History of AI from 40s - 90s

 Alright, so, welcome to the AI course, from today we are going to talk about the technical

aspects of AI.

And before I start I want to mention that while I am going to be presenting the course,

I have I am really standing on the shoulders of giants in other words, there are a lot

of lot of people who have taught AI taught AI really, really well taught AI probably

much better than me, taught AI I have learned AI lot from them and so on so forth.

And I have learned from many such people and the slides that I will be using will be stolen

from different places of course with you know, some, some permission, some permission, I

would say.

And so, because of that, you know, we will be sort of taking interesting bits from wherever

I found and combining them in the course, I should also point out that I am going to

credit all the people that have taken slides from in the first slide, not on individual

slides, but you know, I really, really thank them for developing this course for me and

for themselves.

And for the field further, okay.

Now, AI course has been is being taught in pretty much every department in the world,

every top department in the world at least.

And if you look at the AI courses in different institutions, you may find sometimes the overlap

between topics is limited.

There are many ways to look at a broad field like AI you know, I remember that when I would

go to triple AI conferences as a student.

There will be 8 or 9 parallel tracks running and sometimes as a young student, even though

I was a researcher in AI, I may only understand things about one track and not understand

anything about the other 7 tracks.

It is a really broad field.

And the reason it is a broad field is that AI is not just a technical subject, and we

will understand it more than more over time.

But it comes from a philosophy.

AI is not a problem, which has one solution or 2 solutions.

AI that is a general problem, philosophical problem, which has many, many, many frustrations

into concrete problems, which have many, many, many different solutions.

So triple AI conference, or an AI conference often thinks of itself as an umbrella conference,

which brings all these people together.

Because AI has so many diverse sub fields, any course cannot do justice to all of them

any limited time course.

And therefore, if you look at the book, which is the Russell and Norvig book, by the way,

how many of you now have access to the book?

Very good.

Very good.

I am very happy to see that at least 20% of you have access to the book.

This is only the second class by next week, I hope all of you will have a copy of the

book with you, you as you must have seen for the people who have had looked who have looked

at the book, it is a big fat book and there is no way we are going to cover the big fat

book in the class.

So, any instructor has to make some calls on what it is that they want to cover in the

course.

Now, I think of it as a breadth course A breadth course, therefore, may not be deep.

So usually the trade off is between broad and shallow versus deep and narrow.

If you are one person who is broad and deep, you know all the power to you, I respect you.

But most people are not such people.

Most people are either on this side of the coin or that sort of thing.

And in fact, you are undergrad students.

So it is your time to learn something about everything.

It is your time to be broad.

Once you start doing a research project.

Once you go deeper into a particular assignment.

Once you go into a specific, you know, course project or what is not, that would be the

time for you to go deeper into something.

And you can only go deeper into a few things not too many?

So I am taking a view that we this would be an introduction to the breadth of ideas in

AI.

So I am going to introduce an idea and that idea if you are able to capture it, the core

of it and have it with you, it is going to help you in wide variety of settings, not

just in this specific topic in which we learned that particular idea.

So if you are a general computer scientist, not interested in AI in the long run, you

will still get a point of view and a set of general tools that will help you in attacking

a new problem and I am assuming that almost all of you will go into the industry or somewhere

there you will be solving problems and not saying that you will be solving computer science

problems also.

You will be solving problems technical problems at some level and this AI way of thinking

is going to help you and that particular way of thinking will in introducing the Next week,

For now, we are just motivating them and talking about what is AI and the history of it.

And if you are a serious AI enthusiast if you want to make AI career either in research

or in further study and you know a job and so on so forth.

This would give you an introduction to a lot of topics.

This will teach you about search then you can go and take a heuristic search course.

This will teach you about some basic networks, probabilistic graphical models and you can

take Ferro zinc law in probabilistic graphical models or Daphne Koller.

You know about Daphne Koller.

Daphne Koller was one of the founders of Coursera.

She was a professor at Stanford.

