1. Software Agent -communication-

2. Outline Overview Speech act theory Agent communication languages Summary 1/35

3. Example: Personal Assistant Agent (PA) Task: organize an evening with Heikki What is needed? –Participants’ calendar info –Info about restaurants, pubs, etc. –Weather forecast  Multi-agent System 2/35

4. Example: Organizing Evening (1) Calendar PAA DA Who takes care of Heikki’s Calendar? It is HCA and address is http:…; MTP is…; Presentation is …; CL is…; Ontology is …; … HCA Give me Heikki’s free evenings in this week? Heikki is free on Tuesday and Friday evening 3/35

5. Example: Organizing Evening (2) Calendar PAA HCA I want to book Friday evening Tuesday evening booked No, I don’t want to book Friday evening I want to book Tuesday evening 4/35

6. Example: Sending a Message to an Agent Allocate time from Heikki’s calendar Find out Heikki’s calendar agent Who takes care of Heikki’s Calendar? To: DF@HAP Query Heikki Calendar ServiceDF@HAP 5/35

7. Layered Model 6/35

8. Agent Communication Language Definition –An agent communication language (ACL) allows agents to communicate with each other about their positions and actions –An agents communication language (ACL) provides agents with a means of exchanging information and knowledge Earlier attempts at seamless exchange of information and knowledge between applications –e.g. Remote Procedure Calls, Remote Method Invocation, CORBA Differences between the above and ACLs are –Semantic complexity –ACLs can handle propositions, rules and actions instead of simple objects with no semantics associated with them –An ACL message describes a desired state in a declarative language, rather than a procedure or a method 7/35

9. Agent Communication Agents have conversations (as opposed to exchange single messages), e.g. –Task-oriented –Negotiations A higher level conceptualization of an agent’s strategies drive an agent’s communicative behavior Message type of agents : Speech acts 8/35

10. Speech Acts (1) Communication in MAS is inspired by speech act theory Speech act theories are pragmatic theories of language –They attempt to account for how language is used by people every day to achieve their goals and intentions –The origin of speech act theories are usually traced to the work of the philosopher John Austin Austin noticed that some utterances are like ”physical actions” that appear to change the state of the world. e.g. –Declaring war –”I now pronounce you man and wife” In general, every thing we utter is uttered with the intention of satisfying some goal or intention –A theory of how utterances are used to achieve intentions is a speech act theory 9/35

11. Speech Acts (2) Austin distinguished 3 different aspects of speech acts: –Locutionary act - act of making an utterance e.g. saying ”please make some tea” ’ –Illocutionary act – action performed in saying something e.g. he requested me to make some tea –Perlocution – effect of the act e.g. he got me to make tea Searle identified 5 different types of speech acts: Speech Act TypeDescription and Example Representatives or Assertives Informing e.g. ”It is raining”. Directives Speaker tries to make the hearer do something e.g. ”Please close the window” Commisives Commits the speaker to future action e.g. ”I will close the window” Expressives Speaker expresses a mental state e.g. ”Excuse me”, ”congratulations” Declaratives Effect some changes e.g. ”I name this city Trondheim” 10/35

12. Speech Acts (3) In general, speech acts can be seen to have 2 components: –A performative verb e.g. Request, inform –Propositional content e.g. ”the window is closed” Speech ActPlease close the doorThe door is closedIs the door closed? Performativerequestinforminquire Content the door is closed 11/35

13. Speech Acts (4) How does one define the semantics of speech acts? When can one say someone has uttered, e.g. a request or an inform? How can the properties of speech acts be represented such that planning systems could reason about them? Speech acts are treated as physical actions –Actions are characterised via preconditions and postconditions –Semantics for request: request(s,h,  ) Pre: –s believes h can do  (you don’t ask someone to do something unless you think that they can do it) –s believes h believes h can do  (you don’t ask someone unless they believe they can do it) –s believes s wants  (you don’t ask someone unless you want it!) Post: –h believes s believes s wants  (the effect is to make them aware of your desire) 12/35

14. Agent Communication Language (ACL) ACLs allow agents to effectively communicate and exchange knowledge with other agents Three important aspects –Syntax: How the symbols of communi cation are structured –Semantics: What the symbols denote –Pragmatics: How the symbols are inte rpreted AgentiAgentj Convert to transport form Message delivery/transportation service Goal G Intent I Perform- ative Message (Meaning is a combination of semantics and pragmatics) 13/35

