Advanced

Autonomous Agents and the Concept of Concepts

Davidsson, Paul (1996)
Abstract (Swedish)
Popular Abstract in Swedish

En autonom agent kan definieras som ett system som med syfte att utföra uppgifter kan interagera med sin omgivning genom sina egna sensorer och effektorer. En ny arkitektur för autonoma agenter inspirerad av teorin för antecipatoriska system föreslås. Det är en hybrid arkitektur i det avseende att den kombinerar planerande förmåga, som behövs för att lösa långsiktiga problem, med reaktivt beteende, som är nödvändigt för att klara enklare rutinmässiga problem vilka ofta kräver kort reaktionstid. Den grundläggande idén är att låta en metanivå-komponent utnyttja en modell av agenten och dess omgivning för att förutse framtida tillstånd. Dessa prediktioner används sedan för att påverka beteendet hos... (More)
Popular Abstract in Swedish

En autonom agent kan definieras som ett system som med syfte att utföra uppgifter kan interagera med sin omgivning genom sina egna sensorer och effektorer. En ny arkitektur för autonoma agenter inspirerad av teorin för antecipatoriska system föreslås. Det är en hybrid arkitektur i det avseende att den kombinerar planerande förmåga, som behövs för att lösa långsiktiga problem, med reaktivt beteende, som är nödvändigt för att klara enklare rutinmässiga problem vilka ofta kräver kort reaktionstid. Den grundläggande idén är att låta en metanivå-komponent utnyttja en modell av agenten och dess omgivning för att förutse framtida tillstånd. Dessa prediktioner används sedan för att påverka beteendet hos en reaktiv komponent. En speciell variant av denna arkitektur som betraktar alla agenter (inklusive sig själv) som varande helt reaktiva har implementerats. Karakteristiskt för denna variant är också att man delar in agentens möjliga tillstånd i önskvärda respektive icke önskvärda tillstånd. När metanivå-komponenten genom sina prediktioner upptäcker att agenten kommer att nå ett icke önskvärt tillstånd, försöker den modifiera den reaktiva komponenten så att den undviker att detta tillstånd uppkommer. Empiriska resultat från simuleringar med både en och flera agenter indikerar att beteendet för denna typ av antecipatoriska agenter är överlägset det för motsvarande reaktiv agenter. Autonoma agenter utgör även det ramverk i vilket representation, bildning och inlärning av konkreta begrepp (d.v.s. begrepp som representerar kategorier av konkreta begrepp, t.ex. "stol") har studerats. Innan man undersöker dessa aspekter bör man emellertid först ta ställning till några mer fundamentala frågor om begrepp som t.ex.: Vad menas med att ha ett begrepp? Vilka funktioner har begrepp? Vad känner vi till om kategorier och hur påverkar det begreppsbildningen? Genom att försöka besvara dessa frågor kan man säga att vi studerar själva begreppet begrepp. Eftersom ovanstående frågor sällan har diskuteras inom området Artificiell Intelligens (AI) är mycket av materialet hämtat från närliggande forskningsområden som kognitiv psykologi och filosofi. Ett av huvudmålen med avhandlingen är just att samla ihop de olika teorier som utvecklats inom de olika kognitiva vetenskaperna för att bilda en stabil grund för framtida AI-forskning. De mest kända teorierna för begreppsrepresentation utvärderas sedan i ljuset av svaren till ovanstående frågor. Slutsatsen blir att inget av dessa representationssätt är tillräckligt kraftfullt för att kunna understödja alla de önskvärda funktionerna. Det är till och med så att det verkar orealistiskt att någon "monolitisk" representation skulle kunna vara tillräcklig. Baserat på denna insikt presenteras en ny teori för sammansatta begreppsrepresentationer där de olika komponenterna motiveras av de funktioner som begrepp bör fylla. När det gäller inlärning av begrepp så identifieras några av de krav som varje autonomt begreppsinlärningssystem måste uppfylla. Dessa krav utgör också grunden för en evaluering av de existerande teorierna för begreppsinlärning. En metod för att få en godtycklig inlärningsalgoritm att uppfylla ett av dessa krav presenteras sedan. Kravet består i att algoritmen måste lära sig karaktäristiska klassificeringsbeskrivningar, d.v.s. algoritmen måste inte bara kunna diskriminera mellan de begrepp den redan känner utan även kontrastera dessa mot de begrepp som den inte känner. Experimentella resultat visar att denna metod även fungerar bra i praktiken, bl.a. för att det till skillnad från tidigare metoder för inlärning av karaktäristiska beskrivningar är möjligt att kontrollera generaliseringsgraden. Slutligen skisseras en ny modell för autonom begreppsinlärning som har potential att uppfylla alla de tidigare diskuterade kraven (bl.a. integreras övervakad och oövervakad inlärning). (Less)
Abstract
This thesis has two main themes, autonomous agents and concepts. An autonomous agent is a system capable of interacting with its environment via its own sensors and effectors in order to accomplish some task. Arguments against both purely reactive and purely deliberative agent architectures are presented in favor for hybrid approaches. A novel hybrid approach based on the concept of anticipatory systems is suggested. The basic idea is to let a meta-level component ``run'' a world model faster than real time to make predictions of future states. These predictions are used to guide the agent's behavior on a high-level, whereas the low-level behavior is controlled by a reactive component. A specialization of this architecture, a linearly... (More)
This thesis has two main themes, autonomous agents and concepts. An autonomous agent is a system capable of interacting with its environment via its own sensors and effectors in order to accomplish some task. Arguments against both purely reactive and purely deliberative agent architectures are presented in favor for hybrid approaches. A novel hybrid approach based on the concept of anticipatory systems is suggested. The basic idea is to let a meta-level component ``run'' a world model faster than real time to make predictions of future states. These predictions are used to guide the agent's behavior on a high-level, whereas the low-level behavior is controlled by a reactive component. A specialization of this architecture, a linearly quasi-anticipatory agent architecture, which treats all agents in the domain (itself included) as being reactive, has been implemented. According to this approach, the state space is divided into desired and undesired regions. When the meta-level component detects that the simulated reactive system has reached an undesired state, it modifies the actual reactive system in order to avoid reaching this state. Results from both single and multi-agent experiments indicate that the behavior of such