How probabilistic graphical models book and her courses are the most famous course in

this particular area.

You will learn a little bit about planning and decision making from there you can take

a full blown course on reinforcement learning, for example, Ferro zinc laws teaching a reinforcement

learning 800 level course in our department this semester.

Now you are not ready to take that course without doing the AI which gives you the introduction.

I mean, you can always pick it up, but eventually this course prepares you to take that advanced

course, if you are interested in and so on so for.

So, there are many, many sub topics of AI and this particular course is going to introduce

all of our many of them to you get.

The other decision as an instructor we have to make is, what do we focus on even within

the topics that we are going to explain.

And here, you all know about theory and practice, theory and applications those are the 2 extremes

of any scenario.

But there are 2 more steps in the middle the way I think about it, there is the theory,

there is the modeling, there is the algorithm and then there is the application in my view.

And let us think about this from an example.

And this is very important, this broad understanding of any problem is very important.

So, let us say I am I am in 2004 or 2003 and I am trying to create a MapQuest or a Google

Maps.

And there I need to decide that I have the whole map of the world, I need to figure out

that I want to go from location A to location B, what is the best way to reach location

B from location A?

Now, this is a real problem, this is a real application.

The first thing you will have to do is to make it into a computational problem.

You are not going to talk about which car am I driving you are not going to talk about

you know whether there is a gas station on this particular road not for now, I am going

to abstract out all this information.

I am not going to talk about whether there is a path hole in the road I am not I have

to abstract out all that information into a simplified approximated computational problem.

And what will be such a computational modeling can you think about it, it is very easy you

have done this graphs, you will take the model the computational model of a graph now graph

has node and edges.

So, you will say for my problem what is the node?

So, you could say that any intersection is the node see, we are talking about roads and

locations, we are not modeling every coordinate in the location.

We are abstracting it out if we start modeling every coordinate then it will be impossible

every little point we can model that we have different abstracted out.

So, in our model, we will have graphs and node will be an intersection the edge will

be a load segment that connects 2 intersections then, we can additionally have a weight on

the edge and the weight could be in this case, length is one possibility some people will

say time now you have to say oh in the morning the time is different in the evening the time

is different in the daytime the time is different.

If there is an accident the time is different.

If it is raining, the time is different, which time it is an abstraction.

Initially, we will say I do not want to worry about morning evening raining, I will just

take some average time.

So I will keep my Google cars going all around the city at different points in time and in

each road.

However time it takes I am going to just average it out or take the not if not average, you

know one deviation or whatever.

Then now I have converted into a graph.

Now notice this is called can you guess?

Modeling, this is modeling, taking a problem and converting it to a computational problem.

This is modeling this you have to do in everything in the world.

If you are a computer scientist, you will never get a computation problem almost never

unless with a theoretical researcher which will start from model you will always in the

world, you will get some real problem you do not even know how to start modeling it

into a competition problem.

Often you would see that for a real application a lot of the magic is a modeling because once

you model somebody some time has already solved that problem.

Once you model this is a graph and model on the graph as a shortest path problem.

Now, the all the literature on shortest path problems becomes applicable here.

You do not have to really be creative in solving this specific problem.

You just not have to read up the read up and check what are the various shortest path algorithms?

You know, there is a there is a Bellman Ford algorithm there is a Dexter's algorithm there

will be 20 others.

You just quickly go look at which one is the right one for this fit.

Bellman Ford is much better if it is if you have negative edges.

We do not have negative edges.

Nobody's gone a gain time.

We were not time travelling here, so you do not need Bellman forward, let us do something

simpler and so on.

I am not getting into the specifics of how Google and Microsoft made their map engines.

That is called the algorithm.

We model it is a problem, then we chose an algorithm for it, we may have to innovate

there.

But that is the algorithm, then we implement the algorithm.

And when we implement the algorithm, we have to decide on, you know, what is the full graph.