15. Communication Levels Requirements for an ACL –Syntactic translation between languages –Semantic content preservation among applications The concept must have a uniform meaning across applications. –Ability to communicate complex attitudes about their information and knowledge. Agents need to question, request, etc. Not about transporting bits and bytes. Syntax Format of information being transferred Semantics Meaning of the information Communication Method of interconnection 14/35

16. Origins of ACLs Knowledge Sharing Effort (KSE), funded by ARPA –Central concept: knowledge sharing requires communication, which in turn requires a common language. KSE focused on defining that common language KQML: Knowledge Query and Manipulation Language –Language for both message formatting and message handling protocols KIF: Knowledge Interchange Format –Language for expressing message content 15/35

17. KIF Creation of a common language for expressing properties –Intended to express contents of a message; not the message itself –Based on first-order logic Using KIF, it’s possible to express: –Properties of things in a domain (e.g. Michael is a vegetarian) –Relationships between things in a domain (e.g. Michael and Janine are married) –General properties of a domain (e.g. Everybody has a mother) Example –Relation between 2 objects: The temperature of m1 is 83 Celsius: –(= (temperature m1) (scalar 83 Celsius)) –Definition of new concept: An object is a bachelor if this object is a man and not married: –(defrelation bachelor (?x) := (and (man ?x) (not (married ?x)))) –Relationship between individuals in the domain: A person with the property of being a person also has the property of being a mammal: –(defrelation (person ?x) :=> (mammal ?X)) 16/35

18. KQML: Overview An “outer language” that defines a set of performatives (communicative acts), such as ask, reply Performatives form the core of the language: –Determine the kinds of interactions one can have with KQML-speaking agents –Identify the protocol to be used to deliver the message –Signify that the content is an assertion, a query, a command or another speech act –Describe how the sender would like any reply to be delivered Communication Mechanics of communication, e.g. Sender, receiver. Performatives (message layer) Content of communication, e.g. a KIF expression Content Logic of communication, e.g. ask, tell. e.g. (ask-if:sender agenti :receiveragentj :language Prolog :ontologygenealogy :content“spouse(adam, eve)”) 17/35

19. KQML: Categories of Performatives CategoryPerformatives Basic Queryevaluate, ask-if, ask-one, ask-all, ask-about Multi-response Querystream-about, stream-all Responsereply, sorry Generic informationaltell, achieve, cancel, untell Generatorstandby, ready, next,rest, discard, generator Capability-definitionadvertise, subscribe, monitor, import, export Networkingregister, unregister, forward, broadcast, route 18/35

20. KQML: Example (evaluate :sender A : receiver B :language KIF: ontology motors :reply-with q1 :content (val (torque m1))) (reply :sender B : receiver A :language KIF: ontology motors :in-reply-to q1 :content (= (torque m1) (scalar 12 kgf))) (stream-about :sender A : receiver B :language KIF: ontology motors :reply-with q1 :content (m1) (tell :sender B : receiver A :in-reply-to q1 :content (= (torque m1) (scalar 12 kgf))) (tell :sender B : receiver A :in-reply-to q1 :content (= (status m1) (normal))) (eos :sender B : receiver A :in-reply-to q1) 19/35

21. Facilitators KQML environments (may) contain facilitators that help make the communication protocol transparent Facilitators: a special class of agents that perform useful communication services such as: –Maintain registry of service names –Forward messages to named services –Routing messages based on content –Provide matchmaking between information providers and seekers –Provide mediation and translation services 20/35

22. Facilitators: Example (1) AB F ask(X) tell(X) Point-to-point protocol A is aware that it is appropriate to send a query about X to B There are several ways to achie ve this via a Facilitator. 21/35

23. Facilitators: Example (2) AB F tell(X) subscribe(ask(X)) tell(X) Using the subscribe performative Request that Facilitator F monitor for the truth of X. If B subsequently informs F that it believes X to be true, then F can in turn inform A 22/35

24. Facilitators: Example (3) AB F tell(X ) ask(X) broker(ask(X)) advertise(ask(X)) tell(X ) Using the broker performative A Asks Facilitator to find another agent which can process a given performative 23/35

