Artificial Intelligence 01 introduction

Artificial IntelligenceIntroduction

Andres Mendez-Vazquez

May 15, 2016

1 / 48

Outline

1 MotivationWhat is Artificial Intelligence?Acting humanly: The Turing Test ApproachImplications of the Turing TestSome Issues About the Turing TestThinking HumanlyThinking Rationally: Use of LogicAct Rationally

2 Strong AI vs. Weak AIDefinitionAgainst Strong AISearle’s Chinese Room

3 History of AIThe Long DreamModern TimesFragmentation Years

2 / 48

Outline

3 / 48

Introduction

The ConceptInformally, we have four main fields to define AI

Systems that think like humans Systems that think rationally

Systems that act like humans Systems that act rationally

This divide the field in two main groupsA human-centered approach, an empirical science, involvinghypothesis and experimental confirmation.

A rationalist approach involves a combination of mathematicsand engineering.

4 / 48

Introduction

4 / 48

Introduction

4 / 48

Introduction

4 / 48

We have a PROBLEM!!!

Did you notice the following?There is not a single viable definition of intelligence...

OOPSSS!!!

5 / 48

Actually the situation is much worse

Something NotableAt MIT’s "Brains, Minds and Machines" symposium, 2012

I Chomsky contends that many AI theorists have gotten bogged downwith such things as statistical models and fMRI scans.

He told themAI developers and neuroscientists need to sit down and describe theinputs and outputs of the problems that they are studying.

I Something that they do not actually do.... OOPSSS!!!

6 / 48

We have harsher words

Sydney BrennerGeneticist and Nobel-prize

He went to say thatHe was equally skeptical about new system approaches to understandingthe brain.

He went to say thatThe new AI and neuroscientist approach is some “form of insanity”

7 / 48

Brenner’s Criticism

An unlikely pairSystem Biology - a computational and mathematical modeling ofcomplex biological systemsArtificial Intelligence - attempts for “intelligence” in machines

ProblemBoth face the same fundamental task of reverse-engineering a highlycomplex system whose inner workings are largely a mystery.

Why?Although ever-improving technologies yield massive data related tothe system!!!

I Only a fraction of it is relevant!!! Question Which one?

8 / 48

This is good but....

The ControversyIt will keep raging for the foreseeable future!!!

Therefore, we will use this classificationSystems that think like humans Systems that think rationally

9 / 48

This is good but....

The ControversyIt will keep raging for the foreseeable future!!!

Therefore, we will use this classificationSystems that think like humans Systems that think rationally

9 / 48

Instead

We will look at this classificationAs a way to human beings try to solve problems...

Thus, we have the following new hierarchySystems that think like humans Systems that think rationallySolving problems as humans Solving problems using logic

Basically Solving Problems

⇓Systems that act like humans Systems that act rationally

Resulting of solving problems as humans Resulting of solving problems logically

10 / 48

Instead

We will look at this classificationAs a way to human beings try to solve problems...

Thus, we have the following new hierarchySystems that think like humans Systems that think rationallySolving problems as humans Solving problems using logic

Basically Solving Problems

⇓Systems that act like humans Systems that act rationally

Resulting of solving problems as humans Resulting of solving problems logically

10 / 48

Outline

11 / 48

The Turing Test

You have...I A human judge engages in a

natural language conversationwith one human and onemachine, each of which triesto appear human.

I All participants are placed inisolated locations.

I If the judge cannot reliablytell the machine from thehuman, the machine is said tohave passed the test.

12 / 48

The Turing Test

12 / 48

The Turing Test

12 / 48

The Turing Test

12 / 48

Outline

13 / 48

Implications of the Turing Test

Passing the Turing Test has implications in the following fieldsNatural Language Processing

I The machine needs to understand what you are saying.

Knowledge representationI A precise talk needs a good knowledge representation of the subject.

Automated ReasoningI Without logic who cares what are you saying

Machine LearningI Learn to adapt depending on the data.

14 / 48

Total Turing Test’s Implication

Total Turing TestIt uses a video signal so that the interrogator can test the subject’sperceptual abilities.

Computer VisionI It is used to perceive objects.

RoboticsI A way to manipulate objects and to move in the environment

15 / 48

Outline

16 / 48

Is the Turing Test Relevant?

Some researchers have pointed out that the Turing test is not enoughto talk about intelligent machines.

In the most extreme John Searle, professor of philosophy at UCBerkeley published “The Chinese Room” paper.He claimed that Strong AI is not even possible!!!

17 / 48

Is the Turing Test Relevant?

Some researchers have pointed out that the Turing test is not enoughto talk about intelligent machines.

In the most extreme John Searle, professor of philosophy at UCBerkeley published “The Chinese Room” paper.He claimed that Strong AI is not even possible!!!

