my name is Steve angles I’m one of the
associate professors in the department
of computer sign
my main thing is to ask or pose one
question of everyone today and that is
what is the difference mainly behind
computer intelligence and human
intelligence now I’ve been working in
artificial intelligence for a while you
know basically since the 90s and ever
since I started in AI there’s always
been a standard test to determine
whether computers have reached human
level intelligence
so for the longest time when I was
growing up that was chess you know if a
computer was able to beat humans at
chess they would be intelligent enough
to be considered to be smart and so this
continued for a while until about 1997
IBM created deep blue deep blue managed
to beat the world reigning champion in
chess Garry Kasparov and so it wasn’t
that people suddenly thought that
computers were intelligent
they certainly decided you know what we
need to change the bar we need to find
other tests to determine whether
computers are actually considered to be
intelligent or not and there was various
ones some of them we’re trying to say
what can they know things as well as
humans do you know can we actually ask
some questions about certain subjects
and have them answer in intelligent ways
and so that was considered to be a new
standard of course there’s no way that
computers would be able to hit this
until about 2011 IBM again came up with
the idea of Watson Watson was able to
read through all sorts of unstructured
text and be able to not only understand
what it was reading but be able to
answer questions intelligently about it
so what they do they put it on Jeopardy
and it beats the reigning jeopardy
champions people who had 170 episodes in
a row and beat them tidally so they
changed the bar again they decided we’re
going to go back to some games games
where humans have never been able to be
beaten by computers even the amateur
players of certain games will never be
able to be taken down by the smartest
computer and of course the game for this
case was go and so for a while that was
considered to be you know this was
something that was never going to be
beaten until earlier this year google
deepmind and alphago managed to beat the
world reigning champion ago so basically
every time there’s a standard a test for
computer intelligence
eventually you know some computer beats
it and this is usually the point where
my friends ask me ok so at what point
are computers going to start taking over
the world
when is the singularity going to happen
and they’re going
slave mankind but you know then they
start thinking you know what it’s not so
bad when it comes to this race between
human and computer intelligence you know
humans intelligence will always be
better or superior than computers and
you know then they start thinking about
you know the less harmful robots and
they don’t feel so bad about you know
being taken over by computers and you
see this a lot I new ever comes to
talking about what are the limitations
that computers have you always hear the
exact same things well computers are
human creations so they’ll never
outstrip their creators Computers can’t
think for themselves right or they have
no imagination no creativity now the
first ones silly I mean babies are human
creations and yet they managed to get
intelligence over a certain amount of
time and we know that there’s a lot of
computers now that can think for
themselves that can make decisions you
know intelligent ones even better than
humans can but it’s the last one whether
computers are capable of imagination
that’s the one that I want to address
and for this I’d like to refer back to
one of those robots taking over the
world movies I Robot in there there’s a
scene where Will Smith is interviewing
one of the robots and he asks can a
robot write a symphony can a robot turn
a canvas into a beautiful masterpiece
and if anyone’s actually seen the movie
you know the response from the robot was
can you and yeah I know it seems a
little flippant but really this gets at
the crux of what it means to tell if a
computer’s intelligent or not it’s can
the computer do a task as well as a
human can this is actually called the
Turing test and this was invented by
Alan Turing and the idea was I want to
be able to say for a given task if a
computer can do as well as a human can
then it is as intelligent at least at
that task so this is where the imitation
game came from can you imitate a human’s
behavior we see a lot of these kinds of
examples these days one of them is
speech recognition I’m not talking about
having full conversations with computers
I’m just saying can you say something to
your phone and have it know what you’re
saying and for this you know speech
recognition you know Siri is pretty good
I mean it’s not perfect but then again
if you looked at how humans perform at
this task you would know that humans are
really bad at this task – so when it
comes to the Turing test
maybe computers aren’t perfect but
they’re at least as good and performance
because the humans would be navigation
if you need to get from point A to point
B our computers are good as good at
doing this task as humans are some would
say better right they can take in more
bits of information about traffic flow
about the time of day then humans would
ever be able to think of and while we’re
on the subject of navigation about
driving you know can you actually have
