Artificial Intelligence Agents and Environment PDF

Artificial Intelligence Agents and Environment By: Dr. Hend Shaaban  2  Recap  What are Agents ? AGENDA  What are environment?  Real time examples of AI.  Summary UNI...

Artificial Intelligence Agents and Environment By: Dr. Hend Shaaban  2  Recap  What are Agents ? AGENDA  What are environment?  Real time examples of AI.  Summary UNIT 1 TOPIC 1 P ANANTHI , AP/CSE​  3 WHAT IS AI ? UNIT 1 TOPIC 1 P ANANTHI , AP/CSE​ HUMAN VS COMPUTER 01 02 03 Knowledge : Intelligence: Common sense: Knowing any Application of Application of information knowledge knowledge in correct time  HUMAN Vs COMPUTER Computer – Human – Natural Artificial Intelligence Intelligence  Artificial defines "man- made" Artificial Intelligence Intelligen defines "thinking power" ce hence AI means "a man- made thinking power." TYPES OF ARTIFICIAL INTELLIGENCE BASED ON ABILITY NARROW AI ⮚ Perform a dedicated task with intelligence ⮚ Cannot perform beyond its field or limitations  Perform any intellectual task like a human GENERAL AI System which could be smarter and think like a human by its own.  ⮚ Perform any task better than human with cognitive properties. ⮚ Hypothetical concept of Artificial Intelligence SUPER AI  Thinking Thinking Humanly rationally Systems that Systems that think like think rationally humans AI 4 Main Acting Acting Approaches Humanly rationally Systems that act Systems that like humans act rationally  Thinking ⮚ Cognitive science brings together AI humanly: computer models ⮚ Psychology of how human mind The cognitive works. modelling ⮚ Example: General Problem Solver approach Thinking rationally: The laws of thought approach Right thinking -Logic `"Socrates is a man; all men are mortal; therefore Socrates is mortal''  Turing Test Acting humanly Intelligent program that act human to fool Turing test Computer capabilities ⮚ Natural language processing ⮚ Knowledge representation ⮚ Automated reasoning ⮚ Machine learning ⮚ Computer Vision ⮚ Robotics Acting rationally: The rational agent approach Acting rationally means acting so as In this approach, to achieve one's AI is viewed as the goals, given one's study and beliefs. construction of rational agents.  AI system – agent + environment. AI- AGENTS & Agents act in their ENVIRONME environment. NTS Environment may contain other agents.  Acting upon that Percepts the environment environment through SENSORS through ACTUATORS AGENT AGENT  Percept : Input at any given instance of time Percept Sequence : Complete Agent history of agent perceived Terminolog Agent Function : Mapping of agent sequence to an action. Internally - By agent program y Externally - Trying all possible actions and recording its action in response.  Must sense Intelligent Must act Agent Must be rational, and autonomous  Three forms of intelligent agent  Human agent  Robotic agent  Software agent  Three forms of intelligent agent ​AGENT/ELEMENT Sensors​ Effectors/actuators​ Eyes, ears, and other Human agent​ Other body parts​ organs​ Robotic agent​ Cameras and infrared wheels, lights, speakers, ​ range finders ​ Various motors​ ​Software agent ​encoded bit strings ​Programs   RATIONAL AGENT?  An agent doing right thing at any instant. GOOD  WHAT IS THE RIGHT THING? BEHAVIOUR :  Will depends on the consequence of the action of the THE CONCEPT agent's behaviour. OF  PERFOMANCE MEASURE ! RATIONALITY  Evaluates the given sequence of environment state  Agent = Architecture + Agent Program Structure Architecture = the machinery that an agent executes on. Physical sensors + Actuators of agents Agent Program = an implementation of an agent function.  Simple Reflex Agent Model-based reflex agent Types of AI Goal-based agents Agents Utility-based agent Learning agent   Choose actions only based on the current percept Simple  Condition-Action Rule Reflex  Eg: Autonomous air conditioner Agent  Ex: if car-in-front-is-braking then Initiate- braking Simple Reflex Agent   It keeps track of the part of the environment Model  It maintains an internal state that depends on the percept history and thereby reflects on Based observed aspect of the current state. Reflex  Updating the current state, 1. Information of how world evolves independently Agent 2. Information of how agent's action affect the world  Eg: vacuum cleaner Model-based reflex agent  Goal based agent Goal information desirable situation. Choose among multiple possibilities-> goal state. Eg : Self driving car Goal-based agents USES UTILITY FUNCTION PERFORMANCE MEASURE EG: CONTENT FIND EXACTLY HOW HAPPY RECOMMENDATION SYSTEM THEY WOULD MAKE THE AGENT Utility based agents Utility based agents  Learning element: It is responsible for making improvements by learning from the environment. Critic: The learning element takes feedback from critics which describes how well the agent is doing with respect to a fixed performance standard. LEARNING Performance element: It is responsible for selecting AGENT external action. Problem Generator: This component is responsible for suggesting actions that will lead to new and informative experiences. Eg: Spam filter LEARNING AGENT  Highest performing agents Example: RATIONAL Chooses the shortest path with AGENT low cost for high efficiency  Nature of  In task environments are essentially environme the problem to which rational agent are the solution. nt  P E A S Performance Environment Actuators Sensors Task environment specification  PERFORMANCE AGENT TYPE MEASURE ENVIRONMENT ACTUATORS SENSORS Reaching correct destination, minimizing Camera, fuel consumption & Road, traffic, Wheels, Steering, Speedometer, Taxi driver wear & tear, trip time pedestrians and Horn, brake Accelerometer and violation of traffic customers Etc., rule, maximizing safety and passenger Example  Properties of task environment Fully observable Partially observable 1. Agent's sensors gets access to 1. Agent's sensors does not complete state of environment at get access to complete state of each point of time environment at each point of time 2. Sensors detects all aspects choice of actions 2. Sensors detects noisy and incomplete observations 3. No need to maintain internal state 3. May maintain internal state 4. Eg: Chess or tic-tac-toe 4. Eg: Autonomous vehicle  Properties of task environment Deterministic Stochastic / Probabilistic 1. Next state is completely 1. It is influenced by randomness determined by the current state and uncertainty and action executed by the agent 2. The same initial state and action, 2. Eg: Board games : Chess or tic- outcome may differ due to tac-toe probabilistic factors 3. Eg: Dice rolls, weather prediction  Properties of task environment Episodic Sequential 1. Agent's experience is divided 1. The current decision will affect all into atomic steps i.e., the agent the future decision. receives a percept and then performs an action 2. Eg: Autonomous vehicle 2. Next episode is not affected by previous episode. 3. Eg: Maze problem  Properties of task environment Static Semi Dynamic Dynamic 1. State of the 1. Certain aspects or 1. State of the environment elements of the environment can remains constant environment can change over time throughout the change over time, even without the agent's interaction but the changes do direct actions of the not occur agent 2. Eg: Puzzle independently or autonomously. 2. Eg: Video player game 2. Eg: Recommendation system  Properties of task environment Discrete Continuous 1. The state and action spaces are 1. The state and action spaces are finite and well-defined. infinite or uncountably large. 2. Environment consists of a limited 2. The transitions between states number of distinct states, and and the effects of actions are the agent's actions also have a typically governed by finite set of options. mathematical functions and can involve a wide range of possible 3. Transitions between states and values. the effects of actions are typically deterministic and can 3. Eg: Robot control task be represented using discrete values or categories. 4. Eg: Chess  Properties of task environment Known Unknown 1. The agent has a complete and 1. The agent has limited or no prior accurate information about the information about the environment. environment's dynamics, rules, or possible states and actions. 2. Eg: Board games 2. The agent needs to explore and learn about the environment through interactions and observations. 3. Eg: Real world robots  Properties of task environment Competitive Collaborative 1. An agent is said to be in a 1. An agent is said to be in a competitive environment when it collaborative environment when competes against another agent multiple agents cooperate to to optimize the output. produce the desired output. 2. Eg: Chess game 2. Eg: Autonomous cars  Properties of task environment Competitive Collaborative 1. An environment consisting of 1. An environment involving more only one agent. than one agent. 2. Eg: Maze 2. Eg: Foot-ball game  Agent Single Multi Agents agent agent Competitiv Cooperativ e multi e multi agent agent