. The system integration attempts like OSI and ODA failed largely because of the time it took to produce the standards.


Back to FIPA homepage


Dr. Leonardo Chiariglione +39 11 228 6129/6116/5111
Vice President, Multimedia +39 11 228 6299/6190/5520
CSELT leonardo.chiariglione@tilab. com
Via G. Reiss Romoli, 274
I-10148 Torino (ITALY)