Administrative Issues Login into learn.usc.edu and make sure Login - PDF document

Administrative Issues � Login into learn.usc.edu and make sure Login into learn usc edu and make sure that CSCI561a is listed as one of your courses. � Web page: � Web page: � http://www-scf.usc.edu/~ csci561b/ � http://den.usc.edu Acting Humanly: The Full Turing Test • Problem: 1) Turing test is not reproducible, constructive, and amenable to mathematic analysis. 2) What about physical interaction with interrogator and environment? Trap door 1

Last time: The Turing Test http://aimovie.warnerbros.com http://www.ai.mit.edu/projects/infolab/ This time: Outline � Intelligent Agents (IA) Intelligent Agents (IA) � Environment types � IA Behavior � IA Structure � IA Types IA T 2

Attributes of Intelligent Behavior � Think and reason � Use reason to solve problems � Learn or understand from experience � Acquire and apply knowledge � Exhibit creativity and imagination � Deal with complex or perplexing situations � Respond quickly and successfully to new situations. it ti � Recognize the relative importance of elements in a situation � Handle ambiguous, incomplete,or erroneous information Intelligent Agents Interface Tutors Search Agents Presentation User Agents Information Interface Management Agents Agents Information Brokers Network Navigation A Agents t Information Filters Role- Playing Agents 3

What is an (Intelligent) Agent? � An over-used, over-loaded, and An over used over loaded and misused term. � Anything that can be viewed as � perceiving its environment p g � acting upon that environment What is an (Intelligent) Agent? � PAGE (Percepts, Actions, Goals, PAGE (Percepts Actions Goals Environment) � Task-specific & specialized: well-defined g goals and environment 4

Intelligent Agents and Artificial Intelligence � Example: Human mind as network of p thousands or millions of agents working in parallel. Agency sensors effectors Agent Types Agent research fall into two main strands: Agent research fall into two main strands: � Distributed Artificial Intelligence (DAI) – Multi-Agent Systems (MAS) (1980 – 1990) � Much broader notion of "agent" (1990’s – present) 5

Rational Agents Ho How to design this? to design this? Sensors percepts ? Environment A Agent actions Effectors A Windshield Wiper Agent How do we design an agent that can wipe How do we design an agent that can wipe the windshields when needed? � Goals? � Percepts ? � Sensors? S ? � Effectors ? � Actions ? � Environment ? 6

Grand Challenge � Autonomous Driving Autonomous Driving Interacting Agents Collision Avoidance Agent (CAA) � Goals: Avoid running into obstacles � Percepts ? � Sensors? � Effectors ? � Actions ? � Environment: Freeway 7

Interacting Agents Lane Keeping Agent (LKA) • Goals: Stay in current lane • Percepts ? • Sensors? • Effectors ? Effectors ? • Actions ? • Environment: Freeway Conflict Resolution by Action Selection Agents • Override: Override: • Arbitrate: • Compromise: • Challenges: 8

The Right Thing = The Rational Action � Rational Action: The action that maximizes the � Rational Action: The action that maximizes the expected value of the performance measure given the percept sequence to date � Rational = Best ? � Rational = Optimal ? � Rational = Omniscience ? (Having total R ti l O i i ? (H i t t l knowledge) � Rational = Clairvoyant ? (The sixth sense) � Rational = Successful ? Behavior and performance of IAs � Perception (sequence) to Action Perception (sequence) to Action Mapping: f : P * → A � I deal mapping: 9

Look up table obstacle Distance Distance Acttion Acttion sensor sensor agent 10 No action 5 5 Turn left 30 Turn left 30 degrees 2 Stop Closed form � Output (degree of rotation) = Output (degree of rotation) = F(distance) 10

Behavior and performance of IAs � Performance measure: Performance measure: � (degree of) Autonomy: f) A t (d How is an Agent different from other software? � Agents are autonomous � Agents are autonomous , � Agents contain some level of intelligence , � Agents don't only act reactively , but sometimes also proactively 11

How is an Agent different from other software? � Agents have social ability , � Agents may cooperate with other agents � Agents may migrate from one system to another Environment Types � Characteristics � Accessible vs. inaccessible � Deterministic vs. nondeterministic � Episodic vs. nonepisodic (Sequential) 12

Environment Types � Characteristics � Hostile vs. friendly � Static vs. dynamic � Discrete vs. continuous Environment types Environment Environment Accessi Accessi Determinis Determinis Episodic Static Episodic Static Discrete Discrete ble tic Operating System Virtual Reality Offi Office Environment Mars 13

Structure of Intelligent Agents � Agent = architecture + program Agent = architecture + program � Agent program: the implementation of f : P * → A, the agent’s perception- action mapping function : Skeleton-Agent( Percept ) returns Action memory ← UpdateMemory(memory, Percept ) Action ← ChooseBestAction(memory) memory ← UpdateMemory(memory, Action ) return Action Using a look-up-table to encode f : P * → A � Example: Collision Avoidance p � Sensors: 3 proximity sensors obstacle � Effectors: Steering Wheel, Brakes sensors � How to generate? agent � How large? � How to select action? 14

Using a look-up-table to encode f : P * → A obstacle b t l sensors � Example: Collision Avoidance agent � Sensors: 3 proximity sensors � Effectors: Steering Wheel, Brakes Using a look-up-table to encode f : P * → A � How large: � How large: � How to select action? � Is it an autonomous agent? (by using the look up table) 15

Agent types � Reflex agents g � Reactive: No memory � Reflex agents with internal states � Goal-based agents � Goal information needed to make decision Agent types � Utility-based agents y g � Learning Agent 16

Reflex agents Question � Design a group of mobile robots that Design a group of mobile robots that stay together and move around using reactive (reflex) agents? 17

Reflex agents w/ state (model-based reflex agent) Goal-based agents 18

Utility-based agents Learning agents Performance standard Critic feedback Changes Learning Performance element element Learning Learning Knowledge l d goal Problem generator 19

Information agents � Manage the explosive growth of information. Information agents � Examples: Examples: � BargainFinder comparison shops among Internet stores for CDs � FIDO the Shopping Doggie (out of service) � Internet Softbot infers which internet facilities (finger, ftp, gopher) to use and when from high-level search requests. � Challenge: ontologies for annotating Web pages (eg, SHOE). 20

Example: ALADDIN project Autonomous Learning Agents for Decentralized Data and Information Network � http://www.aladdinproject.org/technolo http://www aladdinproject org/technolo gies.html � Online evacuation algorithm � Based on sensors, provide guidance to people. � Rescue planning � Detecting injured and update all (using NN) Summary � I ntelligent Agents: I ntelligent Agents: � Anything that can be viewed as perceiving its environment through sensors and acting upon that environment through its effectors to maximize progress towards its goals . � PAGE (Percepts, Actions, Goals, Environment) PAGE (P t A ti G l E i t) � Described as a Perception (sequence) to Action Mapping: f : P * → A � Using look-up-table, closed form, etc. 21

Summary � Agent Types: Reflex, state-based, goal- Agent Types: Reflex state based goal based, utility-based, Learning � Rational Action: The action that maximizes the expected value of the maximizes the expected value of the performance measure given the percept sequence to date 22