17 / 48

Recently

Eugene GoostmanThe computer program designed by a team of Russian and Ukrainianprogrammers.

Against 30 JudgesIt was able to fool them 33% of the time

HoweverGraeme Hirst (University of Toronto) et al. dismissed the test because theTuring Test requires 50%.

18 / 48

Recently

18 / 48

Recently

18 / 48

Outline

19 / 48

Thinking Humanly: Cognitive ApproachSome Rea searchers

They think that we should understand the human mind.I Question: Understanding how the human mind solve problems and

react to the environment?

Three ways of doing thisThought’s InspectionPsychological experimentsBrain Imaging

I Also known as Cognitive Brain Imaging...

ExampleNewell and Simon used the traces of their General ProblemSolver (GPS) to compare the traces generated by humansubjects when solving the same problem.

20 / 48

Drawbacks of the Cognitive ApproachThought’s Inspection

To do this is quite difficult because you require snapshots of thethought process...

Psychological experimentsStatistics are quite iffy!!!Reproducibility Problems!!!Bias Problems!!!

Cognitive Brain ImagingResolution problem

I PET and MRI work at the range of mm, but you have in a cubicmm 1,000,000 neurons!!!

Difference Between IndividualsReproducibility and Replication Problems

21 / 48

Outline

22 / 48

Thinking Rationally: Use of Logic

Development of the formal logic in thelate 19th and early 20th century hasgive us:PROBLEM!!!

What?A precise notation about allkinds of thing in the worldand their relations betweenthem.

23 / 48

Thinking Rationally: Use of Logic

Development of the formal logic in thelate 19th and early 20th century hasgive us:PROBLEM!!!

What?It is not easy to takeinformal knowledge andstate in the way thelogical system need it.There is a big adifference between beingable to solve a problemin principle and doing inpractice.

23 / 48

Outline

24 / 48

Acting Rationally

Rational AgentsIn this approach, the agent acts so it can achieve its goals, givencertain beliefs about the environment.

It needs toIt needs to be able to make inferences.It needs to be able to act without inferences (HeuristicTriggers).

Norving and Company claim!!!It is more amenable to scientific development than

I Human behavior based models.I Human though based models.I After all the standards of rationality are clearly defined.

25 / 48

Acting Rationally

25 / 48

Acting Rationally

25 / 48

Acting Rationally

25 / 48

Acting Rationally

25 / 48

Acting Rationally

25 / 48

Acting Rationally

25 / 48

Outline

26 / 48

Strong AI vs. Weak AI

Strong AIStrong AI is artificial intelligence that matches or exceedshuman intelligence.

Weak AIWeak AI system which is not intended to match or exceed thecapabilities of human beings.

27 / 48

Strong AI vs. Weak AI

Strong AIStrong AI is artificial intelligence that matches or exceedshuman intelligence.

Weak AIWeak AI system which is not intended to match or exceed thecapabilities of human beings.

27 / 48

Outline

28 / 48

Arguments Against Strong AI

ArgumentsThe first argument against strong AI is that it is impossible forthem to feel emotions.The second argument against strong AI is that them cannotexperience consciousness.The third argument against strong AI is that machines neverunderstand the meaning of their processing.The fourth argument against strong AI is that machines cannothave free will.The fifth argument against strong AI is that God createdhumans as intelligent persons.

29 / 48

Outline

30 / 48

What is this?

Chinese RoomThe Chinese room was introduced in Searle’s 1980 paper "Minds, Brains,and Programs", published in Behavioral and BrainSciences.

Something NotableIt eventually became the journal’s "most influential target article".It is still generating an enormous number of commentaries andresponses.

David Cole, Philosophy Professor at University of Minnesota Duluth“The Chinese Room argument has probably been the most widelydiscussed philosophical argument in cognitive science to appear in the past25 years"

31 / 48

What is this?

31 / 48

What is this?

31 / 48

What is this?

31 / 48

Against Strong AISearle’s Experiment

Suppose that artificial intelligence research has succeeded inconstructing a computer that behaves as if it understandsChinese.Suppose, says Searle, that this computer performs its task soconvincingly that it comfortably passes the Turing test inChinese.Now, a human is in a closed room and that he has a book withan English version of the aforementioned computer program.

ThenThen, a human are given Questions in Chinese, and He or Shesimply answers them using the book.

Question!Does He/She understand Chinese?

32 / 48

IMPORTANT

The Chinese RoomIt is the most damaging argument against “Strong AI”!!!

Even with the criticism against itIt is still a lingering question that the people in AI still cannnot answer!!!

33 / 48

IMPORTANT

The Chinese RoomIt is the most damaging argument against “Strong AI”!!!

Even with the criticism against itIt is still a lingering question that the people in AI still cannnot answer!!!