Now, when I am going from IIT Delhi to AIMS or hauz khas metro station, do I need to have

in my graph the edge from, you know, the highway edge from Seattle to Portland?

I do not the whole graph of the world is huge.

We do not need the whole graph for every problem.

So how do we cut it slices dices, where does the map get stored, which part of the map

do we put into memory etcetera.

All of these parts would be parts of the application solution they also always interact with the

algorithm obviously.

And last but not least, what can I prove about them?

can I prove that I will always get the shortest path under what conditions will I always get

the shortest path etcetera that will be called the Theory.

So, now, do you understand with an example, this is called you know explanation for example,

application modeling algorithm theory you can I guess I do not know which what will

be the right order the order is because you can prove theorems about the problem, what

is the complexity class of the problem, you can prove theorems about the algorithm.

So, therefore, theory interacts with both the modeling as well as the algorithm itself.

The algorithm is implemented in the actual application.

Now, in this course, we will tilt towards modeling and algorithms in your assignments

you would do application and we will not very much bother about theory.

And again, this is a choice once in a while I will introduce some theorems.

Maybe I may prove one, but mostly we will the book has theorems and proofs, we will

talk about the term but not talk about proof more often than not, we will mostly focus

on how it gets applied, how a new problem can get modeled how a problem a model may

have solutions, how what are the properties of each solution.

And so on so forth.

This is part that we are going to focus on in this particular course.

And of course, you can have any AI course where you go deep into the math and the theorems

and the theory and that is not what we are going to do.

Okay, any questions on this?

All right.

So what I am going to do today is I am going to talk about just start talking about the

history of AI.

So, in terms of the introduction, we will first talk about the history of AI the president

of AI why is there so much interest in here?

Then we will start asking the question, what is AI, right?

AI as a phenomenon is actually a very complicated question lot of people disagrees on what is

AI, so we will try to have our best definition possible and some recurrent themes in the

field of AI.

This would be our introduction, then we will start talking about the technical content

so, the initial first few lectures are going to be just introducing this subject for you

and let us start from one of the salient events way back in 1940's.

Where an ENIAC computers was constructed, designed, developed now, ENIAC stands for

electronic numerical integrator and computers.

It is a first electronic general purpose computer and it was Turing complete Turing complete

if for people who are advanced enough there is a Turing machine which the mathematical

model and technically with a fixed amount of tape the whatever to the machine can do

ENIAC can do too.

And it was capable of being reprogrammed to solve a large class of numerical problems

at the time.

Now, let us go back to the 40's By the way, this is one computer just so you know and

it is not it is much worse than the computers you and I have on our phones so offcourse,

we know this right.

What would I use the computer for if I am in the 40's and I suddenly have something

that can do computations was applied logistics that is a good suggestion you require probably

more algorithms to be developed before you can do logistics, but that is a reasonable

suggestion.

Yes, code breaking yeah code that I think I was trying some code working at the time

I do not know I do not think ENIAC itself was used for code breaking, anybody else.

So it was designed to compute artillery firing tables for the United States Army, but you

both had the right intuition in mind.

It was being used for defense purpose by the way this is not just the 40's.

I will claim that a lot of progress in AI has happened because of defense and was and

enemies.

This is I mean, we can smile about it but which in every country have you see the country

which puts in more money for science and less money for defense, at least some of the countries

which have more money, like significant amount of money.

We all know US government policies, I mean, they offcourse put a lot of money for science,

but they usually put money for science for defense, or engineering for defense is right

from the very beginning AI has been heavily funded by defense agencies.

And, of course, we are talking 40's World War has just happened and obviously defenses,

you know, in on minds of everybody.

So they would create use the, this computer for artillery firing tables and it was said

at the time that it could do things it could compute a trajectory of a projectile in 30

seconds.

We are a human computer and by the way, we do not hear about the word human computer

now, have you heard this word human computer?

You have there was a time when they were there was a job called human computer.

There is a beautiful movie about NASA and three women, does somebody remember the name

of the movie?

Nobody remembers, I am sure somebody has seen it.

