Well come in everybody hear me
OK All I'm OK to go ahead think.

OK this is it's my real
pleasure to introduce

my colleague my boss my
classmate turns out Lance and

I go way back to the spring
of one nine hundred ninety S.

just we didn't know each other so
I don't know if it counts as way back but

we were in a class together
taught by my Ph D.

advisor John Spencer at MIT
in have the lecture notes

to prove it but
it's about Lance my introduction is

not going to end in six tenths
of a second sorry I figured he

deserves quite a good solid introduction
so Lance is professor and chair of

School of Computer Science at the Georges
in storage Institute of Technology as you

all know is research isn't the theory
of computing focusing on competition of

complexity in a petitions other topics
such as Game Theory and micro economics.

He got his Ph D.

a while ago a minute given any
dates working with Michael sips or

before he joined Georgia Tech
in two thousand and

twelve Lance was a professor
at Northwestern and

before that at University of Chicago
he founded in two thousand and two and

quarter's to this day the competition
complex to blog the first major

terror computer science book is popular
science book The Golden Ticket P N P And

the search for the impossible is
the loosely based on his widely down or

article from the communications
of the A.C.M..

The P.N.P. problem is arguably the most
important open problem computer science

has perplexed irritations mathematicians
included for several decades

simply stated it asked whether every
problem who solutions can be who solution

can be quickly checked by a computer can
also be quickly solved by a computer.

Last might speak a little
more on that later.

Is but the golden ticket provides
a nontechnical an induction to this P.

and P.

question its rich history and
its other the mic implications for

everything we do with computers and
beyond Lance has worked

in some barriers of competition complexity
theory such as interacting proves our

calls Kolmogorov complexity
is giving a guest lecture

in my information theory class tomorrow
on Kolmogorov Complexity thank you Lance.

Working with his former student Carsten
learned and colleagues Howard Karloff and

Norm listen Lance developed an algebraic
technique show that the so-called

parliament of a matrix.

Some of you might have heard of
the determinant of a matrix dominant

is a similar quality much
much harder to compute.

Than surety as a prominent has upon
a meal time interactive proof system

which implied that the polynomial
time hierarchy has parliament time

interactive proves let me cut to.

The proof techniques they
use was very useful and

was used by ideation
mere Here is the ass of.

A cryptosystem many of
you might have heard of

to establish that the complexity
classes IP for interactive proofs and

peace space for problems that can be
solved in polynomial space are the same

his work on a tractor proof system
sometimes the base lower bounds for

satisfy bill to lead to his
election in two thousand and

seven a fellow of the CME Association for
Computing Machinery In addition he was and

then a soft presidential faculty
fellow and a Fulbright scholar and

he has served the professional
organizations A.C.M. and

I triple in many capacities of the years
and Kerr currently sits on the computing

research assertion board of
directors maybe not anymore but

today he is going to talk about
the title you see up there thank you.

So thanks.

I'm not going to talk about
anything never thought of that.

Don't worry didn't follow any of that.

So this thought kind of came about.

When starting to work on a second book you
know basically with the side and I was at

some point talking to Paul Pogo Bart about
this and he said wow that's a great idea

you should give a talk about it and we can
tie it in with the seven bridges like.

And the advisory board meetings so

here I am so thanks thanks Paul and
will miss you in general when you.

Go to the.

Republic of Texas or.

Did you just become a US citizen and
I are like live in a country I am the.

Stars I want to thank and
this is this is going to get the this

talks going to kind of give a very
broad overview of computer science and.

Another specialist in all computer
science I'm a theorist so

what they want to think some of the
factory that has a great discussions with

about some of these topics.

Stuff on Mars you true.

And.

Me Loesch prove all of it.

OK so.

So this this this proper came up.

Actually it was going to also so
when I was making the slides this happens.

Mark Zuckerberg was grilled by
senators and representatives because.

You know information now was taken
from people on Facebook was used by

people trying new manipulate
the election and so

and then he started really
hits you that this this field.

Computer science it's really
gotten to this point where

where we have things like this happening.

And so you know part of you know
might I'm not going to go too much

into the the politics in that fix and
stuff a little bit but but

I really wanted to kind of the idea of
this talk is kind of show you that you

noted what actually happens you know and
all this stuff that's made this possible

is a lot of great ideas and
thoughts from computing and

then kind of need to kind of go through
some of them you know as much as I can I

mean six tenths of a second is pretty
quick when even an hour is not really

enough to do everything justice so
this came up it started when I was as

mention those a professor Northwestern two
thousand and eleven and we were trying to

create a computer science division
within the E.C.S. department and

then a question came up which was
you know what is computer science.

So and this is a question made out of the
mess we ask ourselves a lot you know what

is computer science you know is
that a science is there generic

is it both is it something different to
kind of view it is something different

it's kind of hard to explain and so I was
trying to think of OK how do you explain

what computer science is and we're first
going to tell you what it isn't it isn't

programming OK with you know that's this
kind of like when you talk to people and

they say of computer science it's
just you know writing little code but

that's really not what computer
science is any more than

you know mathematics is about solving
different you know taking derivatives of.

