Lec-1 CS188

CS 188: Artificial Intelligence
Instructor: Satish Rao

CS 188: Artificial Intelligence
Instructor: Satish Rao

Course authors: Dan Klein, Pieter Abbeel.
Also: Stuart Russell, Anca Dragan...
Stuff
Course Signup and Piazza
Course Page:
I Tentative Schedule.
I Homeworks, policies,lecture slides, course policies, etc.
Course Page:
https://edge.edx.org/courses/course-v1:Berkeley+CS188+2017 FA17/info
Course Page:
https://edge.edx.org/courses/course-v1:Berkeley+CS188+2017 FA17/info
Communication:
I Piazza.
I Head TA email: mabaroudi@berkeley.edu
I Professor email: satishr@cs.berkeley.edu
Course Organization
Lectures: Tuesday/Thursday.
Course Organization
I Show up!
I Uh..Professor...no room.
I Webcast.
I Also, previous versions available.
Course Organization
I Show up!
I Uh..Professor...no room.
I Webcast.
I Also, previous versions available.
Discussions.
I Attend whichever. Best if choose one to get to know your TA.
I First week crowded, eases off.
I Get to know your TA.
I Some will be designated exam prep sections.
Course Workflow
Homework.
I Generally due Mondays.
I Based on class material.
I Welcome to solve together, submit alone.
Course Workflow
Homework.
I Generally due Mondays.
I Based on class material.
I Welcome to solve together, submit alone.
Academic Integrity.
I Autograded. Limited submissions. Can be reset at office hours?
I Should earn basically 100%.
Course Workflow (continued)
Projects.
I 5 Slip days
amongst all
projects.
Max of 2 per
project.
I 6 projects.
Groups: 1-2
people.
I Autograded.
Projects.
I 5 Slip days
amongst all
projects.
Max of 2 per
project.
I 6 projects.
Groups: 1-2
people.
I Autograded. Locally autograded.
Projects.
I 5 Slip days
amongst all
projects.
Max of 2 per
project.
I 6 projects.
Groups: 1-2
people.
Artificially
Projects.
I 5 Slip days
amongst all
projects.
Max of 2 per
project.
I 6 projects.
Groups: 1-2
people.
Artificially Intelligent
Projects.
I 5 Slip days
amongst all
projects.
Max of 2 per
project.
I 6 projects.
Groups: 1-2
people.
Artificially Intelligent Teaching!!!
I Go
Projects.
I 5 Slip days
amongst all
projects.
Max of 2 per
project.
I 6 projects.
Groups: 1-2
people.
I Go get
Projects.
I 5 Slip days
amongst all
projects.
Max of 2 per
project.
I 6 projects.
Groups: 1-2
people.
I Go get 100%.
Start early,
Projects.
I 5 Slip days
amongst all
projects.
Max of 2 per
project.
I 6 projects.
Groups: 1-2
people.
I Go get 100%.
Start early, vote often!

Projects.
I 5 Slip days
amongst all
projects.
Max of 2 per
project.
I 6 projects.
Groups: 1-2
people.
I Go get 100%.
Start early, vote often!

Course FunFlow (mostly)
Contests. (Optionally: mostly.)

I Submit your own agents and compete with each other!
I Cool names. Well perhaps not so cool, but nerd cool.
I AlphaGhost
I PacLivesMatter
I myTeam.py
I Twopac
I Shots and Goggles.
I Eh
I Mr Silly and His Best Friend.
Course Terror.
Course Terror.
Exams.
I Midterms: Oct 2 and November 9. (Tentative)
Course Terror.
Exams.
I Final. Wednesday, December 13, 2017.
Course Terror.
Exams.
I Final. Wednesday, December 13, 2017.
Exam prep sections will be announced as semester progresses.
Course Resources.
TAs!!! And me.

Course Resources.
TAs!!! And me.

Office Hours.
Course Resources.
TAs!!! And me.

Office Hours.
I Use these.
Course Resources.
TAs!!! And me.

Office Hours.
I Use these.
I Strongly attended near due dates.
Course Resources.
TAs!!! And me.

Office Hours.
I Use these.
I Go at other times!
Course Resources.
TAs!!! And me.

Office Hours.
I Use these.
I Go at other times!
I Some will be exam prep!
Prerequisites
Prerequisites
I 61A, 61B and 70.

Prerequisites
I 61A, 61B and 70.

