Agtivity

Software Agent Definition Issues

Let us know if you have any answer to any of these issues.

What is the full list of characteristics of software agents.
How can software agents be categorized.
Is there a "natural" taxonomy for software agents.
Should software agents be considered separately from robots or should a software agent merely be redefined as a "disembodied robot".
How is the software in a robot conceptually different from a software agent that roams a network.
How does a software agent differ from a "bot".
Are spiders and web crawlers software agents?
To what extent should software agents be considered to be completely separate from traditional programs, as opposed to a subset.
Should any of the following characteristics be required of software agents?

Intelligence

Mobility

Autonomy

What infrastructure requirements must be met by an "Agent OS" as opposed to the underlying OS used by non-agents. For example, mobility or intelligence.
How to categorize these radically different "forms" of agent:

"Merely" autonomous programs (e.g., daemons).

Multi-agent AI programs where the agents are not independent OS-level processes, but possibly merely data packets.

Autonomous, mobile agents which roam the internet and act "intelligently" on behalf of a user for extended periods of time without human intervention.

Programs which execute at a fixed site but access resources across the internet and otherwise act "intelligently" on behalf of a user for extended periods of time without human intervention.

Could an agent be merely a data packet sent to a remote server that "interprets" the packet and sends it on based on the results of processing the packet. Ultimately, that is what a Java program/agent actually is. But conceptually, could an agent merely be a group of agreed-upon data elements that a somewhat domain-specific server program can interpret and send around the network. And if that is not an agent, what is?

Does the "code" really have to be shipped around with the agent's state? After all, the code probably does not change, unless it is self-modifying, or an agent server tampers with it. Conceptually, an agent being sent over the net could be merely the data and state of the agent and an "id" that can be sent to an "agent repository" to retrieve the agent's code if it is not already present on the target system.

Characteristics of agents can be categorized as:

Internal. Not externally visible. For example, a measure of the agent's "intelligence".

External. Such as visible or measurable "behavior". For example, autonomy, mobility, and "apparent" intelligence.

Activity-specific. For example, shopping agents.

Domain-specific. For example, manufacturing, medicine, or telecommunications.

Implementation-specific or Infrastructure-specific. For example, Java, Mozart, KQML, XML.

Please contact us with any questions or comments.

Hit Counter

Updated: September 03, 2001 12:33:15 PM -0600