Integrals and it's really really know
much much more interesting than that.

So what do you do well why don't
we just Google it and say.

OK what is computer science Google what
is it what is computer science and

you know the answers you get
are kind of a little bit.

Not that useful Google's
answer is the study in

study of principles the use
of computers and we keep.

And.
Science is the study of theory expert

Tahsin and engineering or form the basis
for the design and use a computer.

So it's not you know these kind first
of all it's it's is fixed focus

on the computers so we know what
is that you know and also you know

computer science isn't really just about
computers I mean it's really about a lot

it's about of much broader thing about the
new information and how we deal with it.

But then I you know one thing I noticed
you know is this thing is that you

know we typed this question in we got this
answer in zero point five six seconds and

that's where the tower talking from
point six seconds what happened

I mean you really you know you're used
to it now write something in the Google

you get an answer back and and
it happens relatively quickly and

but in you know really think about
what's going on when you do this

what what what's happening
in when you really and

I start thinking about it realized that's
what computer sciences it's all the ideas

that made Google be able to give you
an answer in six tenths of a second.

And so when you get in the Tiffany
like this what is a good

academic like me do about it we don't.

So this is actually I dug up my old tweet
from of February two thousand and eleven

asked what is computer science answered
everything happens when you ask a question

to Google until you get a result I love
the one reply I got five years later.

Picked over one thousand and
sixteen you know really simple and

that's why it's still the best
interview question so

if you're interviewing at Google and
you get this question if.

You are whatever and they would write
a soft You don't say Google but.

That's it so
why does my talk is going to be

a very over in no review what
a small taste in really.

To me a lot but

it's going to be a small taste of what
happens in zero six tenths of a second so

first you know let me we have of all we've
got to find out now where do you start so

I want to do a little bit history before
we get there so you can start a lot of

different places are going to start in
one thousand forty seven with the first

transistor a transistor is kind of the
basic building block block of computing.

This is.
Simple you can you can do a few things of

it logic operations do a little bit
of memory things like that and.

So nine hundred forty seven
it was you know they finally

Bell Labs created this thing
called the transistor.

And then And then in.

Nine hundred sixty five this is
a Gordon Moore is a cofounder of Intel and

he came in sixty five he made
this projection he said.

Every year and a half to two years
the number of transistors we can

put on a chip on a computer chip we can
double that every every every year and

a half to two years and

you know this this Moore's Law
actually became more like a mantra and

so people were actually able to do this
and the some extent are still doing it.

So in one nine hundred seventy one this
was the year of the first microprocessor

Intel four thousand and
four at about three thousand transistors

in two thousand and
five this is the Pentium D.

has about.

Two hundred thirty million transistors.

Current ships of about three
billion transistors but

I take two thousand five for
a reason because.

When you put more transistors on a chip
you make them smaller and smaller so

you can you can you can change the voltage
quicker so it makes the computers

faster but that's kind of stopped
in two thousand and five because

it's good to start generating too much
heat it's basic answer if you try to make

it faster the chip might melt and you can
keep it cool things like that so we really

had to limit computing really went through
this kind of change around the mid.

To part of the you know around
two thousand and five give or

take a couple years and it really
started to see this transformation

in the transformation first that led
to what's called multi-core computing

which is put in multiple
processors on a chip so

this penny ante that actually was a dual
core has the two processes on the chip but

it also led to putting a lot of computers
in one place so which led to be.

A big data centers and

what we what the normal in that when you
think of cloud the cloud this is basically

what the cloud is big big data
centers spread around the world for.

A typical data center has about fifty
to one hundred thousand computers.

And now.

And and then it is this a lot
of things started to happen

with this I mean Amazon created some data
centers to sell some of their stuff and

then they realized they had.

Access capacity so they decided to sell
it rent their computers to other two

to other companies and that turned
out to be an incredible thing so

now you could start you could be a
business you don't have to create your own

computing environment you know just
a couple people could Ransom's some.

Computing power at Amazon and
that that just created this huge boom.

Of.

Moving computing you know just basically
into the cloud services and now today.

Mostly computing is done on the cloud you
notice if you do use a computer these days

you do almost everything through either
a browser or nap on the phone and

that's because really all it all the
browser is doing the showing your stuff

with all the computing going
to the cloud so it's B.

and that's become a big transformation.

And that's kind of where our story.

Really begins.

So.

The first thing is kind of interesting is.

Is.

That there's a lot out there as a color
while analogies between the cloud and

the chip so this is something that
I mean as proof I've pointed out

is that if you think about what
they're both doing a very similar and

they were both in case putting a lot of
computing power in different pieces but

there are also issues of cooling and

power and turning on power certain
areas and often certain areas that at

different times to make these things work
is officially as possible I mean there's

different extremes you know obviously that
is a lot more power than the chip but

because of these things you have to you
want to put your data centers near cheap

power and where it's easy to cool things
like your day amahs is a good example.