33 / 48

Funny Observations

Something NotableMost of the discussion consists of attempts to refute it.

Something Notable"The overwhelming majority," notes BBS editor Stevan Harnad, "still thinkthat the Chinese Room Argument is dead wrong."

It is more, Pat Hayes - An important AI researcher pointed out thatCognitive science ought to be redefined as "the ongoing research programof showing Searle’s Chinese Room Argument to be false"

34 / 48

Funny Observations

34 / 48

Funny Observations

34 / 48

There is even a novel (Hugo Award Finalist)Blindsight

A hard science fiction novelBy PhD Marine-Mammal biologist Petter Watts

35 / 48

The human race confronts its first contact with terrifyingconsequences:

Conscious is not necessary... and the universe is full withnon-conscious intelligence!!!And the only way to survive is to allow an Hominid Vampire Branch(non-conscious) to exterminate the rest!!!

36 / 48

The human race confronts its first contact with terrifyingconsequences:

Conscious is not necessary... and the universe is full withnon-conscious intelligence!!!And the only way to survive is to allow an Hominid Vampire Branch(non-conscious) to exterminate the rest!!!

36 / 48

Other Arguments Against AIThere are other people

Penrose’s ArgumentI In “The Emperor’s New Mind (1989),” he argues that known laws of

physics are inadequate to explain the phenomenon of consciousness.I Highly Criticized because of the following claims

HowUsing a variant of the Turing’s Halting Problem to demonstrate that a system canbe deterministic without being algorithmic.

In addition, he claimed that consciousness derives from deeper level, finerscale activities inside brain neurons (Orch-OR theory).

HoweverA discovery of quantum vibrations in microtubules by AnirbanBandyopadhyay of the National Institute for Materials Science in Japan.

I It “could” confirm the hypothesis of Orch-OR theory.37 / 48

For more, read...

ArticleHameroff, Stuart; Roger Penrose (2014). "Consciousness in the universe:A review of the ’Orch OR’ theory". Physics of Life Reviews 11 (1): 39–78.

38 / 48

Book based in this theory: Hyperion

Hyperion By Dan Simmons - Here a Nucleus of AI are trying to usehuman brains to simulate their own consciousness

39 / 48

Outline

40 / 48

History of AI: In the Beginning

Antiquity:Greek myths of Hephaestus and Pygmalion incorporated theidea of intelligent robots (such as Talos) and artificial beings(such as Galatea and Pandora).Sacred mechanical statues built in Egypt.

384-322 B.C.Aristoteles described the syllogism a method of mechanicalthought.

800 A.D.Jabir ibn Hayyan develops the Arabic alchemical theory ofTakwin, the artificial creation of life in the laboratory.

41 / 48

In the Beginning

1206Al-Jazari created a programmable orchestra of mechanicalhuman beings.

1495-1500Paracelsus claimed to have created an artificial man out ofmagnetism, sperm and alchemy.Leonardo created Robots for Ludovico Sforza.

42 / 48

In the Beginning

42 / 48

In the Beginning

42 / 48

In the Beginning

42 / 48

Outline

43 / 48

Modern Times1600 -1650

John Napier discovered logarithms.Wilhelm Schickard created the first mechanical calculatingmachine.Pascal developed the first real calculator. Addition andsubtraction were carried out by using a series of very lightrotating wheels. His system is still used today in car odometerswhich track a car’s mileage.

1818Mary Shelley published the story of Frankenstein.

1822-1859Charles Babbage & Ada Lovelace worked on programmablemechanical calculating machines.

44 / 48

Modern Times

1861Paul Broca, Camillo Golgi and Ramon y Cajal discover thestructure of the brain

1917Karel Capek coins the term ‘robot.’

1938John von Neuman’s minimax theorem.

1950Alan Turing proposes the Turing Test as a measure of machineintelligence.

45 / 48

Modern Times

45 / 48

Modern Times

45 / 48

Modern Times

45 / 48

Modern Times1956-1974

The Golden Years – The Promise of an intelligent MachineI Movies like “The Forbin Project” promised computers with Strong AI

1974-1980First AI winter

I It is shown that many problems in AI are NP-Complete.I Many projects are stopped in AI.

1980-1987AI Revival Experts Systems, Knowledge Revolution.

1987-1993Second AI winter

I Fall of the Expert System Market and the LISP Machines.

46 / 48

Outline

47 / 48

Present - The Fragmentation Years - AI is still goingthrough a Winter

1993-PresentThe Fragmentation Years

I Computer VisionI RoboticsI Machine LearningI Fuzzy LogicI Bayesian NetworksI Evolutionary MethodsI etc

48 / 48

Artificial Intelligence 01 introduction

Engineering

Transcript of Artificial Intelligence 01 introduction