It is about three African American women who are employed in NASA as human computers or

at least as some some form of computers.

So basically, their job was and we will find a name if somebody remembers or can google

it, let me know.

So we there, their job was to do computation.

They would compute the these projectiles and they may take about 10 hours to compute one

projectile and this thing can do the ENIAC could do the same thing in 30 seconds, right?

So notice that right from there, and offcourse, now human computers are is not a job and not

surprising.

There is also a beautiful story about a woman who really got a bunch of these women together

and train them as human computers in a time where these women did not have much to eat

and so on that is a different, beautiful book.

But anyway, so that was then Right and now that this was the time when computers were

coming into the fold, right it was a reality, not just a figment of somebody is imagination.

They did not ask the question, can I use computers for world processing or can I use computers

for excel like spreadsheet management, or maintaining my database?

In 1950, Alan Turing, who is considered the father of computer science, you all about

Alan Turing right some of you may have seen The Imitation Game Yeah.

You must see that movie beautiful movie, he said in his seminal paper that I propose to

consider the question can machines think?

That is why we call him the Father, because that particular question that he asked in

1950 when there is one big computers or you know, some of these computers have just started

coming out and the only thing they are doing is computing the trajectory of projectiles

and so on so forth.

He is starting to think as far into the future as possible and being somewhat of a luminary

and writing these seminal papers thinking about these philosophical questions, and he

asked the question at Okay, let us say a machine can think, how would I know that it thinks,

I think about this question for a minute.

Suppose I claim that this podium which is made of wood thinks has the ability to think.

Without casting any judgment on what wood is made often is I mean, what is wood and

whether we have any prior on whether this is living or not living, let is not worry

about this.

Let us say I make this claim that this podium things how would you and I even assess whether

this podium things or not?

The Let us take another example, whenever they asked such philosophical questions in

AI, we relate them back to our life, suppose I have to ask the question, does my friend

think, what would I do?

I would ask them a question.

I would ask them a question which I think my friend should give a correct answer for.

If I want to believe that this person really thinks this is a good question, it is an intelligent

question.

If I ask an intelligent question, and my friend is able to give me an answer, then I would

make the claim this, my friend things suppose my friend could not speak what would you do?

Written answers, suppose my friend did not have any hands Sorry, another person could

write for them how they cannot speak, they do not have hands?

Create a situation that they have to make a decision and decision would be seen by some

movement of hand, blinking of eye in a movement of arm moving of legs, whatever, right?

Suppose they did not even have an ear and an eye.

Now I know why we do not want to visualize such a person and I it is, it is becoming

really very gloomy at this point.

But think of it as a thought experiment.

This is probably what Alan Turing also thought about at the time I do not know, hopefully.

Suppose I have a person who cannot see who cannot we observe the environment?

Who cannot make a difference to the environment so, does not have hands does not have voice

does not have eyes to blink does not have any motors has a brain inside is it possible

that this friend, this persons not friend this person still things, but we will have

no idea about it, is it possible?

Come on at least not one way or the other.

Yes, it is quite possible that and you may have seen and you sometimes I met a person,

you know who had Parkinson's disease, but the Parkinson's disease had, you know, gone

to a certain extent that there was full paralysis and so, the person was not able to communicate

very much and we did not know very well, whether the person is able to observe the discussion

or not, whether the person is able to understand what is going on around their surroundings

or not this is even possible in a human in todays world.

It is possible that their brain was working perfectly they just had no way to communicate

it is possible that their brain was not working at all and we just have no idea.

So, the point is not working at all means not working in terms of the understanding

perception of the man.

So, the point that Alan Turing came back with is that it is possible that this podium things

we have no way to know because the podium has no way to communicate with us.

So, you will overtime see that communication with the world by listening producing an output

not listening by ear necessarily by observing or by looking at the environment perceiving

the moment and then coming out producing an output is critically important for us to judge

whether there is an AI system sitting in there or not Okay, because the brain is meaningless

in absence of the communicators in the body, right and there for he said that the only

way we can figure out data machine things is by asking it questions or creating a scenario

where we can judge its decisions.