But this is kind of an idea that comes up
a lot in computing you take ideas from one

domain like making computer chips and then
you take those ideas those very same ideas

and you apply them to two to something
like how you design data center.

And just.

A minute I'm an interest in approach here
but what really came out of the cloud

was you know a lot of information
goes through the cloud and so

what that did up doing
was creating a lot of.

Data.

So and in the end you know.

And did and that is really sort of what

been you know that it will often we
say you know it's all about the data.

This this is what's really powering
things that are happening now so

that are started.

In the very structured way so.

You know I.B.M.
got its I would say it start but

it really became a big company
when it got the contract to do

they had no source of the curate even when
I wrote about started in the thirty's and

to go there wasn't really doing
computing the times mostly tabulating.

But there was already kind of where you
know computing was coming in to deal with

big data and our data originally was
very structured you had very things very

well ordered in certain columns and rows.

And then this wasn't and
then you really needed but

then you would have a lot of databases
I mean a lot of of of information so.

We developed things like called relational
databases which lets you connect

different data but having that we know
maybe like you want to know our student

to take in a certain
class of a certain time

you can kind of put all those things
together is a lot of very powerful tools

that went into developing these
kind of relational databases.

And so that was this was a big you know
things that happened in the seventies and

eighties for example.

And I want to have that then
you can do things like.

Book airline tickets or

buy things on Amazon like my book or
you by the way book signing after that.

But.

But the data today is different it's
not this really well structured

that all nicely placed in
well connected it's messy

it's social media posts it's
videos it's e-mails and

other text it's cars that
are you know it has these cars

drive around map in it taking pictures of
everything and mapping everything in it.

For a little baby's
forced to wear bow ties.

That may be right there.

The.

Internet or things you know said
to is in your in your house

measuring everything that's happening.

So it's no longer all this data
kind of nice and clean and

it's a how do you deal with data this
been a big thing we got all this data but

now it's not so nice anymore how
do we deal with with with that so.

First example want to talk to
you about is just web pages.

So the Web kind of started in the early
ninety's and you know we're actually had.

I mean the Internet goes back
to the sixty's actually but but

in the early ninety's when we
really started having web pages and

in the early days was actually pretty hard
to find a way to it if you wanted and

you had to search we had good
engines that could crawl and

give you millions of examples
of what you wanted but.

He said that this test that
would search for Holiday Inn and

it would give me like a Holiday Inn and
Buffalo New York and

I want to give me the how the in website
just give me some random holiday in.

And then these two students in

Stanford had this kind of interesting
idea we started a Web page and just.

Start clicking on links at random and
that takes you to another web page that

takes you around and if you look at the
web page that happens most often that's

the web page that year that is
going to be the most important and

you know that sounds like
a kind of a crazy Snit do but

there's a way to model this mathematically
using nice tricks when you're outre and.

And to not to be a really incredibly
useful way this actually works and

you know it's written as a as.

You know it's a show real
research paper written by

it's used in its name search a Brin and
Larry Page and

now you know this is became a seven
hundred billion dollars company so

that's it just one you know clever idea
can really drive a lot of things but

even web pages have a lot of structure to
it you know links that link to each other

what about like pictures for example
videos so I like this one you know we got.

The clever way that's how
they made money right.

And I actually like Amazon you know
what the damage done is product and

it's not it's not selling
stuff it's the cloud and

that's where the make all
the money which is an artist

a little fuzzy when you blow it up but
he way these are dogs and muffins.

And I love this feature what.

Can you tell which a dog or a germ up and.

So how do you tell with dogs and often.

How to teach a computer how
the tell the doll itself

how do we tell the difference
from the dog's in a muffins and.

So you know actually we this is
an idea actually came from us

humans you know we have these
neurons in our brain and neurons.

Basically neurons fire and

have enough things fire into another
neuron little fire and no feeling fire all

around your brain until you get this
notion of either dog or muffin so.

I don't know are not a neurology and an
hour an hour and a computer scientist OK.

So I'll give you the computer science
version OK so they so people develop.

What are called the deep deep learning
known or neural networks as we say and

the idea here is you take a picture
of a dog and you feed it in a sickly

you know one bit for each pixel of the
picture where it's really what's going on.

Around.

And then this network has different
weights of different wires and

then the you think of these neurons is
either firing or not firing the various

functions I mean what simple find a bit
and out at the output it should say

dog and if you feed a number off and
at the output it says muffin.

And so you do it what you do is
you run a bunch of examples and

it's not going to get it right and
then when it gets it wrong you know you

adjust the weights and
there's a nice nice ways to be.

Just the way it's quickly so

they call back propagation and use partial
derivatives and all these things but.

Great into centers is that short term and
and

it turns out that you can actually
use this idea especially with modern

technology to make this thing work really
quickly and this learning thinking works

extremely well I mean it's amazing
how well that you can do like.

That you can actually
recognize pictures this way.

