ECE 4524 Artificial Intelligence and EngineeringApplications

Spring 2018 – CRN 12983TR 3:30-4:45 in Randolph 220https://filebox.ece.vt.edu/~ECE4524

Today’s Schedule:

I What is AI?

I Introduce the Bounded Rational Agent

I Course objectives and content

I Course administration

What is AI?

ar-ti-fi-cial adjectivehumanly contrived often on a natural model : MAN-MADE

in-tel-li-gence nounthe ability to learn or understand or to deal with new or tryingsituations : REASON; also : the skilled use of reason (2) : theability to apply knowledge to manipulate one’s environment or tothink abstractly as measured by objective criteria (as tests)

Human intelligence resides in the mind, the product ofelectro-chemical activity in our brain.

Some statistics:

I 3 lbs

I 100 billion neurons

I 100 trillion connections(synapses)

1

Further reading:

I Pinker, ”How the Mind Works”

I Marcus, The Birth of the Mind: How a Tiny Number of GenesCreates the Complexities of Human Thought.

1Image by Patrick Lynch, CC BY 2.5

What kind of intelligence are we talking about

Howard Gardner, 1983 ”Frames of Mind: The Theory of MultipleIntelligences”

I musical

I visual

I verbal

I logical

I bodily

I interpersonal

I intrapersonal

I naturalistic

I moral?

How did we acquire it?

30,000 to 50,000 years ago humansbecame behaviorally modern. There aremany theories about what drove thischange. Two examples are:

I Social brain hypothesis

I Sexual selection

Ridley, 2003

Example milestones in human development

I 2 yo - recognize a wide variety of objects

I 4 yo - astounding language comprehension and production

I 6 yo - enough dexterity to tie shoes

I 8 yo - social skills develop, emotional intelligence

Some definitions of AI

Rich and Knight - An attempt to program computers to do thingsthat people can do better

Winston - A study of the computations that make it possible toperceive, reason, and act

Our working definition will be: an attempt to programcomputers to do things that would be said to requireintelligence if people did them.

Often the simplest tasks for humans are incredibly difficultin AI.

Often the simplest tasks for humans are incredibly difficultin AI.

Bill and Mary went out to a restaurant. After being seated theyordered from the menu. Later, they paid the check and went home.

But, often the simplest tasks for computers are difficult forhumans.

After I read the following, try to recall it:

I 1 5 7 4 0 2 4 8 5 1

I 5 4 0 2 3 1 6 6 5 8

But, often the simplest tasks for computers are difficult forhumans.

After I read the following, try to recall it:

I 1 5 7 4 0 2 4 8 5 1

I 5 4 0 2 3 1 6 6 5 8

But, often the simplest tasks for computers are difficult forhumans.

After I read the following, try to recall it:

I 1 5 7 4 0 2 4 8 5 1

I 5 4 0 2 3 1 6 6 5 8

And, humans often make mistakes

Partial list of human biases and cognitive errors

I Confirmation bias

I Framing

I Hindsight bias

I Availability Heuristic

I Anchoring

I IKEA effect

I on and on and on

http://en.wikipedia.org/wiki/List_of_cognitive_biases

The rotating snake illusion (Akiyoshi Kitaoka)

http://www.ritsumei.ac.jp/~akitaoka/rotsnake.gif

This is one of my favorites.

Margaret Thatcher (Peter Thompson, 1980)

This is one of my favorites.

Margaret Thatcher (Peter Thompson, 1980)

Course Topics

I State-space search, solve complex problems by defining astate, transitions between states, and search for a solution

I Knowledge representation and reasoning, represent facts anduse automated reasoning to arrive at conclusions

I Dealing with uncertainty and inference, uncertainty asprobability and statistical inference

I How to make systems learn, create models with parametersthat can be found given examples

The Rational Agent

The course is centered on the concept of a rational agent.

I Agents perceive their environment through sensors and acton the environment using actuators.

I A rational agent is one that makes decisions to optimize itsperformance given the percept sequence and any priorknowledge.

This tuple of descriptions, Performance, Environment, Actuators,Sensors, denoted PEAS characterizes a system.

