An adaptive agent oriented software architecture

Babak Hodjat, Christopher J. Savoie, Makoto AmamiyaLecture Notes in Computer Science 1998 · DOI · OpenAlex

Abstract

A new approach to software design based on an agent-oriented architecture is presented. Unlike current research, we consider software to be designed and implemented with this methodology in mind. In this approach agents are considered adaptively communicating concurrent modules which are divided into a white box module responsible for the communications and learning, and a black box which is the independent specialized processes of the agent. A distributed Learning policy is also introduced for adaptability.

1. Introduction

In the late 1990s, software engineering faced a transition point. While Object-Oriented Programming (OOP) provided a robust framework for modularity, it often resulted in passive components that lacked "volition"—the internal drive to evaluate results and adapt behavior. The Adaptive Agent Oriented Software Architecture (AAOSA) was proposed to address this by treating software as a "society" of active, learning agents rather than a static hierarchy of commands.

Traditional agent-oriented systems often functioned as wrappers for legacy code or relied on centralized "facilitators" for message routing, creating bottlenecks. AAOSA aims to build software from the ground up as a network of independent agents capable of cross-language communication and autonomous learning.

2. The AAOSA Architecture

The fundamental unit of AAOSA is the agent, which is bifurcated into two distinct modules: the White Box and the Black Box. This separation ensures that the social and communicative logic is decoupled from the specialized functional logic.

Internal structure of an AAOSA agent, illustrating the separation between the communication-focused White Box and the task-focused Black Box.

2.1 The White Box (Communication & Learning)

The White Box serves as the agent's interface to the rest of the system. It contains several key components:

  • Interpreter: Evaluates incoming requests to determine if the agent can fulfill them or if they should be delegated.
  • Address-Book: A dynamic directory of other agents, allowing for decentralized communication.
  • Rewards Unit: Processes feedback from the environment or other agents to update the agent's internal logic.
  • Interpretation Policy: The decision-making logic that evolves over time based on received rewards.

2.2 The Black Box (Specialized Processing)

The Black Box contains the "Process"—the specific algorithmic implementation of the agent's task (e.g., a route-finding algorithm or a speech recognition engine). The rest of the system interacts only with the White Box, making the internal workings of the Black Box irrelevant to the broader network. This "transducer" approach allows heterogeneous codebases to be integrated seamlessly.

3. Distributed Learning and Adaptability

Adaptability in AAOSA is achieved through a distributed reward propagation system. When a task is completed, feedback is passed back through the chain of agents involved. Positive feedback increases the likelihood of selecting the same path in the future, while negative feedback prompts the agent to update its Interpretation Policy and Address-Book.

This mechanism allows the system to optimize itself at runtime. For instance, if a new, more efficient agent is added to the network, existing agents will eventually "learn" to route requests to it as it consistently yields higher rewards.

4. Comparative Analysis

The paper evaluates AAOSA against the Open Agent Architecture (OAA), a prominent framework of the era.

  • Modularity: AAOSA achieves higher modularity by partitioning routing and coordination logic among individual agents rather than concentrating it in a central facilitator.
  • Resilience: The decentralized nature of AAOSA's address books eliminates the single point of failure inherent in facilitator-based systems.
  • Parallelism: Agents are designed as concurrent modules, naturally supporting multi-platform execution and parallel processing across distributed networks.

5. Conclusion and Future Work

AAOSA represents a shift toward emergent software behavior. By embedding learning into the architectural fabric, software becomes capable of evolving to meet user needs without manual reconfiguration. Future research directions include "artificial evolution," where agents that fail to provide value are automatically phased out, paving the way for self-healing and self-optimizing software ecosystems.

Summary updated

BibTeX
@inproceedings{hodjat1998adaptive,
  author = {Babak Hodjat and Christopher J. Savoie and Makoto Amamiya},
  title = {{An adaptive agent oriented software architecture}},
  booktitle = {Lecture Notes in Computer Science},
  year = {1998},
  doi = {10.1007/bfb0095256},
}

Loading related content...

Figure