And then this is one again one of the kind
of neat things that I like about computer

science is that this neural net idea
you know works for computer vision but

it works for so many it's really
the basically the same techniques that it

works in so many different domains so
you know it's a separate talks muffins but

it also machine translation
is now mostly done

through the machine learning models
these people are new models for

it recognition I mean if you were to tell
I mean you can talk to an Alexa I mean my

my goal is a Google home and it
understands me better than most humans do.

And a voice recognition I would never
thought would be basically a solved

problem by now.

We use in a two to diagnose patients
either through scan through symptoms.

Recently we've seen in these last
couple years we've seen neural nets

trained to play games like go in chess and
without it basically all we

do is tell the tell the computer what
the rules are it plays against itself or

uses these these learning techniques
in a you know clever ways and

it's now become the best way
the best computer get to see humans

it's the best computer programs the play
chess and go are now these things that.

As well as not be on the blade any human.

Just from learning on it's own.

So in some sense if humans try to get
involved you actually just kind of

get away.

And driving cars such a cars kind of goes

in there's a lesson though here in
that these things are great but

they're not perfect so and
they're not perfect in a game of chess OK

maybe a loser is a game that's not perfect
itself driving cars people can get hurt or

killed and that as happened
actually in the last couple weeks.

So it's it's I think people sometimes get
used to these things working really really

well but you really have to keep
in mind that in some instances

it can be very dangerous
on the other hand.

You know computers don't they don't
get tired and get drunk they don't get

distracted so so
I mean it also saves lives so

though it's much harder to quantify
You know when lives are saved.

But is something that we
have to keep you can mind.

So now to do deep learning because deep
learning and all the learning techniques

to become so powerful it's almost like
reversing Now how do we make computers

work so well they're trying to do
learning and dealing with big data so.

We go back to data center where you've got
lots of computers you can make them faster

but you can make them work together so
there's this nice idea called Map Reduce

you basically you take it take something
you want to process you know the dog and

you spread it out through many
different computers and this is called

the mapping is happy do so you map it out
it does a little bit of processing and

then you collect it back in to go machine
and you process it this is called

This is neat idea called nappy do since
been used I just learned a lot of other.

Ideas when you really have to process
a lot of information spread throughout.

And now more recently vaccine
moved to different models where

you stick this stuff in
memory now as opposed to

spread around different
computers because that way.

You can actually process things a lot
faster something like call the like spark

example where you actually
do this within memory but

the other thing that we've changed is
actually seen changes in hardware so

this is called a tensor T.P.O.
a tensor processing unit is something that

Google developed and it's its main purpose
this is a computer chip but it what it

does of process is tensors but we're
basically what's used for machine learning

is that's where a lot of machine learning
comes from from from processing in.

To the sensor processing unit is is
something that that you now use to plug in

and and
accelerate machine learning applications

right there in the hardware and
that's become an hour another recent trend

where you know computer use the we're
used to the programming here and

then you build the operating system here
needed hardware here but now it's all kind

of it all kind of happened and now you're
programming the hardware you know people

created the Feiss that just all they
do is specialized minded coins for

example in other big thing that
people try to do to make money.

And even just read in that.

Facebook just now just today and when
it's like if I talk changes every day.

That they're that they're going
to create or all the chips

that the machine learning and other of
the things they need them to do so and

then there's something called F.P.G.A.
is rehashes program directly the chip

I mean it's become a big it's kind of
interesting you take the program and

ideas now you have to apply and
directly to the hardware.

OK so now you have all this things that
happen on a computer you know it it does

all these things and it does all these
tricks to understand what the question is

but a machine learning a lot as you know
searching through the internet doing

the Page Rank out with them you know
where to find the rand things but

how does it get that it how does
it get that information to you so

you know it's happening in
the cloud you're sitting here

how do you get this information
going back and forth and.

Well we use something you might
have heard of called the Internet.

It's of the Internet here but I kind
of want to talk a little bit about you

to the network itself
which I always which is.

Kind of spend a lot of time learning
about last summer a little kind of

the first chapter I'm trying to write for
this book.

And he has that kind of comes in layers so
I think he is really kind of view of

the Internet at various different
levels of granularity so at the top

we have what's called the application
layer and it's basically you know

the browser you're using is communicating
with some company like Google.

The A with the very limited things or
you should think of it maybe like.

Alexa you're talking to
communicating with Amazon and

Amazon has got color buildings
that's why we want them in Atlanta.

And then below the application layer we
have something called the transport layer

transport layer actually takes the process
using nets like this example here

takes the process Netflix like
aptitudes in on your phone connecting

with the you know some Netflix and
in that this is actually

Netflix is actually rent space
on Amazon servers which always

amazes me because they're a competitor
with kind of the video but Netflix doesn't

want to build their own cloud so they
just rent space members on the cloud and.

How all of it all works so

you're actually going into the into
into some program that's running on.

Some machine and Amazon and you're
connecting your process with that process

and so the movies can stream to you
can watch them on your phones or

on your T.V. and
then below the transport layer is you know

connecting the two machines themselves so
that the transfer layer protocol is

the live nice way accounting application
is like you know when you see this.