PEAS == EASP

I Environment: the aspects of the environment the agentinteracts withobservable? deterministic/stochasticepisodic/sequential static/dynamicdiscrete/continuous single/multiple agent

I Actuators: what actions can the agent take, affects theenvironment

I Sensors: what percepts can the agent acquire

I Performance: embodies the criterion for success of an agentsbehavior as seen by its effect on the environment

There are a variety of basic agent architectures.

I Simple Reflex

I Model-based

I Goal-based

I Utility-based

I Learning

Real systems usually end up as a hybrid among these.

Bounded Rational Agents

Rational does not mean omniscient. Agents are bounded by:

I prior knowledge - what information has been provided by thedesigner

I time/energy - how much time (or energy) is available forexploration, learning, and decisions (actions to be taken).

A rational agent takes the action that maximizes performancegiven resource constraints, e.g. the information and time available(Herbert Simon).

PEAS as a programming framework

A Problem is:

I An Environment

I An Agent with Sensors and Actuators

I A Performance measure

Consider a general programming interface that takes a problemdescription and simulates the agent.

Demos

I 8-Puzzle

I First-Order Logic

Course Schedule

I State-space search (1/23 - 2/8)

I Problem Set 1, Quiz 1

I Knowledge representation and reasoning (2/13 - 3/13)

I Problem Set 2, Quiz 2

I Dealing with uncertainty and inference (3/15 - 4/3)

I Problem Set 3, Quiz 3

I How to make systems learn (4/5 - 4/24)

I Problem Set 4, Quiz 4

Prerequisites

I ECE 2574 for knowledge of data structures and basicalgorithms, complexity theory, and programming ability. Iexpect you to be able to read, understand, and write programsof a substantial nature independently.

I Probability (STAT 4714) is needed for the uncertainty andlearning portions of the course.

Software

We will be using Python (2.7) forprogram implementations.

I Python is a dynamic, strongly typed language with clear,readable syntax

I high level dynamic data types

I extensive standard libraries and third party modules forvirtually every task

I has a REPL for fast experimentation

See http://www.python.org/ for pointers to documentation.Other resource links are available on the course website.

Texts and Resources

(AIAMA) Stuart J. Russell, PeterNorvig, Artificial Intelligence: AModern Approach, Prentice Hall;2nd or 3rd edition

The books website is also very helpful with code, demo links, etc.http://aima.cs.berkeley.edu/

Course Activities

I Readings - Each meeting has a section of the text, see thereading guides to help you

I Warmups - due before (usually) every lecture, test basicunderstanding of assigned reading

I Lectures - I try to make these as interactive and interesting aspossible

I Problem Sets - core of your learning, consists mostly of bothexercises and engineering design problems (programming)

I Quizes

Note the course is four credit hours.

Grading

The grades will be computed as follows:Warmups: 5% (Extra Credit)Problem Sets: 68% (15% each + PS0)Quizes 32% (8% each)

Collaboration and Late Policy

I All assignments must be turned in via Canvas by due date andtime

I No late assignments will be accepted, with the followingexception: you get three free late days (24 hour periods)during the semester to accommodate emergencies. To use oneor more of these, just submit the assignment via Canvas asnormal.

I All graded work is expected to be the original work of theindividual student unless otherwise directed.

Communication

Course Website: https://filebox.ece.vt.edu/~ECE4524

I syllabus, schedule, notes, etc.

I primary way materials are distributed.

Piazza: https://piazza.com/

I forum/wiki like software for QA, polls, announcements

I replaces email listserv, but has a configurable email digest

I good mobile apps

I use it to ask (and answer) questions

Canvas: https://vt.instructure.com

I take warmups

I submit problem set solutions

I grades posted

Course Goals

At the end of the course you will be able to ...

I formulate a problem as state space search and write aprogram to solve it

I capture knowledge using logic-based grammar an writeprograms that can automatically reason

I write programs that can handle uncertainty

I write programs that can learn from experience

Questions ?

Next Actions

I Start PS0 - this is largely to get started with python

I Read through the Python Tutorial

I Take warmup before noon on Thursday 1/18.