What is an Agent? � The main point about agents is they are autonomous : capable of acting independently, LECTURE 2: exhibiting control over their internal state INTELLIGENT AGENTS � Thus: an agent is a computer system capable of autonomous action in some environment in order to meet its design objectives An Introduction to MultiAgent Systems SYSTEM http://www.csc.liv.ac.uk/~mjw/pubs/imas output input ENVIRONMENT 1 2 What is an Agent? Reactivity � If a program’s environment is guaranteed to be fixed, the � Trivial (non-interesting) agents: program need never worry about its own success or � thermostat failure – program just executes blindly � UNIX daemon (e.g., biff) � Example of fixed environment: compiler � An intelligent agent is a computer system � The real world is not like that: things change, information is incomplete. Many (most?) interesting environments are capable of flexible autonomous action in dynamic some environment � Software is hard to build for dynamic domains: program � By flexible , we mean: must take into account possibility of failure – ask itself � reactive whether it is worth executing! � pro-active � A reactive system is one that maintains an ongoing � social interaction with its environment, and responds to changes that occur in it (in time for the response to be useful) 3 4 Balancing Reactive and Goal-Oriented Proactiveness Behavior � Reacting to an environment is easy (e.g., � We want our agents to be reactive, stimulus → response rules) responding to changing conditions in an � But we generally want agents to do things appropriate (timely) fashion for us � We want our agents to systematically work � Hence goal directed behavior towards long-term goals � Pro-activeness = generating and � These two considerations can be at odds with attempting to achieve goals; not driven one another solely by events; taking the initiative � Designing an agent that can balance the two � Recognizing opportunities remains an open research problem 5 6 1
Social Ability Other Properties � The real world is a multi -agent environment: � Other properties, sometimes discussed in the context of we cannot go around attempting to achieve agency: goals without taking others into account � mobility : the ability of an agent to move around an electronic network � Some goals can only be achieved with the � veracity : an agent will not knowingly communicate false cooperation of others information � Similarly for many computer environments: � benevolence : agents do not have conflicting goals, and that witness the Internet every agent will therefore always try to do what is asked of it � Social ability in agents is the ability to interact � rationality : agent will act in order to achieve its goals, and will not act in such a way as to prevent its goals being achieved with other agents (and possibly humans) via — at least insofar as its beliefs permit some kind of agent-communication language , � learning/adaption : agents improve performance over time and perhaps cooperate with others 7 8 Agents and Objects Agents and Objects � Main differences: � Are agents just objects by another name? � agents are autonomous: agents embody stronger notion of autonomy than � Object: objects, and in particular, they decide for themselves � encapsulates some state whether or not to perform an action on request from another agent � communicates via message passing � agents are smart: � has methods, corresponding to operations capable of flexible (reactive, pro-active, social) behavior, that may be performed on this state and the standard object model has nothing to say about such types of behavior � agents are active: a multi-agent system is inherently multi-threaded, in that each agent is assumed to have at least one thread of active control 9 10 Objects do it for free… Agents and Expert Systems � Aren’t agents just expert systems by another name? � agents do it because they want to � Expert systems typically disembodied ‘expertise’ � agents do it for money about some (abstract) domain of discourse (e.g., blood diseases) � Example: MYCIN knows about blood diseases in humans � It has a wealth of knowledge about blood diseases, in the form of rules � A doctor can obtain expert advice about blood diseases by giving MYCIN facts, answering questions, and posing queries 11 12 2
Agents and Expert Systems Intelligent Agents and AI � Aren’t agents just the AI project? � Main differences: Isn’t building an agent what AI is all about? � agents situated in an environment: MYCIN is not aware of the world — only � AI aims to build systems that can information obtained is by asking the user (ultimately) understand natural language, questions recognize and understand scenes, use � agents act: common sense, think creatively, etc. — all MYCIN does not operate on patients of which are very hard � Some real-time (typically process control) � So, don’t we need to solve all of AI to build expert systems are agents an agent…? 13 14 Intelligent Agents and AI Environments – Accessible vs. inaccessible � When building an agent, we simply want a � An accessible environment is one in which system that can choose the right action to the agent can obtain complete, accurate, perform, typically in a limited domain up-to-date information about the � We do not have to solve all the problems of environment’s state AI to build a useful agent: � Most moderately complex environments a little intelligence goes a long way! (including, for example, the everyday physical world and the Internet) are � Oren Etzioni, speaking about the commercial inaccessible experience of NETBOT, Inc: “We made our agents dumber and dumber � The more accessible an environment is, the simpler it is to build agents to operate in it and dumber…until finally they made money.” 15 16 Environments – Environments - Episodic vs. non-episodic Deterministic vs. non-deterministic � In an episodic environment, the performance of an agent is dependent on a number of � A deterministic environment is one in which discrete episodes, with no link between the any action has a single guaranteed effect — performance of an agent in different scenarios there is no uncertainty about the state that � Episodic environments are simpler from the will result from performing an action agent developer’s perspective because the � The physical world can to all intents and agent can decide what action to perform purposes be regarded as non-deterministic based only on the current episode — it need � Non-deterministic environments present not reason about the interactions between greater problems for the agent designer this and future episodes 17 18 3
Recommend
More recommend