H.T.T.P. when you type it in I protect
stands for protocol or something like

that that's where the application or the
transfer layer is something called T.C.P.

transfer control protocol never is later
something called the IP the Internet

protocol and
that's where you get these IP addresses.

So you what you want to do is
have a different number for

every computer trouble is you ran out
of numbers so you create more numbers.

And so now you know because.

We're actually trying to get back to
where we can figure these layers again.

Network layer.

Now in the network layer I think
something that I think is to me is.

What I call the most interesting statement
I've heard about the Internet which is

the Internet works so well because it
doesn't have to it's one of my favorite

quotes Now what does this mean well when
you make make a phone call the phone

company makes a connection no I'm in
the old days of physical connection but

later on the logic and kind of promise
is that you'll have this good connection

with someone on the other side you
can have this conversation back and

forth it would have a guarantee
that you'll have a connection

when you send a letter in a send
a letter the post office is

basically guarantee now it's going to
get you but doesn't always work for

them that they're promising to
stick a letter in the mailbox and

it will get to its intended recipient so
for.

For the network layer for the computer and
we don't do that we don't make any

promises into the protocol you know
a trough is you know it's like a best

a best effort kind of thing now in a tries
to get the information that after a while

if your information gets lost or
keep seem to go around in a circle.

Or whatever it does is say OK
give up I just drop it and

it goes away I mean it is I mean
just give up trying to send it and

you're saying well how could how does the
Internet going to work well if you can't

you can't even trust your good
information point to point B.

but that's you know that's where
the transport layer came in and.

The transport layer takes counted that
make sure that the things you know

in the in the network layer things
that come not misc at all are coming

are all in order to break things low
pieces and they might come in order.

The transport layer works to try to
put everything in the right order and

make things work.

But it's kind of me to
these different layers and

the idea that makes this work is
that you really this network layer.

Is that work layer really doesn't have
to succeed at all they can just try and

see what happens.

And then below the network layer
is something called the link where

the link layers because you're not
actually connected you know net and

your phone is not directly connected
to the cloud it's going through.

It's going through some cell tower
maybe your wife I would just make this

some router which is connected to one
thousand nine hundred four a sense of use

eight hundred eighty or whatever company
use that might go through another Wow And

that goes into the cloud and then you know
it goes to some machine in the cloud and

oddly enough there's probably a lot more
layers than you're you're seeing here in

the process so the link layer is just the
kind of protocols to get if one machine to

another in a kind of
just forwards it along.

And then you have these different of
course different protocols like your

wife I would be one protocol and three G.

would be another protocol in
either net which connects you know

things that are that are directly
connected is another protocol and

somehow get these things
from one end to the other.

And that's all the layers I don't
talk about that are there or

something underneath the called
physical layer you know how the actual

bits actually move but
we believe that the let you all engineers.

This is a computer science.

But you know when you study link layers
you know it's how things are connected and

what else have we seen recently
some of us connected is

is that to the seven bridges so.

This is I took this picture a couple
days ago from the fifth floor of

the Mason building a civil engineering
building this is between Mason and and

how we of you haven't seen it yet and
it's basically you should think of.

The you should think of it
because this sidewalk as a river.

And then these is kind of land masses and
these things are seven bridges and

the question is is can you walk
across every bridge exactly once.

So.
If you haven't seen this thing a lot of

your is in it but if you haven't seen it
please please go check it out it's kind of

cool but it's based on an actual town or

what used town called a con and
spurred in that that that

was in Prussia So it turned out what you
speak all Conan's bird now and Russia.

Now called Kalin grad in Russia.

And it had these seven bridges in fact if
you if you think about it this diagram and

this diagram are basically the same.

And they were actually wondering it was
a good game you know could you cross every

bridge exactly once and
then in seven hundred thirty six

Leonard Euler came in actually

a mathematician who came in to strike this
problem into something called a graph and

then showed in fact that you can't do it
and only did that you can't do it but

you gave an exact characterization
of exactly when you could do it and

so and that started an area
called Graph Theory and

graph algorithms that plays all a big
role in helping us do things you know for

a lot lots lots of applications
of graph theory but

particular helps us
design better networks.

And also even analyze the Internet and
see the pieces in our connects and

what areas are important and so
understanding graphs can really

help us understand anything that's kind of
connected in relationships relationships

like there's a relationship say in
Facebook how are things connected and

how your friends are you connected in all
these things are are very well tied in

the graph theory so it's and that's kind
of anything that came out of this bridges

which is you know what inspired you
know we need to get this thought today.

On talk a little bit about.

You know so many it when
transformation you don't know we

have to I want to look at about
cryptography and why this is important

because I'm a Suppose you're connecting to
your bank and you're sending a password

Well what happens if there's
some bad guy listen then maybe

if we know why five you're doing this
from you know from a coffee house.

And and.