computers take over the wheel these days
yes these days there’s this feeling that
with you know the advances in
self-driving cars it’s going to be a lot
safer to be in a car that’s driven by a
robot then it’s going to be in the car
driven by a human especially with smart
cities that are going to incorporate
information from all these different
sources let’s get back to creativity can
computers be as creative as humans and
this is something that’s come up a lot
more recently because with the advent of
the advance of neural networks and the
ability to learn off of you know various
sources of data if you have enough data
you can usually learn how to do a task
as well as a human can and so this study
is well then if we can learn how to
drive if we can learn how to do speech
recognition then why can’t we teach them
creativity with enough data our computer
should be able to take all the things
that humans can do regardless of what
the task is and be able to do it as well
as a human can as long as we have enough
data it’s all just a matter of having
sources things that you can use to draw
from in order to create this corpus of
knowledge so this gets into the research
that I’ve been doing I start off as a
natural language programmer when I first
started in AI which meant that I was
doing a lot of things with English text
and what we were doing were things that
could actually read through English text
and understand the things that is
reading and create models for to be able
to recognize the patterns within it well
we use these same models not just to
record and classify text but to be able
to generate it so we actually said you
know I can take in a bunch of Obama
speeches and we combined all the
different like language and word
constructs that are used and created new
original speeches in Obama’s style
there’s other people who have done
things like chatbots where if you’re
playing a game online and you’re talking
to somebody are you really talking to a
person or are you talking to this AI
that’s going a really good imitation
there was an example of someone who
actually made a research paper generator
where it actually created a journal or a
journal article and submitted it to a
conference and got accepted so this is
the sort of thing where computers are
able to create new interesting original
text and this led into my research on
music generation I have software where
you can feed in music files and it’ll
actually read through and understand the
patterns in the music and be able to
generate original music in that style in
real time forever at first it wasn’t
great it would play things that didn’t
have a lot of direction it was more like
a distracted jazz pianist but over time
we’ve managed to build these things and
now it’s actually has heard a larger
scale progression so that it now has the
sense of a start and middle and an end
and it actually has like an overall
musical flow this is one of those things
where you know now I do work in video
game design people use the software in
order to generate background music for a
level and then as you move from one
section of a level to another the music
will change as it goes from one model to
the next this continued into other areas
I worked also another video game areas
like terrain generation so we would
actually take the the 3d terrain from a
video game level and we would train off
of it and be able to generate new
original video game levels that looked
the same or fit in that same kind of
environment so this was our contribution
to things like visual art and anyone
who’s seen some of the work that’s going
on in neural networks has seen that
certain images have been done where
neural networks were trained off of
existing images and create new original
pieces of art things that no one has
ever seen before and some of these
neural networks when you don’t give it a
specific task or a specific domain and
you just let it roam free create some
things that we’ve never seen before
you know the AI equivalent of dreams so
so then this brings us back to the idea
of okay so computers are obviously able
to generate art but is it any good well
what I opposed to you is is human aren’t
any good I mean this is the thing that’s
been argued for a lot of the artwork
that you’ve seen on the right hand side
you know people have contested what
makes a good piece of art even the piece
that’s on the bottom right was generated
by Coco the gorilla is it art
I’m not sure and neither am i here to
answer that question my whole thing is
just to say well you know what if AI has
been making advances in all these
different areas one of the next ones
that’s going to be advancing into is
creativity because really I mean when we
talk about what creativity is one major
thought is it’s really just intelligence
having fun and whether it’s a human
intelligence or an artificial
intelligence the same rules apply if you
let this thing train off of existing
pieces and see where it can go it can
create all sorts of interesting things
so I believe that our future is not just
going to be limited by you know what
kind of things computers are able to
make we’re actually going to be the ones
who are teaching these new generations
of computers how to do new things and in
the end I feel that this is going to be
the next thing the next impossible task
that computers are going to be able to
defeat thank you very much
[Applause]