And he devised the very famous during test, right?

I am happy many of you know about it, it has become part of, you know, folklore of AI and

folklore of computer science and we will talk more about it at the time a little later,

but he said that the Turing test would be a good test to figure out whether machine

things are not.

Notice by that time the field of AI is still not born, this is still his thought experiment

and he is written this paper, now offcourse, this idea starts to pick up.

And in 1956, four of these people come together to do a two month workshop at the Dartmouth

College in 1956.

These four people are called the founding fathers of artificial intelligence.

They say in the project proposal, we propose a two month 10 man study of artificial intelligence.

So this is the first time the word artificial intelligence starts to get used and it was

given by the very first person on the right John McCarthy, the John was at Stanford.

He worked a lot on chess and he also created or co created the language called Lisp, which

is a declarative programming language, which became very popular in the 70's and 60's.

They say that in two months, we will solve artificial intelligence and they meet together

at Dartmouth College over the summer, and that is where the field was officially born.

Offcourse, they had to spend more than two months for AI to be solved and we have still

not solved it.

So therefore, there is one thing that I feel when I do think about most AI researchers

is that they are highly optimists, heavy optimists.

They feel that we can do something much before anybody in the world is ready to achieve it

and they of course, fail, but that is okay.

They move the field forward and these were the full optimists who thought that they can

solve most of AI in two months, this job, but that is somewhat true also let us talk

about the other three people.

Alan Newell and Herbert Simon, are in the middle they are they were from CMU, they were

working on logic theorists, they are trying to figure out, can we some how create theory

of how to prove theorems?

Can we create an AI system which will be able to prove theorems automatically and lastly,

at the bottom there is Marvin Minsky from MIT who has many contributions including he

built one of the first neural network learning machine and he is my great great grandfather,

at least Academy grandfather.

So my advisors, advisors, advisors advisor is Marvin Minsky, right and he was the person

who started the MIT AI laboratory, so at the time, these four people from Stanford, CMU

and MIT came together and officially started the field.

So now what happened?

Initially, of course, there was enthusiasm people were coming together, you know, trying

to solve these problems.

So if you look at the history, I mean, there are many, many things going on but at the

highest level, you know, Turing tests came in 50's AI itself was born in 56 in 64 so

by the way, language and language communication, all that also was part of AI and Eliza was

the first chatbot way back in 1964.

By the way, we have now we live in a world where chat bots have become extremely important.

We all have chat bots to play with and you know, spend time with kill time with and so

on.

The first chatbot was in 1964 and believe it or not, and it is a funny story, where

somebody on the private messaging system instead of having your own response put Eliza in and

one person started communicating with Eliza and started getting the Eliza started asking

personal questions and this person started answering personal questions and this person

felt like there was a real psychotherapist setting on the other side.

When we when you come to my NLP class, we will talk more about Eliza and how to make

chatbot not in this course, but this happened way back in the 60's.

The first general purpose mobile robot and you can see a picture on the right is shakey

you should google shakey you should learn about it and So on.

Shakey was an amazing project this was this robot which had many many sensors and so on,

it had a camera control unit it had a television camera, it had a rangefinder it had wheels

and so on and at that time, people in Stanford, were using shakey as a platform to do interesting

AI and AI remember in 2016 there was a 50 year celebration of shakey at a AAA conference.

A lot of the people who worked on shakey at the time who was still alive, they came together

and started narrating very funny stories about shakey and so on.

It was really amazing to hear, you know, 50 years down the line shakey was still relevant,

because in the process, they developed one of the most important algorithms of AI called

the a star algorithm and we will talk about a star algorithm and hopefully week three,

you will see that you know what happened 50 years ago, 52 years 53 years ago, was still

extremely important and being used in many, many places including possibly assignment

one.