So in order to deal with this somehow
we don't want to just send a password

through the there were
someone else to list then and

then sign in with with your credentials
and then take your money so

no one cases if you both have this
kind of you both have a key know they

stand a good target for you both have
some secret key you could use that to

encrypt your password turn into
something random and then an.

And then evil you know with this evil guy
this will see random information can't

do anything with it and this is
totally done through something called

an advanced inclusion standard which
they exist it's off in some clever way.

And it's become so important that it's
actually built in you know Intel builds

this into your chips so
it's almost as fast to encrypt things

as this to send normal
normal informations so.

Most everything is encrypted
on the web these days.

But there's a problem with this.

You know I have my key I want to I want
the other never logged on to my bank for

my phone before How's that bank go to

get the other key the key try to send
the key Well then evil guy will see it and

then he'll be able to decrypt everything
or work it all so this is where

public key cryptography came in I think
design magine is ours they are when them.

So I'm going to kind of give a pictorial
version of what happens which is think of

it as a safe so the bank has this safe
it knows the combination to the safe

no one else those combinations safe and
so it takes the safe and

it sends the the open safe to the user
of the phone you know that and

then you take that the phone puts
the key into the safe locks the safe.

Sends a safe over since it's a locked
the evil person can see inside and

I want to spank year you know the bank
knows the combination can open the safe

and pull out the fallout the key that's
the basic idea now of course the problem

is how do you send a safe through
the Internet it's not like we have

teleportation anything like that but you
know just a mathematical tricks actually

number theory comes into play and
he uses the fact that it.

It's easy to multiply two numbers together
but and if they're very large numbers it's

very hard to factor those numbers and
so using that ideas and

similar kinds of techniques from Number
Theory and group theory you can actually

create protocols that actually make
this this this kind of scheme real the.

Work out.

And.

Then once you have the keys you
can equip the password now and

imagine that the target fee is just.

A small piece to security so I mean we.

With cryptography we can
basically you know this

this problem is pretty well study and
we have solutions you know work well but

the problem is is that great of
everything else work perfectly but

you know Joe does always code could
have it could be bugs in your code or

code could allow people to do things that
they weren't planning to do so people you

have to be careful the people who support
you code also humans curiously you could

fool humans in the given passwords to you
know pretending to be something else so.

So they have no cryptography turns out

that part actually turns out to
be relatively easy compared to.

Compared to trying to.

Handle all the security and so
while we can out of the target fee

security is becoming a real challenge
to try to get it do it right.

It's an active area that's an area
that we do a lot of in Georgia Tech so

let me give you the whole picture now.

Of how these things work you start
the application could target fee is

usually considered part of the you
know right below the application

the C C C H E T P S You know in your title
then you've got the security of link and

that that that that has cryptography going
on in the background basically what I

showed you and above that the transporter
layer which connects the the application

with the program and then you have the
network layer which contains the computers

and then eventually the link layer
which connects all of the machines.

Now also last summer I was thinking OK
this is the way things are no no no

I mean that Google decided now why
do we need all these layers so

Google besides cheat and
created something called Quick.

And that's amazing how these things
like change like you know my ideas and

ideas was like you know since it as Larry
said OK I'll describe how Google works but

now Google is changing the rules and and.

So everything I know I start to write a
chapter in a parking lot was that it that

a day so Google Quick basically takes
these three layers application toggle fee

and transport and kind of does it one way
they did the real important thing is that

is that it actually takes a while to
get the information from one computer

to another see really want to minimize the
number of what's called round trip parties

around trip times that back and
forth between computing and

Google Quicken actually limits
you to one and they claim

zero sort of round trips that you need
to actually transmit information so

they really want to be all that do things
very quickly and because it has crypt and

right you know basically you your
deafening using quick if you use a Google

browser like Chrome and going to google
web site you know Google You Tube.

Or whatever else Google owns and.

But now that it did try to make it
a standard so you start to see it in other

browsers and other websites to see if
it actually becomes a standard but

it's actually interesting too
because it has crypto built then

if you mean if your side is secure it
actually with crypto actually now runs

faster than it did without crypto because
if you can use this quick protocol

to get it get it that I think what
part of why Google is trying to do no.

Push people to encrypt everything.

And the I think it's just
quick because it's fast but

the shirt stands for something.

He does not then for
quantum innocently what.

Sure that you do with quantum.

Com So that's most of my talking not a lot
you know it had to and then really think

about everything I just said now
happens in six tenths of a second and

that's pretty amazing you got this all
this processing that's going on all these

answers take your question Parsons your
question sends it over does the processing

sends it back and it gets to you back
in sixty and I think that's just I mean

it's amazing and I left out a lot of
pieces you know as a limit to how more and

I can I can do the stock I wanted to end
up with a couple things One is I can't

really avoid a lot of the AT I mean it's
amazing now that a lot of the ethical

issues the using it has to worry
about computing all of a sudden or

a really popping up and I mentioned
already a couple one is of course

you know what happens when one
company controls a lot of data and

other people might trick people
into giving up that data.

