Agent-Based Systems Agent-Based Systems Where are we? Last time . . . • Reactive and hybrid agent architectures Agent-Based Systems • Criticism of symbolic AI/deliberative architectures • Situated/embodied/behaviour-based intelligence, emergence Michael Rovatsos • Subsumption architecture mrovatso@inf.ed.ac.uk • Hybrid approaches: the best of both worlds? • Horizontal layering: Touring Machines • Vertical layering: InteRRaP Lecture 6 – Agent Communication Today . . . • Agent Communication 1 / 25 2 / 25 Agent-Based Systems Agent-Based Systems Overview of the course Agent interaction and communication • Intelligent autonomous agents • So far, we have dealt exclusively with single agents • Abstract agent architectures • Today’s lecture marks the beginning of the second block of the • Deductive reasoning agents • Practical reasoning agents course syllabus: foundations of multiagent systems • Reactive and hybrid agent architectures • We will be talking about agents interacting in a common • Communication and cooperation environment • Agent communication • Focus will be on different forms of interaction • Methods for coordination environment • Multiagent decision making • Multiagent interactions • Social choice • Coalition formation • Resource allocation communication • Bargaining • Argumentation in multiagent systems • Logics for multiagent systems 3 / 25 4 / 25
Agent-Based Systems Agent-Based Systems Categories of agent interaction Categories of agent interaction • Remember first lecture • Interaction does not always imply action • Non-/Quasi-communicative interaction: • Coordination does not always imply communication • Shared environment (interaction via resource/capability sharing) • Basic typology of interaction: • ”Pheromone” communication (ant algorithms) interaction • Communication: coordination • Information exchange: sharing knowledge, exchanging views • Collaboration, distributed planning: optimising use of resources and competition cooperation distribution of tasks, coordinating execution collaboration • Negotiation: reaching agreement in the presence of conflict • (Human-machine dialogue, reporting errors, etc.) communication 5 / 25 6 / 25 Agent-Based Systems Agent-Based Systems Speech act theory Speech act theory • A speech act can be conceptualised to consist of: • Most multiagent approaches to communication based on speech 1 Locution (physical utterance) act theory (started by Austin (1962)) 2 Illocution (intended meaning) • Underlying idea: treat communication in a similar way as 3 Perlocution (resulting action) non-communicative action • Two parts of a speech act: • Pragmatic theory of language, concerned with how • Performative = communicative verb used to distinguish between different “illocutionary forces” communication is used in the context of agent activity • Examples: promise, request, purport, insist, demand, etc. • Austin (1962): Utterances are produced like “physical” actions to • Propositional content = what the speech act is about change the state of the world • Example: • Speech act theory is a theory of how utterances are used to • Performative: request/inform/enquire achieve one’s intentions • Propositional content: “the window is open” 7 / 25 8 / 25
Agent-Based Systems Agent-Based Systems Speech act theory Speech act theory • Austin and Searle also analysed the conditions under which • Searle (1972) identified following categories of performatives: speech acts can be successfully completed • assertives/representatives (informing, making a claim) • Austin’s felicity conditions : • directives (requesting, commanding) • commissives (promising, refusing) 1. There must be an accepted conventional procedure for the • declaratives (effecting change to state of the world) performative • expressives (expressing mental states) 2. The procedure must be executed correctly and completely • Ambiguity problems: 3. The act must be sincere, any uptake must be completed as far as possible • “Please open the window!” • Searle’s properties for success of (e.g.) a request: • “The window is open.” • “I will open the window.” 1. I/O conditions (ability to hear request, normal situation) • . . . 2. Preparatory conditions must hold (requested action can be performed, speaker must believe this, hearer will not perform action • Debate as to whether this (or any!) typology is appropriate (and anyway) innate to human thinking) 3. Sincerity conditions (wanting the action to be performed) 9 / 25 10 / 25 Agent-Based Systems Agent-Based Systems Speech acts as rational action Speech acts as rational action • Example of the Cohen-Perrault model: • If communication is like action, what should agents say? • Cohen and Perrault (1979) proposed applying planning techniques Request ( S , H , α ) to speech acts (STRIPS-style) pre − can : ( S BEL ( H CAN α )) ∧ ( S BEL ( H BEL ( H CAN α ))) • Pre- and post-conditions would describe beliefs , abilities and wants pre − want : ( S BEL ( S WANT requestInstance )) of participants effect : ( H BEL ( S BEL ( S WANT α ))) • Distinction between “can-do” and “want” preconditions CauseToWant ( A 1 , A 2 , α ) pre − can : ( A 1 BEL ( A 2 BEL ( A 2 WANT α ))) • Identified necessity of mediating acts , since speech acts say effect : ( A 1 BEL ( A 1 WANT α )) nothing about perlocutionary effect • Cohen and Levesque later integrated that in their model of • This has been the most influential approach to using intentions (as previously discussed) communication in multiagent systems! 11 / 25 12 / 25
Agent-Based Systems Agent-Based Systems Agent communication languages KQML/KIF • KQML – Knowledge Query and Manipulation Language • Agent communication languages (ACLs) define standards for • An “outer” language, defines various acceptable performatives messages exchanged among agents • Example performatives: • Usually based on speech act theory, messages are specified by: • ask-if (‘is it true that...’) • Sender/receiver(s) of the message • perform (‘please perform the following action...’) • Performative to describe intended actions • tell (‘it is true that...’) • Propositional content in some content language • reply (‘the answer is ...’) • Most commonly used languages: • Message format: • KQML/KIF (performative • FIPA-ACL (today de-facto standard) :sender <word> :receiver <word> :in-reply-to <word> :reply-with <word> • FIPA=Foundation for Intelligent Physical Agents :language <word> :ontology <word> :content <expression>) 13 / 25 14 / 25 Agent-Based Systems Agent-Based Systems Example KQML/KIF • KQML does not say anything about content of messages (advertise :sender Agent1 → need content languages :receiver Agent2 • KIF – Knowledge Interchange Format: a logical language to :in-reply-to ID1 describe knowledge :reply-with ID2 • Essentially first-order logic with some extensions/restrictions :language KQML :ontology kqml-ontology • Examples: :content (ask • (= > (and (real-num ?x) (even-num ?n)) :sender Agent1 ( > (expt ?x ?n) > 0)) :receiver Agent3 • (interested joe ’(salary ,?x ,?y ,?z)) :language Prolog • Can be also used to describe ontology referred to by interacting :ontology blocks-world agents :content "on(X,Y)")) 15 / 25 16 / 25
Recommend
More recommend