agents is superior to that of the corresponding reactive agents. Autonomous agents also provide the framework in which the representation and the acquisition of concepts are studied. However, these topics should not be studied without first answering some more fundamental questions regarding concepts, such as: What does it mean to have a concept? What functions do, or should, concepts serve? What is known about the nature of categories? Thus, by trying to answer these questions, we investigate the very concept of concepts. Although these topics seldom are discussed within Artificial Intelligence, they have received some attention in related fields, e.g., cognitive psychology and philosophy. One of the main goals of this thesis is to pull together different lines of argumentation that have emerged from the cognitive sciences in order to establish a solid foundation for further AI research. Previous approaches to concept representation are evaluated in the light of the answers to the questions above. It is concluded that none of the existing approaches is able to serve all the desired functions and that it is unrealistic to expect that any monolithic representation would be adequate. Based on this insight, a novel composite representation scheme is presented in which each component is motivated by the functions a concept should serve. Regarding the acquisition of concepts, some of the requirements that any autonomous concept learning system must meet are identified and provide the basis for an evaluation of the existing theories. A method for making any learning algorithm satisfy one such requirement, namely that of representing concepts by characteristic descriptions, is presented together with some promising experimental results. In contrast to previous methods for learning characteristic descriptions, it is possible with this method to control the degree of generalization. In addition, a new model for integrating learning by being told, learning from examples and learning by observation is outlined. (Less)
Please use this url to cite or link to this publication:
author
opponent
  • Jansson, Carl Gustaf
publishing date
type
Thesis
publication status
published
subject
keywords
Artificial intelligens, Data- och systemvetenskap, computer technology, Systems engineering, Adaptive Behavior, Anticipation, Cognitive Modeling, Concept Representation, Concept Learning, Artificial Intelligence, Autonomous Agents, Artificiell intelligens
pages
221 pages
publisher
Department of Computer Science, Lund University
defense location
E-building, LTH, Lund
defense date
1996-05-20 10:15
external identifiers
  • other:ISRN: LUTEDX/(TECS-1006)/1-221/(1996)
ISBN
91-628-2035-4
language
English
LU publication?
no
id
77a5a077-a7ab-4e07-bb44-3016b2f45678 (old id 17635)
date added to LUP
2007-05-24 09:28:02
date last changed
2016-09-19 08:45:03
@phdthesis{77a5a077-a7ab-4e07-bb44-3016b2f45678,
  abstract     = {This thesis has two main themes, autonomous agents and concepts. An autonomous agent is a system capable of interacting with its environment via its own sensors and effectors in order to accomplish some task. Arguments against both purely reactive and purely deliberative agent architectures are presented in favor for hybrid approaches. A novel hybrid approach based on the concept of anticipatory systems is suggested. The basic idea is to let a meta-level component ``run'' a world model faster than real time to make predictions of future states. These predictions are used to guide the agent's behavior on a high-level, whereas the low-level behavior is controlled by a reactive component. A specialization of this architecture, a linearly quasi-anticipatory agent architecture, which treats all agents in the domain (itself included) as being reactive, has been implemented. According to this approach, the state space is divided into desired and undesired regions. When the meta-level component detects that the simulated reactive system has reached an undesired state, it modifies the actual reactive system in order to avoid reaching this state. Results from both single and multi-agent experiments indicate that the behavior of such agents is superior to that of the corresponding reactive agents. Autonomous agents also provide the framework in which the representation and the acquisition of concepts are studied. However, these topics should not be studied without first answering some more fundamental questions regarding concepts, such as: What does it mean to have a concept? What functions do, or should, concepts serve? What is known about the nature of categories? Thus, by trying to answer these questions, we investigate the very concept of concepts. Although these topics seldom are discussed within Artificial Intelligence, they have received some attention in related fields, e.g., cognitive psychology and philosophy. One of the main goals of this thesis is to pull together different lines of argumentation that have emerged from the cognitive sciences in order to establish a solid foundation for further AI research. Previous approaches to concept representation are evaluated in the light of the answers to the questions above. It is concluded that none of the existing approaches is able to serve all the desired functions and that it is unrealistic to expect that any monolithic representation would be adequate. Based on this insight, a novel composite representation scheme is presented in which each component is motivated by the functions a concept should serve. Regarding the acquisition of concepts, some of the requirements that any autonomous concept learning system must meet are identified and provide the basis for an evaluation of the existing theories. A method for making any learning algorithm satisfy one such requirement, namely that of representing concepts by characteristic descriptions, is presented together with some promising experimental results. In contrast to previous methods for learning characteristic descriptions, it is possible with this method to control the degree of generalization. In addition, a new model for integrating learning by being told, learning from examples and learning by observation is outlined.},
  author       = {Davidsson, Paul},
  isbn         = {91-628-2035-4},
  keyword      = {Artificial intelligens,Data- och systemvetenskap,computer technology,Systems engineering,Adaptive Behavior,Anticipation,Cognitive Modeling,Concept Representation,Concept Learning,Artificial Intelligence,Autonomous Agents,Artificiell intelligens},
  language     = {eng},
  pages        = {221},
  publisher    = {Department of Computer Science, Lund University},
  title        = {Autonomous Agents and the Concept of Concepts},
  year         = {1996},
}