ARTIFICIAL INTELLIGENCE

Course Notes

INFOB2KI 2018-2019

Utrecht University The Netherlands

Acknowledgement

These course notes consist of a series of lecture slides which can be found on the

course website of the UU Bachelor course ‘Kunstmatige Intelligentie’:

http://www.cs.uu.nl/docs/vakken/b2ki/schema.html

The series of slides was originally composed by F.D. Dignum (UU), combining

material from various sources. Since 2013, revisions and updates have been done

by S. Renooij (UU).

We acknowledge and are grateful for input by

- Authors of the textbook used (I. Millington & J. Funge; S. Russel & P. Norvig )

- T. Mitchell

- UC Berkeley (D. Klein, P. Abbeel, and others involved in course 188)

- and many others

If we should explicitly mention anyone else on this list, or have unknowingly used

material that we aren’t allowed to use, please notify us.

2

ARTIFICIAL INTELLIGENCE

Lecturer: Silja Renooij

Introduction to course and subject

Utrecht University The Netherlands

These slides are part of the INFOB2KI Course Notes available fromwww.cs.uu.nl/docs/vakken/b2ki/schema.html

INFOB2KI 2018-2019

INFOB2KI� Vakpagina: http://www.cs.uu.nl/docs/vakken/b2ki/

� Boek: Ian Millington & John Funge, AI for Games, 2009, 2de Editie

+ extra materiaal

� Docent: Silja Renooij

� Assistenten: zie http://www.cs.uu.nl/docs/vakken/b2ki/contact.html

� Cijferregeling: – Twee projecten (B,C) P = (PB + PC)/2 � 35% van eindcijfer

– Twee tentamens T= (T1 + T2)/2 � 65% van eindcijfer

– ~ T en P moeten minstens 5 zijn; precieze details (ook voor reparatie) op vakpagina

(http://www.cs.uu.nl/docs/vakken/b2ki/beoordeling.html)

� Tentamens: op 14 december en op 1 februari; beide gesloten boek!

� OER: deelresultaten blijven alleen geldig binnen een academisch jaar.

4

INFOB2KI: verwachtingenJe neemt zelf verantwoordelijkheid voor het leveren van de inspanning die nodigis om dit vak te halen, dwz:

� Je kent en houdt je aan alle regels en deadlines

� Je weet welke stof is behandeld en je hebt ermee geoefend

� Zonodig raadpleeg je zelf aanvullend materiaal

Wij bieden ter ondersteuning:

� de vakpagina: http://www.cs.uu.nl/docs/vakken/b2ki/ , met o.a.

– het weekrooster: een gedetailleerd overzicht van de stof , inclusief college slides, en bijbehorende opgaven

– de studiewijzer (tab literatuur) met verwachtingen per onderwerp, veel-gemaakte fouten, werkcollegeopgaven en uitwerkingen

– oude tentamens

� hoorcolleges, en begeleide practicum- en werkcollegesessies.

Advies: maak hier gebruik van!

5

INFOB2KI: oefenopgaven

� Zijn te vinden in de studiewijzer (relevante opgaven per college aangegeven in het schema (tab weekrooster) op de vakpagina;

� Oude tentamens staan op de pagina met `oefenopgaven’; (zie hier ook voor info over meer uitwerkingen).

� Voor hulp en vragen is er een werkcollege

� Evt kan je zelf goede uitwerkingen aanleveren.

� Help ook elkaar via discussie forum op Blackboard.

6

INFOB2KI: projecten (I)

� op basis van de Pacman projects (Berkeley)

� in Python (� doe deze week de tutorial!)

� met 3 personen (uitzondering: 2)

– zoek vandaag nog practicumpartners, evt via forum op Blackboard!

� Programmeerwerk: middel– beperkt, binnen bestaand raamwerk

– autograder

– moet van helder commentaar voorzien worden, waaruit begrip blijkt

– zie projectpagina voor regels mbt hergebruik code (beinvloedt cijfer!)

� Parallelle vragen: doel– soms extra uitleg, vaak opzet en resultaten van experimenten

– punten voor gemotiveerde toelichting ipv antwoord

– kost veel tijd

– verslag; weegt zwaarst(!)

7

INFOB2KI: projecten (II)

� Niet onderschatten(!): – weinig programmeren, meer theorie, testen, verbeteren,

experimenteren, en verslaglegging!