25. Facilitators: Example (4) AB F ask(X) recruit(tell(X)) advertise(ask(X)) tell(X ) Using the rercruit performative Asks Facilitator to find an appropriate agent to which an embedded performative can be forwarded. A reply is returned directly to the original agent 24/35

26. Facilitators: Example (5) AB F reply(B) ask(X) tell(X) recommend(ask(X)) advertise(ask(X)) Using the recommend performative Asks Facilitator to respond with the ”name” of another agent which is appropriate for sending a particular performative. 25/35

27. Discussion on KQML Weak semantics of performatives –Different implementations of KQML could not interoperate Transportation mechanisms were not defined Lacked the class of performatives: commissives –Difficult to implement multi-agent scenarios without commissives Set of performatives was large and ad hoc –Recently, more efforts have been made to provide formal semantics in terms of preconditions, postconditions and completion conditions (ref: Labrou et. al., 1999.) 26/35

28. FIPA The Foundation for Intelligent Physical Agents (FIPA) is a non-profit association FIPA’s purpose is to promote the success of emerging agent-based applications, services and equipment FIPA operates through the open international collaboration of member organizations: companies, universities and government organizations 27/35

29. FIPA ACL (1) Basic structure is quite similar to KQML: –Performative (communicative act) 20 performatives in FIPA ACL –Housekeeping e.g. Sender –Content the actual content of the message Message structure (Envelope): –Comprises of a collection of parameters –Contains at least the mandatory to and sender parameters Example FIPA ACL message Envelope (transport information) Message body (ACL message) (inform :sender agentA :receiver agentB :content (price good200 150) :language sl :ontologyhpl-auction ) 28/35

30. FIPA ACL (2) Inform and request –Inform and Request are the two basic performatives in FIPA ACL. All others are macro definitions, defined in terms of these –The meaning of inform and request are defined in 2 parts: Precondition: What must be true in order for the speech act to succeed Rational effect: What the sender of the message hopes to bring about InformRequest Content statementaction Precondition Holds that the content is true. Intends that the recipient believe the content Does not already believe that the recipient is aware whether content is true or not Intends action content to be performed Believes recipient is capable of performing this action Does not believe that sender already intends to perform action 29/35

31. FIPA Message Transport Model Agent Platform Agent Communication Channel Agent A Agent Platform Agent Communication Channel Agent B Message Transport Protocol ACL message sent over the Message Transport Service 30/35

32. FIPA Sending Messages 3 options –Via local ACC –Via remote ACC –Direct communication mechanism Agent Platform Agent Communication Channel Agent A Agent Platform Agent Communication Channel Agent B 1 1&2 1 2 3 31/35

33. FIPA Agent Interaction Protocol (1) Ongoing conversations between agents fall into typical patterns. In such cases, certain message sequences are expected, and at any point in the conversation, other messages are expected to follow. –These typical patterns of message exchange are called protocols (request:sender A :receiver B :content some-act :protocol fipa-contract-net ) reject-proposal reason Not-understoodrefuse reason propose preconditions2 cfp action preconditions1 accept-proposal proposal failure reason inform Done(action) cancel reason Deadline for proposals Manager cancels the contract due to a change manager Potential contractors 32/35

34. FIPA Agent Interaction Protocol (2) Potential contractors Manager cfp refuse not-understood propose reject-proposal accept-proposal cancel inform deadline 33/35

35. Comparison between KQML & FIPA ACL Similarities: –Separation of the outer language (performative) and the inner language (content). – Allows for any content language Differences –Communication primitives: KQML – performative FIPA ACL – communicative act –Different semantic frameworks – impossible to come up with an exact mapping or transformation between KQML and FIPA performatives –KQML provides facilitator services; FIPA ACL does not 34/35

36. Summary Agent communication –Speech act theory –Agent communication languages Open issues –How do we verify whether an agent complies with the semantic definition? E.g. If an agent sends a message ( say inform Ф), how do we know whether the agent believe Ф or not –Is sincerity a tall-order for agent society? E.g negotiation –Can we specify semantics without a context? E.g. Query If in an oral examination setting We need a context (protocol) Next lecture: Application & review 35/35