I Lots of math.
Take math self-diagnostic on edge.edex. (not graded.)
Prerequisites
I 61A, 61B and 70.

I Lots of math.
I Lots of programming.
I Project 0 (P0) posted today/soon.
Prerequisites
I 61A, 61B and 70.

I Lots of math.
I Due Tuesday.
Prerequisites
I 61A, 61B and 70.

I Lots of math.
I Due Tuesday.
I Must submit.
Prerequisites
I 61A, 61B and 70.

I Lots of math.
I Due Tuesday.
I Must submit.
I Watch piazza.
Laptops in lecture
(starting next lecture)

I Prefer you dont use laptops/phones in lecture.
Laptops in lecture

I Please sit in back if you do use them.
Laptops in lecture

I Sit up front
Laptops in lecture

I Sit up front !!
Textbook (optional)
Russell and Norvig, AI: A Modern Approach
Textbook (optional)
Russell and Norvig, AI: A Modern Approach
Get Started!
This Week!
Get Started!
This Week!
I Register on edX.
Get Started!
This Week!
I Register on edX.
I Register on piazza main discussion and communication.
Get Started!
This Week!
I Register on edX.
I P0: Python tutorial is out now/soon. (Due Tuesday.)
Get Started!
This Week!
I Register on edX.
I Math self-diagnostic up on edX. Check now! Do again later.
Get Started!
This Week!
I Register on edX.
Also Important.
Get Started!
This Week!
I Register on edX.
Also Important.
I Sections start next week.
Get Started!
This Week!
I Register on edX.
Also Important.
I Regular Office Hours start next week.
Get Started!
This Week!
I Register on edX.
Also Important.
I Regular Office Hours start next week.
I Waitlist Questions Contact Cindy Conners
(csconners@berkeley.edu ) in 379 Soda.
Today
What is artificial intelligence?

Today

What can AI do?
Today

What can AI do?
What does this course do?
Useful now.
John Henry.
Useful now.
John Henry.
Replaced by a steam engine.
Useful now.
John Henry.
Replaced by a steam engine.
AI in the News
Existential threat.
AI in the News
Existential threat.
AI in the News
Existential threat.
Again with Elon
What are they doing.
The OpenAI researchers are

feeding message threads from the
popular website Reddit to
algorithms that build a probabilistic
understanding of the conversation.
If fed enough examples, the
underlying language model will be
good enough to hold a
conversation itself, the
researchers hope.
Center for Human-Compatible AI
Human suffering is bad...

Human suffering is bad...so put them out of their misery!

Human suffering is bad...so put them out of their misery!

Uh oh!
Peek at hardware
Architecture Neurons Num Synapses

Peek at hardware

Fly 100K = 105 10M = 107
Peek at hardware

Fly 100K = 105 10M = 107
AlexNet 650K = 106 60M = 108
Peek at hardware

Fly 100K = 105 10M = 107
AlexNet 650K = 106 60M = 108
Mouse 10M = 108 100M = 1011
Peek at hardware

Fly 100K = 105 10M = 107
AlexNet 650K = 106 60M = 108
Mouse 10M = 108 100M = 1011
Human 100B = 1011 1014 1015
Synapse 1 FLOP (i.e., can fire / not fire once per second)
Peek at hardware

Fly 100K = 105 10M = 107
AlexNet 650K = 106 60M = 108
Mouse 10M = 108 100M = 1011
Human 100B = 1011 1014 1015
human brain requires 1015 flops = 1 petaflop.
Peek at hardware

Fly 100K = 105 10M = 107
AlexNet 650K = 106 60M = 108
Mouse 10M = 108 100M = 1011
Human 100B = 1011 1014 1015
100,000 current CPUs
Peek at hardware

Fly 100K = 105 10M = 107
AlexNet 650K = 106 60M = 108
Mouse 10M = 108 100M = 1011
Human 100B = 1011 1014 1015
100,000 current CPUs costs $5000 / hr on Amazons EC2.
Why take this class?
In old days, cool, fun, the future,

In old days, cool, fun, the future, long years in grad school to have...
These days, cool, fun,
These days, cool, fun, neat applications, ...and
AI Booming in industry
And so on.
And so on.
What is AI?
The science of making machines that.

What is AI?
Think like
people
What is AI?
Think like
people
Act like
people
What is AI?
Think like Think

people rationally.
Act like
people
What is AI?
Think like Think

people rationally.
Act like Act rationally!

people
What is AI?
Think like Think

people rationally.
Act like Act rationally!

people
Rational Decisions
We use the term rational in a specific, techinal way:

Rational Decisions

I Rational: maximally achieving pre-defined goals.
Rational Decisions

I Rationality only concerns what decisions are made.
(not process behind them)
Rational Decisions

I Goals are expressed in terms of utility of outcomes.
Rational Decisions

I Being rational means maximizing your expected utility.
Better Title For Course.

Rational Decisions

I Being rational means maximizing your expected utility.
Better Title For Course.
Computational Rationality.
Maximize Expected Utility!
What About the Brain?
Brains good at rational decisions.


not perfect.

not perfect.
Brains not as modular as software.

not perfect.
hard to reverse engineer.

not perfect.
Brains to intelligence

not perfect.
Wings to flight

not perfect.
Wings to flight
Lessons:

not perfect.
Wings to flight
Lessons:
memory and simulation
guide decisions.
Framework: Rational Agents.
An agent:
An agent: percieves and acts.


A rational agent:

A rational agent:
acts to maximize (expected) utility.

A rational agent:
Percepts, environment, action space guide
rational actions.
Percepts
Sensor
Environment
Agent?
Actuator Actions

A rational agent:
rational actions.
This course:
Percepts
Sensor
Environment
Agent?
Actuator Actions

A rational agent:
rational actions.
This course:
Percepts
Sensor General AI techniques for various problem
Environment
types.
Agent?
Actuator Actions

A rational agent:
rational actions.
This course:
Percepts
Sensor General AI techniques for various problem
Environment
types.
Agent? Recognize when/how a problem can be
addressed with existing technique.
Actuator Actions
Pacman as an Agent
Percepts
Sensor
Environment
Agent?
Actuator
Actions
Demo: pacman-l1.mp4
Demo.
Demo:mplayer
Course Topics!
I Part 1: Making Decisions.

I Fast search/planning.
I Constraint Satisfaction
I Adverserial and uncertain search.
I Part II: Reasoning under Uncertainty
I Bayes nets.
I Decision Theory.
I Machine Learning.
AI
Machine Learning
learning decisions;
sometimes independent
Robots Rational Human-AI

Physically Agents Interaction
embodied. [decisions]
Computer NLP
Vision
A (Short) History of AI
Thinking Machine
Abbreviated History of AI
I 1940-1950: Early days.


I 1943: McCulloch and Pitts: Boolean circuit model of brain.

I 1950: Turings Computing Machinery and Intelligence.
Imitation Game.

Imitation Game.

Imitation Game.
I 1950 1970: Excitement: Look, Ma, no hands!

Imitation Game.
I 1950s: Early AI programs: Samuels checkers program,

Imitation Game.
Newell & Simons Logic Theorist,

Imitation Game.
Newell & Simons Logic Theorist, Gerlernters Geometry
Engine.

Imitation Game.
Engine.
I 1956: Dartmouth meeting. Adopted Term: Artificial
Intelligence

Imitation Game.
Engine.
I 1956: Dartmouth meeting. Adopted Term: Artificial
Intelligence
I 1965: Robinsons complete algorithm for logical reasoning.
Abbreviated History..( a bit more)
I 1970-1990: Knowledge-based approaches.


I 1969-1979: Early development of knowledge-based
systems.

systems.
I 1980-1988: Expert systems industry booms.

systems.
I 1988-1993: Expert systems industry busts.
AI Winter

systems.
AI Winter
I 1990 : Statistical approaches.

systems.
AI Winter
I Resurgence of probability, focus on uncertainty.

systems.
AI Winter
I General increase in technical depth.

systems.
AI Winter
I Agents and learning systems...AI Spring

systems.
AI Winter
I Agents and learning systems...AI Spring
I 2000 : Where are we now?
What Can AI Do?
Play a decent game of Jeopardy?

What Can AI Do?
Play a decent game of Jeopardy? Yes.

What Can AI Do?

Win against any human at chess?
What Can AI Do?

Win against any human at chess? Yes.
What Can AI Do?

Grab a specified cup and put it on a shelf?
What Can AI Do?

Grab a specified cup and put it on a shelf? Yes.
What Can AI Do?

Unload your dishwasher?
What Can AI Do?

Unload your dishwasher? No
What Can AI Do?

Drive safely along the highway?
What Can AI Do?

Drive safely along the highway? Maybe.
What Can AI Do?

Drive safely along Telegraph Avenue?
What Can AI Do?

Drive safely along Telegraph Avenue? No
What Can AI Do?

Drive safely along Telegraph Avenue? No ?
What Can AI Do?

Buy a weeks worth of groceries on the web?
What Can AI Do?

Buy a weeks worth of groceries on the web? Yes.
What Can AI Do?

Buy a weeks worth of groceries at Berkeley Bowl?
What Can AI Do?

Buy a weeks worth of groceries at Berkeley Bowl? No.
What Can AI Do?

Discover and prove a new mathematical theorem?
What Can AI Do?

Discover and prove a new mathematical theorem? Maybe.
What Can AI Do?

Perform a surgical operation?
What Can AI Do?

Perform a surgical operation? No.
What Can AI Do?

Perform a surgical operation? No. It can help though.
What Can AI Do?

Translate spoken Chinese into spoken English in real time?
What Can AI Do?

Translate spoken Chinese into spoken English in real time? Yes.
What Can AI Do?

Write an intentionally funny story?
What Can AI Do?

Write an intentionally funny story? No.
Unintentionally Funny Stories
I One day Joe Bear was hungry.

I One day Joe Bear was hungry. He asked his friend Irving Bird
where some honey was.
where some honey was. Irving told him there was a beehive in
the oak tree.
the oak tree. Joe walked to the oak tree.
the oak tree. Joe walked to the oak tree. He ate the beehive.
The End.
I Henry Squirrel was thirsty.
The End.
I Henry Squirrel was thirsty. He walked over to the river bank
where his good friend Bill Bird was sitting.
The End.
where his good friend Bill Bird was sitting. Henry slipped and fell
in the river.
The End.
in the river. Gravity drowned.
The End.
in the river. Gravity drowned. The End.
I Once upon a time there was a dishonest fox and a vain crow.
The End.
One day the crow was sitting in his tree, holding a piece of
cheese in his mouth.
The End.
cheese in his mouth. He noticed that he was holding the piece of
cheese.
The End.
cheese. He became hungry, and swallowed the cheese.
The End.
cheese. He became hungry, and swallowed the cheese. The fox
walked over to the crow.
The End.
walked over to the crow. The End.
The End.
[Shank, Tale-Spin Systems, 1984]

The End.
[Shank, Tale-Spin Systems, 1984]

Natural Language
Speech technologies (e.g. Siri, Google,

Alexa)
I Automatic speech recognition (ASR)
I Text-to-speech synthesis (TTS)
I Dialog systems
Natural Language
Language processing technologies.

I Question answering.
I Machine translation.
Computer Vision
Karpathy, Fei-Fei, 2015; Donahue et al., 2015, Xu et. al., 2015; and
many more.
In action
Arnold
In action
Arnold
Angst
Game Agents
Classic Moment:
Game Agents
Classic Moment:
May, 97: Deep Blue vs. Kasparov
Game Agents
Classic Moment:
+ First match won against world champion
Game Agents
Classic Moment:
+ Intelligent creative play
Game Agents
Classic Moment:
+ 200 million board positions per second
Game Agents
Classic Moment:
+ Humans understood 99.9 of Deep Blues
moves
Game Agents
Classic Moment:
moves
+ Can do about the same now with a PC
cluster
Game Agents
Classic Moment:
moves
cluster
- 1996: Kasparov Beats Deep Blue
Game Agents
Classic Moment:
moves
cluster
I could feel
Game Agents
Classic Moment:
moves
cluster
I could feel I could smell
Game Agents
Classic Moment:
moves
cluster
I could feel I could smell a new kind of
intelligence across the table.
Game Agents
Classic Moment:
moves
cluster
I could feel I could smell a new kind of
intelligence across the table.
- 1997: Deep Blue Beats Kasparov
Deep Blue hasnt proven anything.
Reinforcement Agents
Game Agent
[Schulman, Chen, Bartlett, Sutskever & Abbeel, 2016]

Angst!!!
Angst!!!
Simulated Agent
[Shulman, Moritz, Levine, Jordan, Abbeel, ICLR 2016]

