so who’s excited about AI yeah oh yeah
who’s nervous about it okay I see a mix
I feel both and there’s good reason
AI is a complex and controversial topic
and we’re hearing all sorts of
conflicting messages killer robots are
gonna take all our jobs and murder us
and I will cure cancer and I sure in
Utopia
this is just marketing height then AI is
nothing but a buzzword there is a kernel
of truth in all these statements but the
reality is more complex and nuanced it’s
natural to feel like if you’re not a
math genius or you don’t have a PhD from
Stanford that you couldn’t possibly hope
to understand what’s going on much less
to get involved I’m here to tell you
that is false it is not only possible
for everyone to get involved with AI it
is actually crucial that you do so I say
this is someone who is an unlikely
person to become an AI researcher and we
need more unlikely people in the field
both to address the harms that are being
caused as well as to take full advantage
of the positive opportunities but first
what is ai ai refers to any technique
that allows a computer to appear to be
acting with some level of intelligence
ai is in products like how Google photos
automatically organizes your picture
collection here it’s grouped together
temples skyline and food into separate
categories ai is in Skype translator
which lets people speaking different
languages communicate in real-time these
are both examples of a particular type
of AI called deep learning in the past
couple of years deep learning has
achieved better than human results on
tasks that we used to think only humans
could do like recognizing pictures and
understanding speech one misconception
about AI researchers is that we’re all
trying to recreate a human brain or to
get a computer to achieve human
consciousness many of us are not many of
us including me are interested in
practical solutions to real-world
problems
like identifying lung cancer on a CT
scan I’ll talk more about some of these
positive applications in a moment but
first I wanna address three aspects of
AI that worry me the first is that AI is
encoding and magnifying human bias
computer vision has repeatedly performed
worse on people of color research
earlier this year from joy ballon weeny
and Tim Nick gabru evaluated computer
vision products from IBM Microsoft and
elsewhere and they found that these
products performed worse on women than
on men and were some people with dark
skin compared to people with light skin
for instance
IBM’s computer vision software was 99.7%
accurate on light skinned men and only
65% accurate on dark skinned women
that’s ridiculous that’s a huge
difference 34% bias is also being
encoded in language and translation
tools like Google Translate if you give
Google Translate pair of English
sentences she is a doctor he is a nurse
translate those into Turkish which has a
gender-neutral singular pronoun and then
back into English they come back with
the genders flip to fit the stereotype
he is a doctor she is a nurse and this
is just one example out of many biases
second area that worries me is that AI
is promoting and incentivizing
conspiracy theories people watch over 1
billion hours of YouTube per day that’s
per day that’s a–that’s huge and part
of YouTube’s success is because of its
recommendation algorithm so when you’re
watching YouTube there’s a side bar that
says up next and has several videos
selected by an algorithm these
recommendations disproportionately
include conspiracy theorists and white
supremacists this phenomena has been
investigated by The Wall Street Journal
and The Guardian and it’s been written
about in the New York oh sorry it’s been
written about and then you are
time’s so what’s going on here this is
not an intentional evil plot
but YouTube’s algorithm is trying to
maximize how much time people spend
watching YouTube because they earn more
money that way and conspiracy theorists
watch way more YouTube than people that
trust a variety of media sources time we
spend on other media sources is bad for
YouTube’s bottom line so Google which
which owns YouTube is earning a lot of
money while pumping pollution into the
rest of society in this case that
pollutions in the form of the political
and social impact of having lots of
people who believe conspiracy theories
and distrust mainstream media sources in
addition to even egging on conspiracy
theorists to come out with more extreme
material because that’s what that’s
what’s rewarded the problems posed by AI
can often feel very new but in this case
this problem of what to do when a
company’s profiting while offsetting its
cost everyone else not something that
economists and public policy experts
have been looking at for ages and we
could use their help here and before I
explain the third area let’s start by
watching this video so you’ll see some
green and red rectangles on the left and
those are identifying people’s heads and
faces even though this video is at a
distance and there are people are in a
crowd I find this pretty creepy AI can
be used by authoritarian governments for
surveillance so currently in China
million people who are members of an
ethnic minority are in internment camps
and surveillance has helped facilitate
that this said this video is from a
Chinese company called deep glint it’s
two years old the technology’s only
continue to advance since then last year
there was research on how to identify
protesters in a crowd even if they’re
wearing hats or scarves to try to
conceal their identities and this is not
just a problem abroad in the United
States earlier this year we learned that
Palantir which is a tech company founded
by billionaire and Trump donor Peter
Palantir had been using New Orleans as a
testbed for its predictive policing
technology for the last six years this
program was so secretive that even City
Council members didn’t know about it
much less have any oversight Amazon is
already selling facial recognition
software to police and the software is
in use by police in Orlando and Oregon
even though the ACLU is raising concerns
about the racial bias and inaccuracy of
Amazon software I think the combination
of a total lack of accountability
transparency or oversight together with
racial bias and existing police data
that will be used to train these
algorithms together with the repeated
failure of computer vision on people of
color makes this a really frightening
combination so part of the reason that
we’re seeing so many unsettling and even
scary applications of AI is because of
how incredibly narrow the group of
people that’s creating it is and how
many of us are not represented by that
group only 12% of machine learning
researchers are women the numbers are
similarly dire when it comes to race geo
diversity and background on top of this
people in tack are often building AI
without any sort of input or
collaboration from fields like sociology
psychology or history that can help us
better understand how Tech and humans
interact I’ve personally experienced how
toxic the environment can be in academic
STEM fields and in the tech industry and
it scares me that the same people
responsible for creating these toxic
environments are the primary ones
creating powerful AI technology I am an
AI researcher but I’m a very unlikely
person to have become an AI researcher
and my goal is to help more unlikely
people find their way into the field I
felt like an outsider my entire life I
could never find the community I was
looking for not in academia and not in
tact so I eventually decided to create
that community for myself for other
outside
for people who have been excluded or who
can’t afford the resources I grew up on
the Gulf Coast of Texas near a cluster
of chemical refineries I attended a poor
predominantly black public high school
that was later ranked in the bottom two
percent of Texas schools I was pepper
sprayed by police while I was at school
and there are some great students and
teachers at my school but even as a
teenager was clear we did not have the
resources or the opportunities that
wealthier schools had and we were not
treated the same unlikely as it was I
completed my PhD in math I did not fit
in like most math PhD programs my
program was almost entirely male I
experienced sexism harassment and
isolation
I watched several of my female friends
drop out due to the culture my thesis
advisor even told me I was too feminine
to be successful and he later stopped
meeting with me it was it was
devastating years later I sold my car my
furniture and most of what I owned to
move to San Francisco with just two
suitcases and a new dream of becoming a
data scientist I did not know anybody
who worked in tech it was not a career
option I’d really been exposed to and I
feel like people in tech were speaking a
foreign language
eventually I achieved my goal and I
landed a job as a data scientist and
software engineer at an up-and-coming
startup that would later go on to become
a multi-billion dollar company and a
household name everyone kept telling me
I’d made it but I was miserable
once again I did not fit in the
environment was aggressive sexist
isolating and I dreaded going to work
each morning during the same period of
time I first got interested in deep
learning in 2012 I read a New York Times
article about an academic group that had
no expert knowledge of biochemistry but
was using deep learning to automatically
design new pharmaceutical drugs
is exactly the type of practical AI I’m
interested in as I told friends and
acquaintances what I was learning many
people told me that they could never
understand AI that they weren’t smart
enough or their brain wasn’t wired the
right way these are all myths this is
completely false for a number of reasons
as I got deeper into the field and
learned the practical techniques needed
to create state-of-the-art models I saw
that in most cases the math theory
wasn’t even necessary to become a
working practitioner I also saw that the
field was unnecessarily exclusive in
2013 I went to a talk that a star in the
field was giving and I asked him a
simple practical question during the Q&A
he not only didn’t answer my question he
told me that nobody who knew the answer
so people like him nobody was writing it
down if you weren’t part of the inside
crowd tough luck but it doesn’t have to
be like that for much of my life I
thought it was a negative that I didn’t
and couldn’t fit in I sometimes even
wish there could be a different person
so I could better fit into the toxic
environments I was in but it turns out
that weasely don’t fit in or valuable
they allowed me to see a gap that the
current system and academia and in tag
and in the AI community isn’t built for
people like me it’s not built for most
of us my different background allowed me
to create something different two years
ago I partnered with Jeremy Howard who’s
a philosophy major with no graduate
degree and no formal machine learning
training and we decided to create a
course to teach state of the art deep
learning with no advanced math
prerequisites and we made this available
completely for free our motto is making
neural networks uncool because being
cool is about being exclusive and that’s
the opposite of what I want we’ve had 10
so most people who take our course are
working professionals doing this in
their spare time and we’ve had tens of
thousands of people take the class and
apply deep learning to problems they
care about whether this is at nonprofits
or our fortune 500 companies launching
startups or winning half as
I’m a despite all these success stories
I continue to hear from experts saying
that what we’re doing is impossible
we’re disproving the myth that you need
the fanciest most expensive computer and
the fanciest most prestigious background
to do cutting-edge work earlier this
year a group of our part-time students
entered a competition against engineers
from Google and Stanford artists like
Google and Intel and these teams at
Google and Intel had access to way more
expensive computers yet our team won
theirs thank you yeah I’m so proud of
them for this this was covered in the
verge and the MIT tech review and
they’re showing that you don’t have to
work at Google or have Google like
resources to do state-of-the-art work
one person who’s taken our courses
Melissa Fabrice Melissa was an English
literature PhD student focused on
American poetry before switching careers
a few years ago she’s now an engineer at
the microlending nonprofit Kiva which
lets people make small loans to
entrepreneurs around the world
Melissa discovered that computer vision
software couldn’t identify pictures of
Kiva users because the software was
primarily trained by a non-white people
Melissa won a prestigious grant to help
fund her project to address this another
person who’s taken our class is Sarah
hooker Sarah was an economics major who
only learned to code a few years before
the class Sarah started Delta analytics
to pair data scientists with nonprofits
Sarah and a team worked with rainforests
connection which puts recycled cell
phones up on trees and endangered
rainforest streams audio and then uses
deep learning to identify chainsaw
noises if someone’s illegally cutting
down the forest they’re experts who say
our students will never get jobs in the
field yet Sarah now works for Google and
is helping to open Google’s first AI
Research Center in Africa these are just
a few stories out of many
other people who have taken our class
have helped farmers in India created
wearable devices for patients with
Parkinsons disease invented new styles
of artwork and music and even had their
work featured on HBO and in Forbes I was
contacted by a Canadian Canadian dairy
farmer who wants to use deep learning to
improve the health of his goats utters I
never would have guessed that AI could
be applied to go daughters but this is
exactly the type of person I want to
help I want to help other unlikely
people other outsiders people sometimes
ask me if it’s dangerous to make AI
accessible to more people what’s
dangerous is having an exclusive and
homogeneous group creating technology
that impacts us all we’ve seen the harms
being caused by companies like Facebook
Google YouTube Amazon Palantir and
others and we need people from more
diverse backgrounds to try to address
those harms we need people from
communities that have been
disproportionately targeted with
harassment to help create our technology
because they have a deeper understanding
of how tech can be weaponized against
the most vulnerable and of what
safeguards we need to put in place and
we also need more people who are experts
about people about human behavior human
psychology and human history the reason
I want everyone involved with AI is that
you know about problems nobody else
knows about and you have skills in a
background that nobody else has I didn’t
know about using old cell phones to
protect the rainforest about the suicide
rate for farmers in India or about the
health of NGO daughters yet these are
all problems my students have worked on
I’m so glad that my students didn’t
believe the myth that AI is only for
math prodigies but that they knew their
we need journalists Boyer’s hospital
administrators hotel sales managers
sociology
CEOs historians and more to understand
the capabilities of AI it doesn’t matter
if you didn’t go to the right school if
math makes you anxious if your friends
and colleagues think you’re the last
person that ever expects be working with
AI AI needs more unlikely people and the
world needs you involved with AI thank
you [Applause]
who’s nervous about it okay I see a mix
I feel both and there’s good reason
AI is a complex and controversial topic
and we’re hearing all sorts of
conflicting messages killer robots are
gonna take all our jobs and murder us
and I will cure cancer and I sure in
Utopia
this is just marketing height then AI is
nothing but a buzzword there is a kernel
of truth in all these statements but the
reality is more complex and nuanced it’s
natural to feel like if you’re not a
math genius or you don’t have a PhD from
Stanford that you couldn’t possibly hope
to understand what’s going on much less
to get involved I’m here to tell you
that is false it is not only possible
for everyone to get involved with AI it
is actually crucial that you do so I say
this is someone who is an unlikely
person to become an AI researcher and we
need more unlikely people in the field
both to address the harms that are being
caused as well as to take full advantage
of the positive opportunities but first
what is ai ai refers to any technique
that allows a computer to appear to be
acting with some level of intelligence
ai is in products like how Google photos
automatically organizes your picture
collection here it’s grouped together
temples skyline and food into separate
categories ai is in Skype translator
which lets people speaking different
languages communicate in real-time these
are both examples of a particular type
of AI called deep learning in the past
couple of years deep learning has
achieved better than human results on
tasks that we used to think only humans
could do like recognizing pictures and
understanding speech one misconception
about AI researchers is that we’re all
trying to recreate a human brain or to
get a computer to achieve human
consciousness many of us are not many of
us including me are interested in
practical solutions to real-world
problems
like identifying lung cancer on a CT
scan I’ll talk more about some of these
positive applications in a moment but
first I wanna address three aspects of
AI that worry me the first is that AI is
encoding and magnifying human bias
computer vision has repeatedly performed
worse on people of color research
earlier this year from joy ballon weeny
and Tim Nick gabru evaluated computer
vision products from IBM Microsoft and
elsewhere and they found that these
products performed worse on women than
on men and were some people with dark
skin compared to people with light skin
for instance
IBM’s computer vision software was 99.7%
accurate on light skinned men and only
65% accurate on dark skinned women
that’s ridiculous that’s a huge
difference 34% bias is also being
encoded in language and translation
tools like Google Translate if you give
Google Translate pair of English
sentences she is a doctor he is a nurse
translate those into Turkish which has a
gender-neutral singular pronoun and then
back into English they come back with
the genders flip to fit the stereotype
he is a doctor she is a nurse and this
is just one example out of many biases
second area that worries me is that AI
is promoting and incentivizing
conspiracy theories people watch over 1
billion hours of YouTube per day that’s
per day that’s a–that’s huge and part
of YouTube’s success is because of its
recommendation algorithm so when you’re
watching YouTube there’s a side bar that
says up next and has several videos
selected by an algorithm these
recommendations disproportionately
include conspiracy theorists and white
supremacists this phenomena has been
investigated by The Wall Street Journal
and The Guardian and it’s been written
about in the New York oh sorry it’s been
written about and then you are
time’s so what’s going on here this is
not an intentional evil plot
but YouTube’s algorithm is trying to
maximize how much time people spend
watching YouTube because they earn more
money that way and conspiracy theorists
watch way more YouTube than people that
trust a variety of media sources time we
spend on other media sources is bad for
YouTube’s bottom line so Google which
which owns YouTube is earning a lot of
money while pumping pollution into the
rest of society in this case that
pollutions in the form of the political
and social impact of having lots of
people who believe conspiracy theories
and distrust mainstream media sources in
addition to even egging on conspiracy
theorists to come out with more extreme
material because that’s what that’s
what’s rewarded the problems posed by AI
can often feel very new but in this case
this problem of what to do when a
company’s profiting while offsetting its
cost everyone else not something that
economists and public policy experts
have been looking at for ages and we
could use their help here and before I
explain the third area let’s start by
watching this video so you’ll see some
green and red rectangles on the left and
those are identifying people’s heads and
faces even though this video is at a
distance and there are people are in a
crowd I find this pretty creepy AI can
be used by authoritarian governments for
surveillance so currently in China
million people who are members of an
ethnic minority are in internment camps
and surveillance has helped facilitate
that this said this video is from a
Chinese company called deep glint it’s
two years old the technology’s only
continue to advance since then last year
there was research on how to identify
protesters in a crowd even if they’re
wearing hats or scarves to try to
conceal their identities and this is not
just a problem abroad in the United
States earlier this year we learned that
Palantir which is a tech company founded
by billionaire and Trump donor Peter
Palantir had been using New Orleans as a
testbed for its predictive policing
technology for the last six years this
program was so secretive that even City
Council members didn’t know about it
much less have any oversight Amazon is
already selling facial recognition
software to police and the software is
in use by police in Orlando and Oregon
even though the ACLU is raising concerns
about the racial bias and inaccuracy of
Amazon software I think the combination
of a total lack of accountability
transparency or oversight together with
racial bias and existing police data
that will be used to train these
algorithms together with the repeated
failure of computer vision on people of
color makes this a really frightening
combination so part of the reason that
we’re seeing so many unsettling and even
scary applications of AI is because of
how incredibly narrow the group of
people that’s creating it is and how
many of us are not represented by that
group only 12% of machine learning
researchers are women the numbers are
similarly dire when it comes to race geo
diversity and background on top of this
people in tack are often building AI
without any sort of input or
collaboration from fields like sociology
psychology or history that can help us
better understand how Tech and humans
interact I’ve personally experienced how
toxic the environment can be in academic
STEM fields and in the tech industry and
it scares me that the same people
responsible for creating these toxic
environments are the primary ones
creating powerful AI technology I am an
AI researcher but I’m a very unlikely
person to have become an AI researcher
and my goal is to help more unlikely
people find their way into the field I
felt like an outsider my entire life I
could never find the community I was
looking for not in academia and not in
tact so I eventually decided to create
that community for myself for other
outside
for people who have been excluded or who
can’t afford the resources I grew up on
the Gulf Coast of Texas near a cluster
of chemical refineries I attended a poor
predominantly black public high school
that was later ranked in the bottom two
percent of Texas schools I was pepper
sprayed by police while I was at school
and there are some great students and
teachers at my school but even as a
teenager was clear we did not have the
resources or the opportunities that
wealthier schools had and we were not
treated the same unlikely as it was I
completed my PhD in math I did not fit
in like most math PhD programs my
program was almost entirely male I
experienced sexism harassment and
isolation
I watched several of my female friends
drop out due to the culture my thesis
advisor even told me I was too feminine
to be successful and he later stopped
meeting with me it was it was
devastating years later I sold my car my
furniture and most of what I owned to
move to San Francisco with just two
suitcases and a new dream of becoming a
data scientist I did not know anybody
who worked in tech it was not a career
option I’d really been exposed to and I
feel like people in tech were speaking a
foreign language
eventually I achieved my goal and I
landed a job as a data scientist and
software engineer at an up-and-coming
startup that would later go on to become
a multi-billion dollar company and a
household name everyone kept telling me
I’d made it but I was miserable
once again I did not fit in the
environment was aggressive sexist
isolating and I dreaded going to work
each morning during the same period of
time I first got interested in deep
learning in 2012 I read a New York Times
article about an academic group that had
no expert knowledge of biochemistry but
was using deep learning to automatically
design new pharmaceutical drugs
is exactly the type of practical AI I’m
interested in as I told friends and
acquaintances what I was learning many
people told me that they could never
understand AI that they weren’t smart
enough or their brain wasn’t wired the
right way these are all myths this is
completely false for a number of reasons
as I got deeper into the field and
learned the practical techniques needed
to create state-of-the-art models I saw
that in most cases the math theory
wasn’t even necessary to become a
working practitioner I also saw that the
field was unnecessarily exclusive in
2013 I went to a talk that a star in the
field was giving and I asked him a
simple practical question during the Q&A
he not only didn’t answer my question he
told me that nobody who knew the answer
so people like him nobody was writing it
down if you weren’t part of the inside
crowd tough luck but it doesn’t have to
be like that for much of my life I
thought it was a negative that I didn’t
and couldn’t fit in I sometimes even
wish there could be a different person
so I could better fit into the toxic
environments I was in but it turns out
that weasely don’t fit in or valuable
they allowed me to see a gap that the
current system and academia and in tag
and in the AI community isn’t built for
people like me it’s not built for most
of us my different background allowed me
to create something different two years
ago I partnered with Jeremy Howard who’s
a philosophy major with no graduate
degree and no formal machine learning
training and we decided to create a
course to teach state of the art deep
learning with no advanced math
prerequisites and we made this available
completely for free our motto is making
neural networks uncool because being
cool is about being exclusive and that’s
the opposite of what I want we’ve had 10
so most people who take our course are
working professionals doing this in
their spare time and we’ve had tens of
thousands of people take the class and
apply deep learning to problems they
care about whether this is at nonprofits
or our fortune 500 companies launching
startups or winning half as
I’m a despite all these success stories
I continue to hear from experts saying
that what we’re doing is impossible
we’re disproving the myth that you need
the fanciest most expensive computer and
the fanciest most prestigious background
to do cutting-edge work earlier this
year a group of our part-time students
entered a competition against engineers
from Google and Stanford artists like
Google and Intel and these teams at
Google and Intel had access to way more
expensive computers yet our team won
theirs thank you yeah I’m so proud of
them for this this was covered in the
verge and the MIT tech review and
they’re showing that you don’t have to
work at Google or have Google like
resources to do state-of-the-art work
one person who’s taken our courses
Melissa Fabrice Melissa was an English
literature PhD student focused on
American poetry before switching careers
a few years ago she’s now an engineer at
the microlending nonprofit Kiva which
lets people make small loans to
entrepreneurs around the world
Melissa discovered that computer vision
software couldn’t identify pictures of
Kiva users because the software was
primarily trained by a non-white people
Melissa won a prestigious grant to help
fund her project to address this another
person who’s taken our class is Sarah
hooker Sarah was an economics major who
only learned to code a few years before
the class Sarah started Delta analytics
to pair data scientists with nonprofits
Sarah and a team worked with rainforests
connection which puts recycled cell
phones up on trees and endangered
rainforest streams audio and then uses
deep learning to identify chainsaw
noises if someone’s illegally cutting
down the forest they’re experts who say
our students will never get jobs in the
field yet Sarah now works for Google and
is helping to open Google’s first AI
Research Center in Africa these are just
a few stories out of many
other people who have taken our class
have helped farmers in India created
wearable devices for patients with
Parkinsons disease invented new styles
of artwork and music and even had their
work featured on HBO and in Forbes I was
contacted by a Canadian Canadian dairy
farmer who wants to use deep learning to
improve the health of his goats utters I
never would have guessed that AI could
be applied to go daughters but this is
exactly the type of person I want to
help I want to help other unlikely
people other outsiders people sometimes
ask me if it’s dangerous to make AI
accessible to more people what’s
dangerous is having an exclusive and
homogeneous group creating technology
that impacts us all we’ve seen the harms
being caused by companies like Facebook
Google YouTube Amazon Palantir and
others and we need people from more
diverse backgrounds to try to address
those harms we need people from
communities that have been
disproportionately targeted with
harassment to help create our technology
because they have a deeper understanding
of how tech can be weaponized against
the most vulnerable and of what
safeguards we need to put in place and
we also need more people who are experts
about people about human behavior human
psychology and human history the reason
I want everyone involved with AI is that
you know about problems nobody else
knows about and you have skills in a
background that nobody else has I didn’t
know about using old cell phones to
protect the rainforest about the suicide
rate for farmers in India or about the
health of NGO daughters yet these are
all problems my students have worked on
I’m so glad that my students didn’t
believe the myth that AI is only for
math prodigies but that they knew their
we need journalists Boyer’s hospital
administrators hotel sales managers
sociology
CEOs historians and more to understand
the capabilities of AI it doesn’t matter
if you didn’t go to the right school if
math makes you anxious if your friends
and colleagues think you’re the last
person that ever expects be working with
AI AI needs more unlikely people and the
world needs you involved with AI thank
you [Applause]
Please follow and like us: