when I was a kid my favorite book was
called the big jump I’m not sure how
many of you have seen or heard of the
big jump but it was a great story about
a king who was able to leap to the top
of his castle in a single bound
one day the king had an idea to have a
contest and give a prize to any one of
his subjects who could make that big
jump so he told them and they all tried
and they all failed to do that and I
think the Kings challenge about making
that big jump is really the same
approach that we should take with
artificial intelligence because there is
so much hype about artificial
intelligence so much jargon and promises
warnings and and we really don’t know
what AI is or what it does so we can’t
make that big leap of understanding into
artificial intelligence and how it can
change our lives maybe for the better
maybe for the worse we don’t know yet I
am no engineer or mathematician as
someone who is immersed in digital and
social media I felt it was imperative
that I tried to get a handle on AI
because it’s having a profound impact on
all of our lives whether we know it or
not and that’s only going to become more
and more and more of an impact as time
goes on so what I wanted to do today is
share my journey of discovery try to
explain and simplify some of the
concepts around AI in order to give all
of you a little bit of a better
understanding of where we are and where
we may be heading ai is not a new
concept it started in the 1950s actually
a computer scientist named John McCarthy
coined the term and he was trying to
figure out a way to get machine
to solve problems more like the way
humans do so I thought if I’m going to
figure out and go back to the roots of
AI I need to find a way to get back to
the 1950s to look for some clues and and
I went there and I found this the magic
8-ball a lot of you recognize it it’s it
was a popular toy and what you did is
you asked the magic 8-ball a question
and it gave you an answer sometimes it
was a really cryptic answer but it was
an answer nonetheless and suddenly it
hit me the magic 8-ball is an early form
of AI the only difference data it didn’t
have any data so no matter how many
times I asked it questions it couldn’t
improve AI artificial intelligence is
all about predictions and really that’s
what the magic 8-ball was – so it was
artificial yes but not intelligent now
children learn by absorbing the world
around them
the imitate patterns they make
assumptions and then they adapt robots
learn in much the same way as children
in the beginning a machine or a AI is a
lot like a child and what it does is it
absorbs its world only its world is made
up of data the data we feed it so
numbers pictures words and it uses its
algorithms which is a term we all hear
all the time in all of the algorithm is
it’s a set of steps that we use that a
machine uses to solve a problem and it
tries to find patterns and then predict
what the right answer will be with the
best statistical probability and it’s
really simple the more data you have the
better statistical predictions and so
you have better results we have three
data types because we’re over
overwhelmed with big data and then three
types are structured data which I think
all of you are familiar with structured
data is like numbers or
what goes into a table unstructured data
those are things like the words we say
or the text and an email or a blog post
pictures unstructured data videos or
unstructured data and then we have semi
structured data and that’s kind of in
between the two of those so the number
of followers you may have on Twitter as
an example of semi structured data these
days because computers are faster and
faster and they have more processing
power and we can store vast amounts of
information in the cloud not these
clouds up there but that cloud we’re
able to interpret and process more
things so beyond tables we can interpret
talk which is called natural language
processing and also what we see which is
called machine learning but we’re
getting into jargon and I want to step
back and look at something simple
something you all know which is search
now search is the number one thing we do
online in fact 92% of all our
experiences begin with search Google
uses something I mean you’re not
surprised that it uses AI to power its
semantic algorithm and the semantic
algorithm understands the relationship
between words so it knows for example
that John Smith is probably a person and
it’s conversational interface not only
understands that John Smith is a person
but it understands the relationship
between that name and personal pronouns
that describe it you can try this at
home
afterwards open up your browser open up
Chrome click on the microphone there and
ask it a question say who is the Prime
Minister of Canada pretty easy question
we all know the answer to that and it
will say the Prime Minister of Canada is
Justin Trudeau then open up a new tab
and say who is he married to not who is
Justin Trudeau marry to but who is he
married to and it will say he is married
to Sophie Gregoire Trudeau then open up
a third tab and say how many children do
they have and it says how many children
they have and you can do this for maybe
one or two more tabs before Google
breaks down
that’s pretty great in a lot of ways
because Google understands our context
right but it’s also a little scary
because Google understands what we want
it kind of gets us and in doing so it’s
starting to build trust we’re starting
to trust it now ai is only as good as
the data the data it’s fed so think of
data as an AI version of a great meal
and you can have a data set that’s
filled with junk food or with healthier
options now imagine if you have a data
set that’s made up of greasy burgers
fries and soft drinks versus one that’s
made up of organic salads which do you
think is better well the answer is it
depends on what you’re using it for so
if for example that greasy fried data
set is used to build an app that’s
trying to get you to to eat at fast food
restaurants that’s what you want to use
but the point is neither one of those is
complete because it’s not representative
of a wider population and where we start
to get into trouble with this is when it
comes down to things like facial
recognition because the quality of
predictions depend on the quality and
the variety of data and otherwise
there’s too many biases and errors and a
lack of data diversity is why Alexa and
Google assistant are having trouble
understanding requests in accented
English or why Amazon’s facial
recognition app matched 28 members of US
Congress to the mug shots of convicted
felons I know it’s pretty scary and
we’ve got to get this right because if
it’s a facial recognition application
innocent lives could be affected in
China right now what they’re doing is
they’re testing a I powered glasses that
will police will wear to identify
suspects on busy streets that sounds a
lot like a server
and stayed here in Toronto at the
University of Toronto researchers are
trying to counteract that and what
they’re doing is they’re using something
called adversarial learning to teach the
AI to identify the image of something
say a turtle and then adjust a few
pixels in that to trick its algorithm
into identifying it as something else
and what they’ve done is they got an AI
to identify a picture of a turtle as a
picture of a gun now if you show human
beings those same two photos someone
that was adjusted and the first one we
can’t tell the difference but you’d see
how the battle grounds between
surveillance States and facial
recognition and facial fooling are
starting to take shape
MIT came out with an interesting
experiment where they used a dataset of
psychopathic personalities to create
something called Norman Norman is a
deviant AI and it’s named after Norman
Bates from psycho and if you see Norman
online don’t book a hotel but beyond all
these statistical models because Norman
and all of the other predictive
technologies they’re using some
scientists are moving into what’s called
embodied learning what is embodied
learning embodied learning goes outside
of the lab and takes in an unpredictable
environment like the world around us
company in Silicon Valley open AI
they’re teaching a robotic arm to grip
objects of various shapes sizes and
weights in much the same way that we
learn to grip facebook created two BOTS
they created a tourist spot and a guide
bot and they had the tourist spot asked
the guide bot questions to find out if
it could get around New York
successfully and it did that I’m not
sure how many of you have heard of
Google duplex that’s Google’s
voice-activated application so Google
duplex calls businesses hair salons and
restaurants to either book reservations
or make an appointment
and it sounds really real complete with
pauses and speech disfluencies those are
those hums and ahhs that all of us
associate with humans in fact the people
at the other end didn’t know they were
talking to a machine
now Google after the excitement caught a
lot of flack for that and in its second
iteration Google duplex had to identify
itself as a digital voice assistant how
many of you have seen Sophia the robot
so Sophia the robot she it does a lot of
publicity interviews with media has been
on a date with Will Smith and kind of
seems real but Sophia is not sentient
even though it says it has feelings it’s
just programmed to respond to a whole
bunch of different commands so for
example if you asked Sophia question
what Sophia will do is and I’m not in
recognition Lille and give it a little
smile just like us so we think it’s
human the head of facebook’s AI said
Sophia is really just smoke and mirrors
but Sophia does give us a glimpse into
our future all of these examples Sophia
duplex the bots I mentioned from
Facebook those are examples of
artificial narrow intelligence single
purpose AI and what that means is it’s
smart but it’s really good at doing just
one thing now a lot of us when we think
of AI we don’t think of single purpose
because that’s not really that scary
what we think about is something called
artificial general intelligence and
that’s really what’s depicted in a lot
of movies and fiction and what is a GI
artificial general intelligence it’s
when machines can do a whole bunch of
different things not just one goal and
they start to be more like us and and
the question is will they become
conscious and you know nobody knows that
the answer to that now there is a
start-up in Silicon Valley that’s
teaching machines not just to make
predictions but to understand conceptual
thinking and reason so it’s starting to
approach artificial general intelligence
in a different way are we going to merge
with machines
maybe Elon Musk talks about a neural
lace that is a connection between our
bodies and brains and computers that
could hopefully provide for better lives
and once it’s not here but it’s not that
far away if you think about pacemakers
and cochlear implants and as machines
start to sound like us more and more and
in some cases look like us and give us
answers quicker than other people do we
could start to trust BOTS more than
people and now think about it if you’re
in your car and a passenger contradicts
your Google Maps directions who are you
gonna believe that begs the question how
can we coexist in a world with machines
and and what types of relationships are
we gonna have with those machines it’s
really the subject of some research I’m
doing right now because has a potential
to make life really really amazing or
really really bad so for example do we
teach children to be polite to machines
when they’re asking a request do we say
hey do you need to say please and thank
you and if we do that are we raising a
generation that can’t tell the
difference between say people and a
blender for example now right now Google
assistant and Alexa both have settings
that reward kids when they’re polite
asking them questions so you have a
situation of BOTS disciplining kids do
we trust them to do that and how is this
going to affect all of our interpersonal
relationships there’s also the manner of
work and all the disruption we are going
to face in the next maybe five to ten
years where so many jobs are displaced
by machines and it’s not just the jobs
you may think driving trucks or in
factories its educated work so marketing
lawyers accountants radiologists
how are we going to retrain people is
that a job for government is it a job
for education is it a job for the
private sector or for all three it’s
something we need to think about max
tegmark is a professor at MIT and he
wrote a fabulous and fascinating an end
kind of scary book called life 3.0 and
in it he poses some very challenging
questions so he asks and something to
consider is it ethical to enslave a
superintelligence so a sentient machine
to do our bidding and what happens if
that machine wakes up even more changes
its goals and becomes our master and
that’s why he says it’s so important
that when we talk about artificial
intelligence we think of it as something
called beneficial AI know what
beneficial AI does is it takes in our
morals our ethics our values and our
purpose and it asks those tough
questions so that we know that as AI
continues to be developed there will be
a place for people and he says we need
to build this into AI now because in the
future it will be too late but back to
the big jump I said everyone failed at
getting to the top of the castle but
that’s not true a little boy was able to
get to the top of the castle and he
figured out a really really novel way of
doing it how did he do it he got up
there by jumping up one step at a time
and I think that’s really the same
approach we need to take with artificial
intelligence because there is an amazing
potential out there to cure disease or
maybe spark creativity but the reverse
is also true and and we can’t just bury
our heads in the sand pretend it doesn’t
exist and then wake up in a few years to
a world we no longer recognize we need
to be smart we need to be critical and
we need to be open and we need to ask
how will a AI affect our work and life
and we need to keep learning about it
and what it can become
what we could become but we need to do
it one step at
time thank you [Applause]
called the big jump I’m not sure how
many of you have seen or heard of the
big jump but it was a great story about
a king who was able to leap to the top
of his castle in a single bound
one day the king had an idea to have a
contest and give a prize to any one of
his subjects who could make that big
jump so he told them and they all tried
and they all failed to do that and I
think the Kings challenge about making
that big jump is really the same
approach that we should take with
artificial intelligence because there is
so much hype about artificial
intelligence so much jargon and promises
warnings and and we really don’t know
what AI is or what it does so we can’t
make that big leap of understanding into
artificial intelligence and how it can
change our lives maybe for the better
maybe for the worse we don’t know yet I
am no engineer or mathematician as
someone who is immersed in digital and
social media I felt it was imperative
that I tried to get a handle on AI
because it’s having a profound impact on
all of our lives whether we know it or
not and that’s only going to become more
and more and more of an impact as time
goes on so what I wanted to do today is
share my journey of discovery try to
explain and simplify some of the
concepts around AI in order to give all
of you a little bit of a better
understanding of where we are and where
we may be heading ai is not a new
concept it started in the 1950s actually
a computer scientist named John McCarthy
coined the term and he was trying to
figure out a way to get machine
to solve problems more like the way
humans do so I thought if I’m going to
figure out and go back to the roots of
AI I need to find a way to get back to
the 1950s to look for some clues and and
I went there and I found this the magic
8-ball a lot of you recognize it it’s it
was a popular toy and what you did is
you asked the magic 8-ball a question
and it gave you an answer sometimes it
was a really cryptic answer but it was
an answer nonetheless and suddenly it
hit me the magic 8-ball is an early form
of AI the only difference data it didn’t
have any data so no matter how many
times I asked it questions it couldn’t
improve AI artificial intelligence is
all about predictions and really that’s
what the magic 8-ball was – so it was
artificial yes but not intelligent now
children learn by absorbing the world
around them
the imitate patterns they make
assumptions and then they adapt robots
learn in much the same way as children
in the beginning a machine or a AI is a
lot like a child and what it does is it
absorbs its world only its world is made
up of data the data we feed it so
numbers pictures words and it uses its
algorithms which is a term we all hear
all the time in all of the algorithm is
it’s a set of steps that we use that a
machine uses to solve a problem and it
tries to find patterns and then predict
what the right answer will be with the
best statistical probability and it’s
really simple the more data you have the
better statistical predictions and so
you have better results we have three
data types because we’re over
overwhelmed with big data and then three
types are structured data which I think
all of you are familiar with structured
data is like numbers or
what goes into a table unstructured data
those are things like the words we say
or the text and an email or a blog post
pictures unstructured data videos or
unstructured data and then we have semi
structured data and that’s kind of in
between the two of those so the number
of followers you may have on Twitter as
an example of semi structured data these
days because computers are faster and
faster and they have more processing
power and we can store vast amounts of
information in the cloud not these
clouds up there but that cloud we’re
able to interpret and process more
things so beyond tables we can interpret
talk which is called natural language
processing and also what we see which is
called machine learning but we’re
getting into jargon and I want to step
back and look at something simple
something you all know which is search
now search is the number one thing we do
online in fact 92% of all our
experiences begin with search Google
uses something I mean you’re not
surprised that it uses AI to power its
semantic algorithm and the semantic
algorithm understands the relationship
between words so it knows for example
that John Smith is probably a person and
it’s conversational interface not only
understands that John Smith is a person
but it understands the relationship
between that name and personal pronouns
that describe it you can try this at
home
afterwards open up your browser open up
Chrome click on the microphone there and
ask it a question say who is the Prime
Minister of Canada pretty easy question
we all know the answer to that and it
will say the Prime Minister of Canada is
Justin Trudeau then open up a new tab
and say who is he married to not who is
Justin Trudeau marry to but who is he
married to and it will say he is married
to Sophie Gregoire Trudeau then open up
a third tab and say how many children do
they have and it says how many children
they have and you can do this for maybe
one or two more tabs before Google
breaks down
that’s pretty great in a lot of ways
because Google understands our context
right but it’s also a little scary
because Google understands what we want
it kind of gets us and in doing so it’s
starting to build trust we’re starting
to trust it now ai is only as good as
the data the data it’s fed so think of
data as an AI version of a great meal
and you can have a data set that’s
filled with junk food or with healthier
options now imagine if you have a data
set that’s made up of greasy burgers
fries and soft drinks versus one that’s
made up of organic salads which do you
think is better well the answer is it
depends on what you’re using it for so
if for example that greasy fried data
set is used to build an app that’s
trying to get you to to eat at fast food
restaurants that’s what you want to use
but the point is neither one of those is
complete because it’s not representative
of a wider population and where we start
to get into trouble with this is when it
comes down to things like facial
recognition because the quality of
predictions depend on the quality and
the variety of data and otherwise
there’s too many biases and errors and a
lack of data diversity is why Alexa and
Google assistant are having trouble
understanding requests in accented
English or why Amazon’s facial
recognition app matched 28 members of US
Congress to the mug shots of convicted
felons I know it’s pretty scary and
we’ve got to get this right because if
it’s a facial recognition application
innocent lives could be affected in
China right now what they’re doing is
they’re testing a I powered glasses that
will police will wear to identify
suspects on busy streets that sounds a
lot like a server
and stayed here in Toronto at the
University of Toronto researchers are
trying to counteract that and what
they’re doing is they’re using something
called adversarial learning to teach the
AI to identify the image of something
say a turtle and then adjust a few
pixels in that to trick its algorithm
into identifying it as something else
and what they’ve done is they got an AI
to identify a picture of a turtle as a
picture of a gun now if you show human
beings those same two photos someone
that was adjusted and the first one we
can’t tell the difference but you’d see
how the battle grounds between
surveillance States and facial
recognition and facial fooling are
starting to take shape
MIT came out with an interesting
experiment where they used a dataset of
psychopathic personalities to create
something called Norman Norman is a
deviant AI and it’s named after Norman
Bates from psycho and if you see Norman
online don’t book a hotel but beyond all
these statistical models because Norman
and all of the other predictive
technologies they’re using some
scientists are moving into what’s called
embodied learning what is embodied
learning embodied learning goes outside
of the lab and takes in an unpredictable
environment like the world around us
company in Silicon Valley open AI
they’re teaching a robotic arm to grip
objects of various shapes sizes and
weights in much the same way that we
learn to grip facebook created two BOTS
they created a tourist spot and a guide
bot and they had the tourist spot asked
the guide bot questions to find out if
it could get around New York
successfully and it did that I’m not
sure how many of you have heard of
Google duplex that’s Google’s
voice-activated application so Google
duplex calls businesses hair salons and
restaurants to either book reservations
or make an appointment
and it sounds really real complete with
pauses and speech disfluencies those are
those hums and ahhs that all of us
associate with humans in fact the people
at the other end didn’t know they were
talking to a machine
now Google after the excitement caught a
lot of flack for that and in its second
iteration Google duplex had to identify
itself as a digital voice assistant how
many of you have seen Sophia the robot
so Sophia the robot she it does a lot of
publicity interviews with media has been
on a date with Will Smith and kind of
seems real but Sophia is not sentient
even though it says it has feelings it’s
just programmed to respond to a whole
bunch of different commands so for
example if you asked Sophia question
what Sophia will do is and I’m not in
recognition Lille and give it a little
smile just like us so we think it’s
human the head of facebook’s AI said
Sophia is really just smoke and mirrors
but Sophia does give us a glimpse into
our future all of these examples Sophia
duplex the bots I mentioned from
Facebook those are examples of
artificial narrow intelligence single
purpose AI and what that means is it’s
smart but it’s really good at doing just
one thing now a lot of us when we think
of AI we don’t think of single purpose
because that’s not really that scary
what we think about is something called
artificial general intelligence and
that’s really what’s depicted in a lot
of movies and fiction and what is a GI
artificial general intelligence it’s
when machines can do a whole bunch of
different things not just one goal and
they start to be more like us and and
the question is will they become
conscious and you know nobody knows that
the answer to that now there is a
start-up in Silicon Valley that’s
teaching machines not just to make
predictions but to understand conceptual
thinking and reason so it’s starting to
approach artificial general intelligence
in a different way are we going to merge
with machines
maybe Elon Musk talks about a neural
lace that is a connection between our
bodies and brains and computers that
could hopefully provide for better lives
and once it’s not here but it’s not that
far away if you think about pacemakers
and cochlear implants and as machines
start to sound like us more and more and
in some cases look like us and give us
answers quicker than other people do we
could start to trust BOTS more than
people and now think about it if you’re
in your car and a passenger contradicts
your Google Maps directions who are you
gonna believe that begs the question how
can we coexist in a world with machines
and and what types of relationships are
we gonna have with those machines it’s
really the subject of some research I’m
doing right now because has a potential
to make life really really amazing or
really really bad so for example do we
teach children to be polite to machines
when they’re asking a request do we say
hey do you need to say please and thank
you and if we do that are we raising a
generation that can’t tell the
difference between say people and a
blender for example now right now Google
assistant and Alexa both have settings
that reward kids when they’re polite
asking them questions so you have a
situation of BOTS disciplining kids do
we trust them to do that and how is this
going to affect all of our interpersonal
relationships there’s also the manner of
work and all the disruption we are going
to face in the next maybe five to ten
years where so many jobs are displaced
by machines and it’s not just the jobs
you may think driving trucks or in
factories its educated work so marketing
lawyers accountants radiologists
how are we going to retrain people is
that a job for government is it a job
for education is it a job for the
private sector or for all three it’s
something we need to think about max
tegmark is a professor at MIT and he
wrote a fabulous and fascinating an end
kind of scary book called life 3.0 and
in it he poses some very challenging
questions so he asks and something to
consider is it ethical to enslave a
superintelligence so a sentient machine
to do our bidding and what happens if
that machine wakes up even more changes
its goals and becomes our master and
that’s why he says it’s so important
that when we talk about artificial
intelligence we think of it as something
called beneficial AI know what
beneficial AI does is it takes in our
morals our ethics our values and our
purpose and it asks those tough
questions so that we know that as AI
continues to be developed there will be
a place for people and he says we need
to build this into AI now because in the
future it will be too late but back to
the big jump I said everyone failed at
getting to the top of the castle but
that’s not true a little boy was able to
get to the top of the castle and he
figured out a really really novel way of
doing it how did he do it he got up
there by jumping up one step at a time
and I think that’s really the same
approach we need to take with artificial
intelligence because there is an amazing
potential out there to cure disease or
maybe spark creativity but the reverse
is also true and and we can’t just bury
our heads in the sand pretend it doesn’t
exist and then wake up in a few years to
a world we no longer recognize we need
to be smart we need to be critical and
we need to be open and we need to ask
how will a AI affect our work and life
and we need to keep learning about it
and what it can become
what we could become but we need to do
it one step at
time thank you [Applause]
Please follow and like us: