Software agent

from Wikipedia, the free encyclopedia

A software agent (also agent or softbot ) is a computer program that is capable of certain ( well-specified ) independent and dynamic (autonomous) behavior. This means that, depending on various states (status), a certain processing procedure takes place without a further start signal being given from outside or an external control intervention taking place during the procedure.

Definition and basics

According to Michael Wooldridge, there is no generally accepted definition of an agent. There is general agreement that an agent must be independent, but beyond that there is little agreement. Wooldridge tries nevertheless to a definition: "An agent is a computer system which is located in a certain environment and which is able to carry out independent actions in this environment in order to achieve its (given) goals." The VDI guideline: VDI / VDE 2653 gives the following definition: “A technical agent is a delimitable (hardware or / and software) unit with defined goals. A technical agent strives to achieve these goals through independent behavior and interacts with his environment and other agents. "

Research on artificial intelligence defines software as an agent if it has the following properties that describe the degree of program autonomy:

autonomous

works independently of user intervention

cognitive

is capable of learning and learns based on previously made decisions or observations

communicative

communicates its states as an effect on its environment

modal adaptive

changes its own settings (parameters and / or structure) based on its own conditions and the conditions of the environment

active

carries out actions on its own initiative

reactive

reacts to changes in the environment

robust

compensates for external and internal disturbances

social

communicates with other agents

Simultaneous external interventions that limit autonomy or change decision-making criteria cannot be ruled out.

If, in addition to the properties mentioned, the ability to automatically change the place of execution (to migrate ), one speaks of a mobile agent . To do this, he needs skills that enable him to adapt to other infrastructure to a certain extent. See migration (information technology) , which describes such mechanisms on the part of humans.
Intelligent agents are characterized by knowledge , the ability to learn, inferences and the ability to change behavior.

A network of a subset of autonomous agents that can communicate with one another is called a population . This communication is influenced by the density and distribution of the agents as well as their grouping and the temporal variance of these parameters.

commitment

Agents are given a wide range of uses in the areas of e-commerce , information research, simulation , performing routine tasks and in autonomous systems. But also complex tasks, for example in automated negotiations, can be taken over by software agents. In the field of simulation, there is the specialty of multi-agent simulation or group simulation with its own software products. The latter is often used in games.

Implementations

There are numerous implementations of agent platforms in the scientific environment. These usually have a special focus, for example intelligent behavior, security, efficient migration.

A detailed overview of current systems, which is funded as a project ( Co-ordination Action ) within the framework of the sixth research framework program of the European Commission, is AgentLink.org.

Among other things, there is the extensive, Java- based agent framework JADE .

Agent types

Agent types differ in the agent architecture (not to be confused with the architecture on which the environment program runs). An agent architecture is the way in which the definition and management of agent behavior takes place. In principle, there is a wide variety of terms, but a division into three largely recognized areas is possible:

Reactive agents

In principle, reactive (or subcognitive ) agents do not have their own knowledge, but only act directly on the basis of their perceptions and without a decision-making process.

The following agent types occur more often in this context:

Simply reactive agent

Is the simplest type. The agent receives sensor information and selects an action based on condition-action rules .

Observing agent

Is an extension of the simple reactive agent. This type of agent already has a memory and collects information about the environment and what its own actions would cause. The condition-action rules are then applied to this overall picture and no longer only to the pure sensor information.

Adaptive agents

Adaptive agents manage a model of their own process and parameter structure. These can be adapted to your own history and recognized or measured external conditions. This enables adaptive regulation and thus, for example, an optimal implementation with regard to resources.

Cognitive agents

Cognitive agents manage a model of their environment in their own data structure. This enables the planning of the actions and ultimately also targeted action. A well-known subclass is the agent definition in the BDI agents by specifying the beliefs , desires and intentions .

The following agent types occur more often in this context:

Goal-based agent

The agent has a target that he is trying to achieve, and based on the sensor information and his knowledge of the consequences of his actions, he decides which action brings him closest to his goal. Since the goal cannot always be achieved in one step, the agent is able to plan.

Benefit-based agent

As a further development of the goal-based agent, the use-based agent also has a target specification. All possible and impossible states are mapped to a real number, which represents the benefits for the agent. He himself has the value -∞. This enables him to decide in situations in which several goals are achievable which actions have the greater benefit or which goals are more desirable. This is particularly interesting when it cannot be said with certainty whether a goal can be achieved. The agent can use it to carry out a risk assessment and will not only follow its main goal.

See also

literature

  • Rolf Grütter: Software Agents on the Web . In: Informatik Spektrum , 2006, issue 1
  • Michael Wooldridge: Intelligent Agents: The Key Concepts . Springer, Berlin / Heidelberg January 1, 2002, p. 151-190 , doi : 10.1007 / 3-540-45982-0_1 .

Web links

Individual evidence

  1. ^ Wooldridge: Intelligent Agents: The Key Concepts . 2002, p. 5.
  2. ^ Professor Michael Wooldridge, Head of Department of Computer Science, Professor of Computer Science, Senior Research Fellow, Hertford College. Retrieved March 1, 2017 .
  3. VDI guideline 2653 sheet 1: Agent systems in automation technology - basics , 2010
  4. AgentLink.org ( Memento of the original from July 23, 2008 in the Internet Archive ) Info: The archive link was inserted automatically and has not yet been checked. Please check the original and archive link according to the instructions and then remove this notice. @1@ 2Template: Webachiv / IABot / www.agentlink.org
  5. Adaptive Agents (PDF; 1.0 MB)
  6. Definition of adaptive agents ( Memento of the original from May 14, 2011 in the Internet Archive ) Info: The archive link was automatically inserted and not yet checked. Please check the original and archive link according to the instructions and then remove this notice. @1@ 2Template: Webachiv / IABot / www.iscid.org
  7. Adaptive Agents and Multi-Agent Systems (PDF; 216 kB)
  8. Cognitive Agents (PDF; 309 kB)
  9. Cognitive Agents ( Memento of the original from January 31, 2012 in the Internet Archive ) Info: The archive link was inserted automatically and has not yet been checked. Please check the original and archive link according to the instructions and then remove this notice. (PDF; 1.5 MB) @1@ 2Template: Webachiv / IABot / www.pst.ifi.lmu.de
  10. Survey of Cognitive and Agent Architectures ( Memento of the original from February 23, 2010 in the Internet Archive ) Info: The archive link was inserted automatically and has not yet been checked. Please check the original and archive link according to the instructions and then remove this notice. @1@ 2Template: Webachiv / IABot / ai.eecs.umich.edu