you
associate professors in the department
of computer sign
my main thing is to ask or pose one
question of everyone today and that is
what is the difference mainly behind
computer intelligence and human
intelligence now I’ve been working in
artificial intelligence for a while you
know basically since the 90s and ever
since I started in AI there’s always
been a standard test to determine
whether computers have reached human
level intelligence
so for the longest time when I was
growing up that was chess you know if a
computer was able to beat humans at
chess they would be intelligent enough
to be considered to be smart and so this
continued for a while until about 1997
IBM created deep blue deep blue managed
to beat the world reigning champion in
chess Garry Kasparov and so it wasn’t
that people suddenly thought that
computers were intelligent
they certainly decided you know what we
need to change the bar we need to find
other tests to determine whether
computers are actually considered to be
intelligent or not and there was various
ones some of them we’re trying to say
what can they know things as well as
humans do you know can we actually ask
some questions about certain subjects
and have them answer in intelligent ways
and so that was considered to be a new
standard of course there’s no way that
computers would be able to hit this
until about 2011 IBM again came up with
the idea of Watson Watson was able to
read through all sorts of unstructured
text and be able to not only understand
what it was reading but be able to
answer questions intelligently about it
so what they do they put it on Jeopardy
and it beats the reigning jeopardy
champions people who had 170 episodes in
a row and beat them tidally so they
changed the bar again they decided we’re
going to go back to some games games
where humans have never been able to be
beaten by computers even the amateur
players of certain games will never be
able to be taken down by the smartest
computer and of course the game for this
case was go and so for a while that was
considered to be you know this was
something that was never going to be
beaten until earlier this year google
deepmind and alphago managed to beat the
world reigning champion ago so basically
every time there’s a standard a test for
computer intelligence
eventually you know some computer beats
it and this is usually the point where
my friends ask me ok so at what point
are computers going to start taking over
the world
when is the singularity going to happen
and they’re going
slave mankind but you know then they
start thinking you know what it’s not so
bad when it comes to this race between
human and computer intelligence you know
humans intelligence will always be
better or superior than computers and
you know then they start thinking about
you know the less harmful robots and
they don’t feel so bad about you know
being taken over by computers and you
see this a lot I new ever comes to
talking about what are the limitations
that computers have you always hear the
exact same things well computers are
human creations so they’ll never
outstrip their creators Computers can’t
think for themselves right or they have
no imagination no creativity now the
first ones silly I mean babies are human
creations and yet they managed to get
intelligence over a certain amount of
time and we know that there’s a lot of
computers now that can think for
themselves that can make decisions you
know intelligent ones even better than
humans can but it’s the last one whether
computers are capable of imagination
that’s the one that I want to address
and for this I’d like to refer back to
one of those robots taking over the
world movies I Robot in there there’s a
scene where Will Smith is interviewing
one of the robots and he asks can a
robot write a symphony can a robot turn
a canvas into a beautiful masterpiece
and if anyone’s actually seen the movie
you know the response from the robot was
can you and yeah I know it seems a
little flippant but really this gets at
the crux of what it means to tell if a
computer’s intelligent or not it’s can
the computer do a task as well as a
human can this is actually called the
Turing test and this was invented by
Alan Turing and the idea was I want to
be able to say for a given task if a
computer can do as well as a human can
then it is as intelligent at least at
that task so this is where the imitation
game came from can you imitate a human’s
behavior we see a lot of these kinds of
examples these days one of them is
speech recognition I’m not talking about
having full conversations with computers
I’m just saying can you say something to
your phone and have it know what you’re
saying and for this you know speech
recognition you know Siri is pretty good
I mean it’s not perfect but then again
if you looked at how humans perform at
this task you would know that humans are
really bad at this task – so when it
comes to the Turing test
maybe computers aren’t perfect but
they’re at least as good and performance
because the humans would be navigation
if you need to get from point A to point
B our computers are good as good at
doing this task as humans are some would
say better right they can take in more
bits of information about traffic flow
about the time of day then humans would
ever be able to think of and while we’re
on the subject of navigation about
driving you know can you actually have