– zie projectbeschrijving voor toelichting wat we verwachten van experimenten

� motiveer en documenteer: – al je keuzes, op een wetenschappelijke manier

– dus bijv niet alleen toelichten waarom je A kiest, maar ook waarom B en C niet.

� Zie belangrijke details op website (project pagina: Projects.pdf)– mbt wat wel en niet submitten, gezamenlijke verantwoordelijkheid en

fraude/plagiaat)

8

Outline

� What is (Artificial) Intelligence?

� Some history + examples

� Different types of AI + current focus

� Game AI: what’s special?

� Course overview

9

What’s AI?

10

What do you call a blonde who has

dyed her hair brown?

First: what’s intelligence?

� Measured by IQ score (?)

� A very general mental capability that, among other things,

involves the ability to reason, plan, solve problems, think

abstractly, comprehend complex ideas, learn quickly and

learn from experience. It is not merely book learning, a

narrow academic skill, or test-taking smarts. Rather, it

reflects a broader and deeper capability for

comprehending our surroundings — "catching on,"

"making sense" of things, or "figuring out" what to do.

Gottfredson, L.S. (1997). "Foreword to "intelligence and social policy"“. Intelligence 24 (1): 1–12.

11

Abridged (pre)history of AI

� 1943 McCulloch & Pitts: Boolean circuit model of brain

� 1950 Turing's "Computing Machinery and Intelligence"

� 1956 Dartmouth meeting: "Artificial Intelligence" adopted

� 1952- 69 Look, Ma, no hands!

� 1950s Early AI programs, including Samuel's checkers program, Newell & Simon's Logic Theorist, Gelernter's Geometry Engine

� 1965 Robinson's complete algorithm for logical reasoning

� 1966 - 74 AI discovers computational complexityNeural network research almost disappears

� 1969 - 79 Early development of knowledge-based systems

� 1971 Terry Winograd’s SHRDLU dialogue system

� 1980 - 88 Expert systems industry booms

� 1988 - 93 Expert systems industry busts: “AI winter”

� 1985 - 95 Neural networks return to popularity

� 1988 - AI becomes a science; resurgence of probability, general increase in technical depth; “Nouvelle AI” instead of symbolic

12

Contributing fields (AI prehistory)

� Philosophy Logic, methods of reasoning, mind as physical system foundations of learning, language,rationality

� Mathematics Formal representation and proof algorithms,computation, (un)decidability, (in)tractability,probability

� Economics utility, decision theory

� Neuroscience physical substrate for mental activity

� Psychology phenomena of perception and motor control,experimental techniques

� Computer building fast computers engineering

� Control theory design systems that maximize an objectivefunction over time

� Linguistics knowledge representation, grammar

13

What is Artificial Intelligence?

The dream of AI has been to build...

“...machines that can think, that learn and

that create.”

Dijkstra (1984):

“The question of whether Machines Can

Think... ...is about as relevant as the question

whether Submarines Can Swim”

14

Strong and weak AI

One may dream about... (strong AI)

� ...computers that can be made to think on

a level at least equal to humans, that can

be conscious and experience emotions.

This course is mostly about... (weak AI)

� ...adding “thinking-like” features to

computers to make them more useful

tools. That is, “not obviously machine like”.

15

Weak AI

Weak AI:

� Flexible category: as soon as we understand the

workings of an AI system, it seems less “intelligent”

� As soon as a part of AI is successful, it becomes an

own research area!

E.g. large parts of advanced search, parts of language

understanding, parts of machine learning and

probabilistic learning etc. etc.

16

AI in the 1990s� 1991 AI logistics planning and scheduling (50,000 vehicles,

cargo, and people) by US forces in Gulf war

� 1995 No hands across America (driving autonomously 98% of the time from Pittsburgh to San Diego)

� 1995 - The emergence of intelligent agents, everywhere

� 1996 AI proved a mathematical conjecture (Robbins conjecture) unsolved for decades

� 1997 Deep Blue defeated the reigning world chess champion Garry Kasparov

� 1996 - NASA's on-board autonomous planning program controls scheduling of operations for spacecraft

� 1999 Proverb solves crossword puzzles better than most humans

17

Deep Blue and AI

Is Deep Blue AI?http://www.cs.uu.nl/docs/vakken/b2ki/Videos/deepblue.wmv

18

What is AI?

Views of AI fall into four categories:

Thinking humanly Thinking rationally

Acting humanly Acting rationally

Most textbooks advocate "acting rationally“

But we will also look at “acting humanly”

19

Thinking humanly: cognitive modeling

� 1960s "cognitive revolution": – information-processing psychology

� Requires scientific theories of internal activities of the brain

� How to validate? Requires

1) Predicting and testing behavior of human subjects (top-down)

or

2) Direct identification from neurological data (bottom-up)

� Both approaches (roughly, Cognitive Science and Cognitive Neuroscience)

are now distinct from AI

20

Thinking rationally: "laws of thought"

� Aristotle: what are correct arguments/thought processes?

� Several Greek schools developed various forms of logic: notation and rules of derivation for thoughts; may or may not have proceeded to the idea of mechanization

� Direct line through mathematics and philosophy to modern AI

� Problems: 1. Not all intelligent behavior is mediated by logical deliberation

2. What is the purpose of thinking? What thoughts should I have?

21

Acting humanly: Turing Test

� Turing (1950) "Computing machinery and intelligence":

"Can machines think?" � "Can machines behave intelligently?"

� Operational test for intelligent behavior: the Imitation Game

� Predicted that by 2000, a machine might have a 30% chance of fooling a lay person for 5 minutes

� Anticipated all major arguments against AI in following 50 years

� Suggested major components of AI: knowledge, reasoning, language understanding, learning

22

2008: Turing Test for Bots

� Contest aim: see if a computer game playing bot can play like a human.

� Can [computers] be programmed to be more fun to play – to play like you

and me? People like to play against opponents who are like themselves –

opponents with personality, who can surprise, who sometimes make

mistakes, yet don’t blindly make the same mistakes over and over. Can a

computer […] seem to have personality, fallibility and cunning?

� First contest: Perth, Western Australia in December 2008.

# of bots to convince 4 out of 5 expert judges of > 50% humanness: none

(some did fool 2 out of 5)

� In 2012 two bots passed the test and fooled the judges for the first time!

(52% humanness)

23

Acting rationally: rational agent

� Rational behavior: doing the right thing

� The right thing: that which is expected to

maximize goal achievement, given the available

information

� Doesn't necessarily involve thinking – e.g.,

blinking reflex – but thinking should be in the

service of rational action

24

Is Watson AI?

Watson Playing Jeopardyhttp://www.cs.uu.nl/docs/vakken/b2ki/Videos/watson.wmv 25

Is Watson AI?

Watson makes (human) mistakes 26

Watson in health care and beyond…..

2012: IBM hopes its Watson will become doctor's sidekick

� broad domain, reason (combine) evidence

� Learn (See overview and some insights on youtube (1 hour movie):

https://www.youtube.com/watch?v=U_KhvJyjZ6c&feature=related)

August 2014: Watson Will Be Probing The Mysteries Of Physics

Deutsches Elektronen-Synchrotron – one of the largest research centers in Germany – has

entered into a collaboration with IBM to enhance its computers’ abilities to crunch the massive

amounts of data being generated every day by its particle accelerators.

June 2018: Project Debater https://www.ibm.com/blogs/research/2018/06/ai-debate/

Currently: anyone can use Watson on the IBM cloud…27

AI: this millenium

� 2001 - Very large datasets: Google gigaword corpus

� 2003 - Human-level AI back on the agenda

� 2004 - NASA's rovers autonomously navigate surface of Mars

� 2007 Checkers is solved by researchers at Univ of Alberta

� 2009 Google builds self driving car

� 2011 IBM Watson wins Jeopardy

� 2011 Rubiks Cube solved by robot in 10 seconds

� 2012 US state of Nevada permits driverless cars

� 2012 bots pass Turing test for first time

� 2013 “AI systems won't be enrolling in kindergarten this year …

But they may soon drive your child there” (journalist @ AAAI)

� 2014 13 year old ‘Eugene’ passes Turing test for first time

� 2017 DeepMind's AlphaGo wins against world Go champion

� 2017 Saudi-Arabia grants citizenship to humanoid Sophia

� 2017 Captcha solved (no longer just in theory)

� 2018 AlphaGo team receives Marvin Minsky Medal for Outstanding Achievements AI 28

AI in the news: just last month

� AI painting auctioned at Christie’shttps://www.christies.com/features/A-collaboration-between-two-artists-one-human-one-

a-machine-9332-1.aspx

� AI creates pornographic arthttps://www.volkskrant.nl/wetenschap/dit-is-porno-volgens-een-computer-~bd1ddfd3/

� AI designs perfumehttps://www.vox.com/the-goods/2018/10/24/18019918/ibm-artificial-intelligence-

perfume-symrise-philyra

� AI makes games

� iBorderCtrlhttps://www.iborderctrl.eu/

� …..

29

Rational agents� An ‘agent’ is an entity that perceives and acts

� For any given class of environments and tasks, we seek the agent with the best performance

� Computational limitations � perfect rationality unachievable30

Reflex agent: simple or model-based; goal-based agent; learning agent

Properties of agents (desired)

� autonomous � independently follows its goals

� social � interaction human/ other agents

� pro-active � goal-directed

� reactive � responds (in time) to changes

Latter two can be at odds with one another!

31

Environments

� (in)accessible

state info is complete, accurate, up-to-date

� (non)-deterministic

actions have guaranteed effect, no uncertainty

� (non)-episodic

no ‘history’

� static (vs dynamic)

only changes due to agent actions

� discrete (vs continuous)

finite number of actions and percepts

Game?semi; non; non; dynamic; continuous

32

AI in Games

33

AI in Games

� Entertainment games

– Strategy games (Turn-Based; Real-Time)

– Role-playing games

– Action games (FPSs; Sports; Platform)

– Simulations

– Puzzles

� Serious games

– Training environments

– Educational games

– Simulations

34

How AI contributes to games

� Tasks– Pathfinding

– Bot behavior (chasing, guarding…)

– Reputation

– Strategic reasoning

– Group behavior

– Learning

– Etc.

� Techniques

� A*

� Rule systems

� Neural networks

� Artificial evolution

� Planning

� Finite state machines

� Blobs of random code

� Etc.

35

AI Tech Situation� As fidelity of worlds grows,

increasing challenge of “just keeping up” , e.g.

– Pathfinding

• Easy on a 2d tilemap with 90 degree walls

• On an arbitrary polygon mesh, not as easy

– Combat posture

• Easy to switch between idle and combat sprites

• managing 100+ bone model, not as easy

� Just keeping old features working is hard

36

“AI” is not the same as visual representation

AI No AI

37

“AI” is not the same as complexity

38

The Complexity Fallacy

� It is a mistake to think that complex AI equals better

character behaviour

– Complex AI might come out looking stupid

– While simple AI might come out looking perfect

“Clever AI was programmed

into the game. The ghosts

would group up, attack the

player, then disperse. Each

ghost had his own AI.”

39

Game AI lies in perception

GamePlayer

Goals BeliefsEmotions

40

Game Architecture

Controllers(Game AI+

Human)

Controllers(Game AI+

Human)

Game StateSimulator

(Physics+Animations)

RendererAgents

(controlled by

Game AI or

Human)

actions

information

information information

updates

41

Animated games:Behavior = AI + physics

Behavior Code

Physics AI

42

The Illusion of “Human” Behaviour

� Game AI is about the illusion of human behaviour

– Smart, to a certain extent

– Unpredictable but rational decisions

– Emotional influences

– Body language to communicate emotions

– Being integrated in the environment

� Due to the increasing realism of game worlds,

evoking the illusion becomes harder every day

43

Game AI vs ‘Serious’ AI

� In game AI, non-obvious cheating is very common:

– adding markers to game world for e.g.

• path-planning, cover recognition, …

– looking through obstacles

– awarding small advantages

– teleportation

– ...

� In game AI, easy way to achieve an effect is acceptable from

the point-of-view of entertainment

� For serious applications, implementing the “right” (academic)

solution is preferable

This course will address techniques from both worlds.44

Course overview

� Introduction (chapters 1,2)

� Pathfinding & Search (chapter 4)

� Learning & Decision making (chapters 5,7,8)

� Uncertainty & Planning (chapter 5)

� Motion, Strategy & Tactics (chapters 3,6,11)

- Book chapters + additional material!

- Slides (per class, in schedule) +

exercises/projects

45