Robotics
Robotics
Robotics
Robotics
+ Part mech. eng.
Robotics
Robotics
+ Part mech. eng.
+ Part AI
Robotics
Robotics
+ Part mech. eng.
+ Part AI
+simulations!
Robotics
Robotics
+ Part mech. eng.
+ Part AI
+simulations!
Technologies
Robotics
Robotics
+ Part mech. eng.
+ Part AI
+simulations!
Technologies
+Vehicles
Robotics
Robotics
+ Part mech. eng.
+ Part AI
+simulations!
Technologies
+Vehicles
+Rescue
Robotics
Robotics
+ Part mech. eng.
+ Part AI
+simulations!
Technologies
+Vehicles
+Rescue
+Help in the home
Robotics
Robotics
+ Part mech. eng.
+ Part AI
+simulations!
Technologies
+Vehicles
+Rescue
+Help in the home
+Lots of automation...
Robotics
Robotics
+ Part mech. eng.
+ Part AI
+simulations!
Technologies
+Vehicles
+Rescue
+Help in the home
In this class:
Robotics
Robotics
+ Part mech. eng.
+ Part AI
+simulations!
Technologies
+Vehicles
+Rescue
+Help in the home
In this class:
+ We ignore mechanical aspects
Robotics
Robotics
+ Part mech. eng.
+ Part AI
+simulations!
Technologies
+Vehicles
+Rescue
+Help in the home
In this class:
+ Methods for planning
Robotics
Robotics
+ Part mech. eng.
+ Part AI
+simulations!
Technologies
+Vehicles
+Rescue
+Help in the home
In this class:
+ Methods for planning
+ Methods for control
Robot learning
Interacting with AI: Very open
Interacting with AI: Very Open
Not so clear utility function
Clear utility function

Decision Making
Applied AI involves many kinds of

automation
Decision Making

automation
+ Route planning, e.g. Google maps
Decision Making

automation
+ Scheduling, e.g. airline routing, military
Decision Making

automation
+ Medical diagnosis
Decision Making

automation
+ Medical diagnosis
+ Web search engines
Decision Making

automation
+ Medical diagnosis
+ Spam classifiers
Decision Making

automation
+ Medical diagnosis
+ Spam classifiers
+ Automated help desks
Decision Making

automation
+ Medical diagnosis
+ Spam classifiers
+ Fraud detection
Decision Making

automation
+ Medical diagnosis
+ Spam classifiers
+ Fraud detection
+ Product recommendations
Decision Making

automation
+ Medical diagnosis
+ Spam classifiers
+ Fraud detection
+ ...
Decision Making

automation
+ Medical diagnosis
+ Spam classifiers
+ Fraud detection
+ ...Lots more!
CS188: Artificial Intelligence
And so it begins...
CS188: Artificial Intelligence
And so it begins...
See you next week!

Lec-1 CS188

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Lec-1 CS188

Hochgeladen von

Copyright:

Verfügbare Formate

CS 188: Artificial Intelligence

Instructor: Satish Rao

Instructor: Satish Rao

Start early, vote often!

Start early, vote often!

Contests. (Optionally: mostly.)

TAs!!! And me.

TAs!!! And me.

TAs!!! And me.

TAs!!! And me.

TAs!!! And me.

TAs!!! And me.

I 61A, 61B and 70.

I 61A, 61B and 70.

I 61A, 61B and 70.

I 61A, 61B and 70.

I 61A, 61B and 70.

I 61A, 61B and 70.

(starting next lecture)

(starting next lecture)

(starting next lecture)

(starting next lecture)

What is artificial intelligence?

What is artificial intelligence?

What is artificial intelligence?

The OpenAI researchers are

Human suffering is bad...

Human suffering is bad...so put them out of their misery!

Human suffering is bad...so put them out of their misery!

Architecture Neurons Num Synapses

Architecture Neurons Num Synapses

Architecture Neurons Num Synapses

Architecture Neurons Num Synapses

Architecture Neurons Num Synapses

Architecture Neurons Num Synapses

Architecture Neurons Num Synapses

Architecture Neurons Num Synapses

In old days, cool, fun, the future,

The science of making machines that.

The science of making machines that.

The science of making machines that.

The science of making machines that.

Think like Think

The science of making machines that.

Think like Think

Act like Act rationally!

The science of making machines that.

Think like Think

Act like Act rationally!

We use the term rational in a specific, techinal way:

We use the term rational in a specific, techinal way:

We use the term rational in a specific, techinal way:

We use the term rational in a specific, techinal way:

We use the term rational in a specific, techinal way:

Better Title For Course.

We use the term rational in a specific, techinal way:

Better Title For Course.

Brains good at rational decisions.

Brains good at rational decisions.

Brains good at rational decisions.

Brains good at rational decisions.

Brains good at rational decisions.

Brains good at rational decisions.