computers take over the wheel these days
yes these days there’s this feeling that
with you know the advances in
self-driving cars it’s going to be a lot
safer to be in a car that’s driven by a
robot then it’s going to be in the car
driven by a human especially with smart
cities that are going to incorporate
information from all these different
sources let’s get back to creativity can
computers be as creative as humans and
this is something that’s come up a lot
more recently because with the advent of
the advance of neural networks and the
ability to learn off of you know various
sources of data if you have enough data
you can usually learn how to do a task
as well as a human can and so this study
is well then if we can learn how to
drive if we can learn how to do speech
recognition then why can’t we teach them
creativity with enough data our computer
should be able to take all the things
that humans can do regardless of what
the task is and be able to do it as well
as a human can as long as we have enough
data it’s all just a matter of having
sources things that you can use to draw
from in order to create this corpus of
knowledge so this gets into the research
that I’ve been doing I start off as a
natural language programmer when I first
started in AI which meant that I was
doing a lot of things with English text
and what we were doing were things that
could actually read through English text
and understand the things that is
reading and create models for to be able
to recognize the patterns within it well
we use these same models not just to
record and classify text but to be able
to generate it so we actually said you
know I can take in a bunch of Obama
speeches and we combined all the
different like language and word
constructs that are used and created new
original speeches in Obama’s style
there’s other people who have done
things like chatbots where if you’re
playing a game online and you’re talking
to somebody are you really talking to a
person or are you talking to this AI
that’s going a really good imitation
there was an example of someone who
actually made a research paper generator
where it actually created a journal or a
journal article and submitted it to a
conference and got accepted so this is
the sort of thing where computers are
able to create new interesting original
text and this led into my research on
music generation I have software where
you can feed in music files and it’ll
actually read through and understand the
patterns in the music and be able to
generate original music in that style in
real time forever at first it wasn’t
great it would play things that didn’t
have a lot of direction it was more like
a distracted jazz pianist but over time
we’ve managed to build these things and
now it’s actually has heard a larger
scale progression so that it now has the
sense of a start and middle and an end
and it actually has like an overall
musical flow this is one of those things
where you know now I do work in video
game design people use the software in
order to generate background music for a
level and then as you move from one
section of a level to another the music
will change as it goes from one model to
the next this continued into other areas
I worked also another video game areas
like terrain generation so we would
actually take the the 3d terrain from a
video game level and we would train off
of it and be able to generate new
original video game levels that looked
the same or fit in that same kind of
environment so this was our contribution
to things like visual art and anyone
who’s seen some of the work that’s going
on in neural networks has seen that
certain images have been done where
neural networks were trained off of
existing images and create new original
pieces of art things that no one has
ever seen before and some of these
neural networks when you don’t give it a
specific task or a specific domain and
you just let it roam free create some
things that we’ve never seen before
you know the AI equivalent of dreams so
so then this brings us back to the idea
of okay so computers are obviously able
to generate art but is it any good well
what I opposed to you is is human aren’t
any good I mean this is the thing that’s
been argued for a lot of the artwork
that you’ve seen on the right hand side
you know people have contested what
makes a good piece of art even the piece
that’s on the bottom right was generated
by Coco the gorilla is it art
I’m not sure and neither am i here to
answer that question my whole thing is
just to say well you know what if AI has
been making advances in all these
different areas one of the next ones
that’s going to be advancing into is
creativity because really I mean when we
talk about what creativity is one major
thought is it’s really just intelligence
having fun and whether it’s a human
intelligence or an artificial
intelligence the same rules apply if you
let this thing train off of existing
pieces and see where it can go it can
create all sorts of interesting things
so I believe that our future is not just
going to be limited by you know what
kind of things computers are able to
make we’re actually going to be the ones
who are teaching these new generations
of computers how to do new things and in
the end I feel that this is going to be
the next thing the next impossible task
that computers are going to be able to
defeat thank you very much
[Applause]
you
Please follow and like us: