Artificial Intelligence Chapter 9

Programming

"Give a man a fish and you feed him for a day"

Learning

"Teach a man to fish and you feed him for a lifetime"

Machine Learning System

- Data and Output are inputs to a computer and a Program is coming out; - Implemented through an algorithm that automatically generates programs by learning data; - A system that improves itself by self-learning;

Expert System

- Data and Program are inputs to a computer and an Output is coming out; - Using human knowledge and experience in the form of specific rules; - It can not improve itself through experience;

Stages of AI

-Applied AI; -Cognitively simulated AI; -Strong AI

Classical Views

-Thinking Humanly; -Thinking Rationally; -Acting Humanly; -Acting Rationally.

Designing and Implementation of AI Projects

1. Define your tasks for the project objective. 2. Illustrate data flow from raw data collection and storage to processing and discovery. 3. Features Selection. 4. Classify machine learning types for the selected feature data. 5. Choose Algorithms. 6. Experimental Design and Implementation by using available open source software frameworks. 7. Run it!

AI

A Machine that thinks and acts like a person

good heuristic

A ___________ must reasonably find the distance to the target, and the computation to find the heuristic should be simple

Frame

A data structure containing information about the object and facts

Semantic Networks

A graph whose nodes represent concepts and whose arcs represent relations between these concepts

a button

A great awakening - defining _________ as "an object with a few holes in the middle", then many errors occur

Deliberative Agent

A more intelligent agent that adaptively responds to the outside world and interacts with the external environment through a sensor and an effector

Reactive Agent

A simple agent that automatically responds to input data according to defined rules, such as condition reflections

AI Landscape

AI, ML, DL

Consistency, Memory, Clear Logic, Accessibility, Longevity

Advantages of Expert System

Easy to encode and understand

Advantages of Semantic Networks

Machine Learning(ML)

Algorithm that automatically generates programs by learning data

Knowledge processing, Machine Learning, Cognitive ability

Base technologies

Adaptive Learning

By imitating the human brain and adapting to a new environment, it accumulates intelligence on its own

Planning

Deciding what actions to perform (and when) to achieve a given objective

Scheduling

Deciding when to perform a given set of actions

Knowledge-Based Approach

Developed as an expert system by intelligent decision-making based on stored knowledge

Data integrity, Time and Cost, Emotionless

Disadvantages of Expert System

May become large and lead to enormous searches

Disadvantages of Semantic Networks

GPGPU(General Purpose GPU)

GPU is developed into __________ which is used for general data processing

Data Driven Approach

How to accumulate many case data and make decisions with knowledge extracted from these data

Big data power

ICT development for Data Producing

Representation, Reasoning, Search

Knowledge processing ->

machines

Learning and intelligence can be precisely described and implemented on ________

Inference and Planning

Logically deduce and infer from the knowledge to get the desired results and make decisions

Third AI Boom

Machine Learning and Deep Learning, Professor Jeffrey Hinton (2012)

Knowledge-Base

Meaning a set of facts and rules, accumulate in advance through the knowledge acquisition system

Algorithm

Open Source Software

System Components

Perception, Knowledge-Base, Inference Engine

Inference Engine

Present the most suitable behavior and answer according to given input data and surrounding environment

Heuristic Search

Rough guess by experience, how to find the most plausible way

volume and velocity

Smartphones & Internet of Things will drive both the _____________ of data

Deep Learning(DL)

Solving complex nonlinear problems based on artificial neural network theory

Search Tree

The ____________ of real-world problems is generally very complex

Perception

The ability to accept external environmental information through visual and auditory functions from the outside world, sometimes with natural language input

perception, cognition, action

Three key steps of a knowledge-based agent

computer intelligence

Turing defines ______________ as the ability to produce human-level performance in cognitive tasks

Singularity

_________ occurs when machine begins to outperform human ability

AI

__________ should augment, not replace human expertise!

Machine learning

____________ can (help to) solve many problems, but is no panacea

Intelligence

ability to think, understand, and act instead of acting instinctively or automatically

Learning

adapt internal representation so that it is as accurate as possible

OWL

adds semantics to RDF

Semantic Web

an effort to create a web that uses the concepts from semantic networks

Turing test

defined by Alan Turing in 1950; -> Person C alternates between object A (machine) and object B (person) for Q&A: if the interrogator C can not distinguish the answer from a human or a machine, the computer has passed the test.

Second AI Boom

early 1980s, Expert Systems

Computing power

powerful parallel and distributed processing power

RDF

representation of information/data for the purpose of sharing

Strong AI

technologies to create human-like intelligence machines

Weak AI

technologies to solve specific problems by imitating human intelligence

Artificial Intelligence

technology that implements intelligence such as human cognitive ability (language, voice, vision, emotion), learning, and reasoning

Graphics Processing Unit (GPU)

the core of the graphics card that plays the role of processing images

First AI boom

the late 1950s, research on search and reasoning, very optimistic

RDF (Resource Description Framework) and OWL (web ontology language)

two main representations for the semantic web

Data Driven Approach

•Common property extraction from signal data •Issues, training, machine learning

Knowledge-Based Approach

•Representing knowledge as a combination of symbols •Understanding Issues, Language, and Knowledge

System Objectives

•Solving real world problems; •Intelligent Service Applications;