As as happened is these are Mark
Zuckerberg is trying to fix Explorer and

how that could affect.

You know if you mean social networks
of both in the last five years

created democracy is and

seem to destroy democracies I mean
it's amazing what could happen.

And then you know I also
mentioned assault driving cars and

you have to be careful I mean even had
like my rabbi starts arguing with me

about you know what happens of a self
driving car has to make a decision between

killing one person or
killing the occupants of the car and.

And I just told them Listen you know

the subtle car is not going to get that
position that started to be distracted so

it's just going to stop them and
convince them.

That you know those that car conversations
you have with you know just general people

and there are a lot of ethical
challenges I mean there is.

Issues of you know how do we how do
we increase diversity in our field

it's not nearly as diverse
as it should be and

how do you make people feel
comfortable and safe in computing.

This issues of privacy and
these issues of fairness you know you

have missed machine learning algorithms to
work in and use in the make decisions but

if they're learning from him to learn
different people's pages this is already

going to give biased results and so
I do make sure things are fair and.

And then and
then the questions about fairness and

privacy you know they teach there
are different for different countries for

different cultures for different you
know who gets to choose what's fair and

what's private when you've got this
Internet around the entire world and

we're all trying to deal with it at those
those are all and it's lots of others out

there all those are all important but it
to me the biggest issue is automation is.

Is you know are computers able to do
more and more things that humans can do.

So this is.

This is this is a test was or the made
a plant MAKE ME is no model three.

Of the way and

then and then I just had to put in the
suite at the on must had the other day.

Yes the excessive automation that has
there was a mistake to be precise my

mistake humans are under rated.

So this so but you know the reality is
is that you have baby for now but these

things car computers again a lot smarter
a lot faster they're going to automate and

not just you know manual labor like you're
creating cars but they're automating.

Just about every
profession I mean you can.

Never really eliminate
a lot of jobs in law firms.

I don't think anyone is even safe
you know even as professors.

You know we have this automated that.

That is happening out there so who knows
where all our jobs are going to go and

you know a lot of people wary.

That the automation is going
to become self-aware and

it will get into this world of
the Terminator but actually my the movie

that I think you are more likely to see
which scares me even more is Wall-E.

where we just all become little nothing
for us to do with sit around and

get that wow the computers do everything.

And then me and.

That's good that's good timing let me and
talk a little bit about the future.

So what if this is my last live a lot of
the pictures are the one on the right is.

A small piece of the fifty bit quantum
computer so it's a real breakthrough and

in fact even more reason this by can
find pictures you know a guy who cleans.

Quantum computing is really
a kind of a really scientifically

mathematically really cool you
know lot of interesting ideas.

Will it actually lead to something
where it can it can do things that.

Have actually be useful in computing.

And then I'd been a jury's out I'm
a bit of a skeptic myself but but

there's certainly a lot of effort and
money going in there.

But I Had To me the more exciting thing is
actually what's a little bit more mundane

which is the thing on the left which is
a new new type of memory that Intel and

Micron developed together called three D.

crosspoint and the ideas that each of
these little cells is is memory and

you can kind of control it by you want
to do all the cell by you know this bar

in this bar that's about all I know but
it's a good like a very different model

that it's it doesn't use transistors and
electron.

So it's what's called nonvolatile which
means it doesn't move it doesn't disappear

when you're in your power goes down you
can it's three dimensional sequential

lot of memory into a small amount of
space and and you could put it very close

to the processor which means you can get
a lot of very fast access out of it and.

And so it actually doesn't even come out I
mean in this form until the fall that if

you're claiming that it's going to come
on a four but it really well enough can

this kind of memory can really
be transformational of this

even a simple idea and if you look at out
say almost like nearly half the the job

talks I'm seeing in my department are
related to memory one way or another and

it's because of new technologies like this
and you know what can you do with it where

you can have if you can store like
huge databases and memory and and

you can really save and recovery time if
you're from Asia as doesn't disappear when

a para goes out they're also going to
have better power control so you can do

a lot more things you don't have to keep
things powered to keep the memory going.

And you can you can also enable what
becoming very important you know

in in something called either fog or dead
computer in a very closely related where

you put the computer in right at the near
the user and you need that because

if you want to do something like virtual
reality you know where you need things

that you know when you move your head is
going to move right away or you want or

a car is trying to decide in order made a
car is trying to decide whether or not it

should stop for a pedestrian you know
six tenths of a second is very quick but

it's not quick enough and so many minutes
at some times even the speed of light

is too slow as they really critically
important to get the computing power for

some applications really
close to the user and

that's what and that's I think
you know least in the short term

you know I think a lot of where
a lot of excitement is going to go.

So with that I will thank you.

In.

The dark where.

You know it's not really
a different structure it's really

I can give you the dark
web is I mean it's.

It's basically just a lot
of the Internet people are.

Try to keep fighting and I try to
discredit our club I guess it's it's.