Then things started to go bad, they started to go down south and since we are in the US

summer is a great word spring is a great word, but winter is a bad word.

So you should think of a winter not from an Indian point of view where you are like wow,

we have winters now, you should think of it from the US point of view where Oh damn, we

have winters now it is going to snow, it is going to be negative minus 20 degrees Celsius

or whatnot I cannot go out my eyeballs are going to freeze and so on so forth.

Right

You have seen those videos where you know you put water in the in the air and the water

freezes mid year we are talking about winters and they were winters and they were not one

but they were two AI winters that came over time.

In 1974 to 80 there was one winter where people were working in language started to feel that

Oh, man, this is not working out.

There was one seminal report which said machine translation is very hard it is not going to

succeed.

A lot of funding in machine translation at the time came from defense agencies, can you

guess which language did they care for at the time?

English Russian, we are in the cold war time so of course all the effort will going into

making sure that we can understand any things that we can intercept in Russian language

right and they said that oh man, machine translation is very hard neural network had come in, but

Marvin Minsky wrote a paper saying that neural network is very hard it cannot even do XOR

function, you know, XOR function from a logic, I hope.

And similarly, there was a report where there was some frustration on speech understanding,

they were able to show that if you speak words in a certain order, you are able to pass it,

I understand it, but otherwise the system cannot do anything and by the way, all these

problems remain unsolved are mostly unsolved until very recently, and they are still not

completely solved, but we have made substantial progress in the last year.

So because of all this, suddenly we were in a place where there was so much excitement

about a earlier and it died down.

Then it started coming back up and the LISP was working there was something called the

experts systems era and even now, some people who are still stuck in the 90's, who teach

the old style AI course, they will teach expert systems we do not talk about expert systems

in this course anymore because those this is what thing from the past.

But the idea of expert system was somebody uses a logical language to encode all knowledge

about their field.

Like if you want to make an automated doctor, you will say that this disease causes these

symptoms this symptom, this disease can be you can be treated by this medicine, etc,

etc.

So, put everything in a logical framework and then given the symptoms, you do some inference

to figure out what disease you may have, what treatment should be given, that was an expert

system.

At the time, they started making specialized hardware for expert system and it became a

half billion dollar industry.

But within a year, all that half billion dollar industry was wiped out, because new machines

came out which were much better and they were general purpose and so on, so forth, and all

this industry died out.

But more importantly, what died out the reason why I went through these winters is the because

AI researchers were diehard optimists, not cautious optimists or realists.

They started saying from the very beginning we are going to solve intelligence, we will

make a machine that will be more intelligence than you, we will defeat you in chess, we

will solve theorems, we will communicate in language, we will understand your speech,

we will do machine translation, they kept making all these claims again and again and

again and again and again and offcourse, defense agencies will do nothing they said, Oh, we

will give you a million dollars, we will give you $2 million.

What happened?

What came out?

What came out?

Nothing I would not sell the money down the drain.

But from the perspective of a person who wants to use the technology, they will feel that

the money is down the drain because nothing substantial, which is workable, has come out.

So AI started getting bad name to the extent that people thought that Oh, AI folks, the

other thing you have to realize is that they were also computer scientists, to the theoreticians

mostly who really cared about proving things.

They were like, give me a problem, I will figure out its complexity class, give me a

problem, I will create an algorithm and prove things about it.

On the other hand this AI folks why interested in working demonstrations?

They will proof, but they were interested in working demonstrations more than the proofs.

Because of which the computer scientists theoretical computer scientists will say, you guys are

not formal enough you guys are not mathematically rigorous enough.

On the other hand, the applications people will say, you guys do not do your guys solutions

do not work.

So they were sandwiched in the middle neither the applications people are happy that you

know you can achieve anything that works nor that theratetises are happy where you can

achieve something you can prove anything about.

So even in the computer science world AI started getting bad name, it started being looked

at as set of people who have these heuristic techniques, which sometimes work.

But have no theoretical basis so therefore, I remember and by the way, some peoples remain

stuck in time, right for example, after being in it for four years, if you go home, your

mom will still think you are a 17 year old.

They would not she would not know that you are 21 now and you have moved on and you have

become a different person.

You know, I remember a beautiful dog, my shamba Nagel, who said that my mom still thinks I

am 14 year old, I am not getting you he said that explicitly.

We remain stuck in time you go to look at your friends who are in the US now they are

notion of India is the notion that was when they left India.

If you go talk to uncles and Aunties who left India 30 years ago, their notion of India

no longer matches with any notion of India today.

But they are stuck in time so we are always stuck in time.

So they were people, even in this department and in India in general, who was stuck in

time they thought that AI has not moved because they will get their PhD in the 90's.

That is where AI was not a very good word, AI became a bad word actually.

In fact, they were people who cut their cords from AI they were topics which were part of

AI and they created their own sub communities and started saying we are not AI we are computer

vision, we are not AI we are natural language processing, we are not AI via machine learning

and I lived through the time therefore I can relate to it I remember that in 2000s when

I was doing my PhD, AAAI, there were five communities which will still say we are part

of AAA five is just a number 3, 4 7 something Like that.

There were many such communities which will only go to their local conferences when will

never come back to AI conference because they thought is AI bad word let us stay away from

it.

Let us say we are whatever we are right so, when I would come back to in IIT Delhi, I

was a student here undergrad when I came back as I was still doing PhD at the time, I told

some of my ex professor, I mean professors not ex professors now, my colleagues today

that I am doing AI they will say, oh, you are doing AI okay.

They will give me the sheepish smile I do not know why they are giving me a smile.

But it was clear that it is not a smile of appreciation and then I have a one person

who I would not name who always speaks his mind thankfully, so he said, you are doing

AI what are you doing in AI?

So I told him, I am working on this Markov decision process model oh, so sa,y you are

doing operations research.

Why did you say you are doing AI now for people who do not know, a model came from operations

research in the 50's.

But you know, over time it got adopted by AI and it is the something we will study later

in the course.

So it is very confused what is going on and then it dawned on me, that to them AI still

the AI of the 80's and 90's.

When we were having a winter and AI had become a bad word.

So this had lasting effects, because and this is an economist 2007 node, which says AI is

associated with systems that have all too often fail to live up to their promises and

in this world, over time AI has now come out.

We will talk about the present in the next class, but this graphic shows you the number

of attendees for several AI conferences and at the 80's it was only AAAI, if you see and

they were almost 5,000 attendees if you see here, they were almost 5000 attendees at HI

conference in 1985 and it went down so rapidly that over time the it was less than 2000,

maybe 50,100 people started coming I remember a time in AAAI had less than 1000 attendees

and we were very worried that we will maybe even have to close the conference we cannot

we cannot fund it anymore.

From there, we have now reached a point where all conferences are starting to see a jump

in the attendees and offcourse, the conference that has one the most in this is new NIPS

or NIPS.

NIPS is the old name now its new NIPS, right Neural Information Processing systems and

notice that this is not even the current figure.

So I remember that for NIPS one year there were 2000 attendees next year, there were

4000 attendees and next year, there were 8000 attendees.

To the point that the year after that the general registration closed down in less than

12 minutes.

But think about World Cup cricket tickets even for India, Pakistan match you will probably

get more than 12 minutes to buy the tickets.

This is new NIPS technical conference we are talking about believe it or not, if you really

want to go to new NIPS you have to buy it in the first minute of registration opening

or you should have a paper in there so, those are the two possibilities.

So AI It is said that Yann Le Cunn is a celebrated in China, Yann Le Cunn wrote on his facebook

page that he was walking in the streets of China and random people started coming to

him and started saying oh, you are Professor Yann Le Cunn so nice to meet you or things

like that, You know this only happens to people like Shahrukh Khan and Vidya Balan in India

I have never felt that I feel so sad about it someday and this is before Yann Le Cunn

during the award I want to say.