It's really just the parts of the Internet
that don't normally see the search engines

hide because it's a lot of nasty things
that go on in there are legal or

unethical or just downright immoral
stuff that happens there so

it's where people are the people create
the sometimes the creator own data centers

or they'll create their own protocols be
a lot of secure protocols and things where

they can do things anonymously can't be
traced they can say what they want to do

they can trade coins to buy a bad
good things bad things and it's just

when it's more of a concept and then I
would say a physical location in the web

right.

Right right.

That was sort of my own
with the Wally story.

Yeah so I mean we're really I think we're
seeing a lot of advances now because

people are actually creating specialized
hardware which is really why you seen this

increase I mean it's not going out I think
it's going to last when you get that much

faster for too much longer but there's
Right now we're coming up with new ways to

do these specialized the original you
started with graphics you wanted graphics

cards in computers and people realized
well you can use these ideas for graphics

uses graphic cards you can also use that
for machine learning so people stop and

if you realized or the graphic operations
aren't enough you have to do this fancier

operations tensor operations that's where
you could write the convolutional now and

they're at the right the convolutions and
the backpropagation and all that so.

What's going to mean for jobs I mean
I've gone to you know I went to.

A few months ago went to this C.R.A.
this Computer Research Association and

all day on the future of work and
it's pretty scary I mean it's.

I mean they're saying you
know maybe half the jobs.

That are currently around today will
go away in ten fifteen years and

then all of progressed from there.

And then the quick question is What do
you do about it you mean can you keep

retraining people in the going to jobs
to be you know in the past it's been

the case that when jobs go away new jobs
take their place it's not clear whether or

not it's going to keep going I mean it's
maybe that will happen maybe not it's

really hard to predict and and then you
talk in the conversations about you know

maybe universal link that
not everyone has to work but

I think people see workers say no they
want to be productive in society so you

know Panem not to be productive in society
that it's not really a solution so I mean.

It.

It's a it's a big problem and there's a
lot of people trying to suggest approaches

but I think it's a conversation that we're
really continuing to have and I'm not

I want to be a Luddite I mean I now think
you can stop in any way the technology.

But but we have to keep in mind
I mean when you're I mean it's

it's an incredible field computing and
I mean I you know I did and

you know when I got into it but boy I mean
it's really become especially the last ten

years this incredible feel a lot
of really interesting science and

engineering going on but also the social
aspects and you know how it's changing

society and it's really amazing it's
doing a lot of amazingly great things but

you really do worry about you know
what the long term effects can be and.

Just so hard to predict and I'm on.

Just something we need to be careful
of and people will watch out for

him and
think about ways how do we mitigate it.

Right.

You know.

OK Tyreese about creating the secret codes
I mean so the problem with security is

something to talk of is part of security
the security is a much broader thing but

security is also for
example you take advantage of.

For example your send a message to
a computer but if you make the message

really really long you might overwrite
memory that wasn't meant to be overwritten

and then you can maybe take control of
the computer and of a standard the net so

you have to write your code in ways
to make sure that doesn't happen.

You know you have secret codes I mean
we were to store the passwords and

how do you store them you have to be
careful about that you have to be and

then there's a whole human risk factor I
mean even talk about in the whole area.

You know human computer interaction
which I don't really get too much but

you know you have to be really careful
Heidi I describe how things happen and

how people are or
interact with computers because it's it's

just far too easy to trick
people into thinking that.

You know someone with
authority is asking for

their password it looks like a real
web page but it's really a fake thing.

To try to prevent that from happening and.

One nice thing about the cloud is that you
know when you're on the cloud then you

have one company really working hard on
these questions as opposed to trying to do

it yourself but still these these are I'm
to come up with these great solutions but

there's it's always hard to get everything
perfect and people who are trying to break

in are really really smart so it's
a battle we have to keep adapting to them.

The Russians I mean you know we're
pretty good at it ourselves too.

Yeah.

Right.

I do I did this is
a great you know we have

to in this is kind of discussion we've
been having in our department how do you

do you how do we rethink the computing
you know the the threads in everything

we've been doing in computing
we were trivial about dozen

years ago you know they don't necessarily
reflect what's going on today so

we just have to constantly rethink it and
there are also a lot of basic principles

that in computing which which I think will
always be true about about abstraction

about information about you know
algorithms and complexity and

modeling and networks of a lot of things
that are that are going to change these

those those first few classes maybe change
what programming language you use but

the basic concepts are the same but
we really do have to rethink and

known everything being separate
I mean the big change for

example is used to be the networking was
completely separate from your princes and

we separate from the computer architecture
these things are kind of merging now so

we need kind of classes on cloud computing
or now we're going to create a new class

on a large chains in crypto currency
things like that so that we need to get

it and now of course we are classed
as a machine learning of evolved and

got humongous So you just have to keep
you know rethinking the curriculum and

and and
evolving it moer Russia Garfield you know.

The computers there really
the idea of computing kind of

really get started until the thirty's and
then even real computers then really

happened until the fifty's digital
computers as we know him so.

So we're still trying to figure it out.

I.