this is a perfectly lovely photo from San Marco square in Venice there’s
San Marco square in Venice there’s
San Marco square in Venice there’s something kind of curious going on here
something kind of curious going on here
something kind of curious going on here if we look a little bit closer
if we look a little bit closer
if we look a little bit closer everyone’s head appears to fall roughly
everyone’s head appears to fall roughly
everyone’s head appears to fall roughly along the same horizon line that’s odd
along the same horizon line that’s odd
along the same horizon line that’s odd not everyone in the world is the same
not everyone in the world is the same
not everyone in the world is the same height right this optical illusion is
height right this optical illusion is
height right this optical illusion is the results of the photographer being
the results of the photographer being
the results of the photographer being average height from the perspective of
average height from the perspective of
average height from the perspective of an outlier the world would look a lot
an outlier the world would look a lot
an outlier the world would look a lot less uniform this is a talk about the
less uniform this is a talk about the
less uniform this is a talk about the role of humans in machine learning but
role of humans in machine learning but
role of humans in machine learning but really this is a talk about the role of
really this is a talk about the role of
really this is a talk about the role of humans in decision-making I’m a designer
humans in decision-making I’m a designer
humans in decision-making I’m a designer at Google I work on products powered by
at Google I work on products powered by
at Google I work on products powered by artificial intelligence and for the last
artificial intelligence and for the last
artificial intelligence and for the last year I’ve also been helping to lead a
year I’ve also been helping to lead a
year I’ve also been helping to lead a company-wide effort to make fairness a
company-wide effort to make fairness a
company-wide effort to make fairness a core component of our machine learning
core component of our machine learning
core component of our machine learning process ok what do I mean by fairness
process ok what do I mean by fairness
process ok what do I mean by fairness well speaking personally it comes down
well speaking personally it comes down
well speaking personally it comes down to two fundamental beliefs I believe
to two fundamental beliefs I believe
to two fundamental beliefs I believe that an unfair product is one that
that an unfair product is one that
that an unfair product is one that doesn’t work for you because of
doesn’t work for you because of
doesn’t work for you because of something you can’t change about
something you can’t change about
something you can’t change about yourself because I believe that
yourself because I believe that
yourself because I believe that technology should work for everybody
technology should work for everybody
technology should work for everybody I know controversial opinion like
I know controversial opinion like
I know controversial opinion like motherhood and apple pie who is going to
motherhood and apple pie who is going to
motherhood and apple pie who is going to disagree with this but if we stake a
disagree with this but if we stake a
disagree with this but if we stake a step back from machine intelligence for
step back from machine intelligence for
step back from machine intelligence for a second and look at the history of
a second and look at the history of
a second and look at the history of traditional product design it’s littered
traditional product design it’s littered
traditional product design it’s littered with examples where people have made
with examples where people have made
with examples where people have made decisions that haven’t worked for
decisions that haven’t worked for
decisions that haven’t worked for everyone and instead have been in line
everyone and instead have been in line
everyone and instead have been in line with defaults the things that go without
with defaults the things that go without
with defaults the things that go without saying the obvious the status quo take
saying the obvious the status quo take
saying the obvious the status quo take for example the way that we do
for example the way that we do
for example the way that we do automotive safety testing it wasn’t
automotive safety testing it wasn’t
automotive safety testing it wasn’t until 2011 that
until 2011 that
until 2011 that was federally mandated in the United
was federally mandated in the United
was federally mandated in the United States for a female body type
States for a female body type
States for a female body type crash-test-dummy
crash-test-dummy
crash-test-dummy to be used in front impact collision
to be used in front impact collision
to be used in front impact collision testing as a result women were 47
testing as a result women were 47
testing as a result women were 47 percent more likely to be severely
percent more likely to be severely
percent more likely to be severely injured in a car accident even today
injured in a car accident even today
injured in a car accident even today those crash-test dummies only represent
those crash-test dummies only represent
those crash-test dummies only represent the fifth percentile of female height
the fifth percentile of female height
the fifth percentile of female height and weight and two out of three front
and weight and two out of three front
and weight and two out of three front impact tests don’t have them seated in
impact tests don’t have them seated in
impact tests don’t have them seated in the driver’s seat
these are surely cards they’re
these are surely cards they’re historically the standard for how
historically the standard for how
historically the standard for how photographers have performed color
photographers have performed color
photographers have performed color calibration the idea goes something like
calibration the idea goes something like
calibration the idea goes something like this
this
this you you take a photo you process it you
you you take a photo you process it you
you you take a photo you process it you try to match up all these colors after
try to match up all these colors after
try to match up all these colors after which in theory you’ll know you’re good
which in theory you’ll know you’re good
which in theory you’ll know you’re good to go but you may have noticed something
to go but you may have noticed something
to go but you may have noticed something a little bit of Miss on this slide up
a little bit of Miss on this slide up
a little bit of Miss on this slide up and so relatively recently shirley cards
and so relatively recently shirley cards
and so relatively recently shirley cards only featured white women now
only featured white women now
only featured white women now unfortunately change didn’t come as a
unfortunately change didn’t come as a
unfortunately change didn’t come as a result of recognizing how
result of recognizing how
result of recognizing how unrepresentative this process was or
unrepresentative this process was or
unrepresentative this process was or some of the obvious disparities and
some of the obvious disparities and
some of the obvious disparities and performance on the photo like the one on
performance on the photo like the one on
performance on the photo like the one on the left
the left
the left it came in 1995 with Kodak gold after
it came in 1995 with Kodak gold after
it came in 1995 with Kodak gold after wood furniture manufacturers and
wood furniture manufacturers and
wood furniture manufacturers and chocolate companies had complained to
chocolate companies had complained to
chocolate companies had complained to Kodak about the limited spectrum of
Kodak about the limited spectrum of
Kodak about the limited spectrum of brown in their magazine advertisements
then there’s the film industry and here
then there’s the film industry and here we actually start to see an example of
we actually start to see an example of
we actually start to see an example of how machine learning can help raise
how machine learning can help raise
how machine learning can help raise awareness about gaps in representation
awareness about gaps in representation
awareness about gaps in representation this is work that a colleague of mine
this is work that a colleague of mine
this is work that a colleague of mine hartwig adam worked on in collaboration
hartwig adam worked on in collaboration
hartwig adam worked on in collaboration with Geena Davis Institute for gender in
with Geena Davis Institute for gender in
with Geena Davis Institute for gender in media using a technique called
media using a technique called
media using a technique called classification they were able to analyze
classification they were able to analyze
classification they were able to analyze every single frame from the top 100
every single frame from the top 100
every single frame from the top 100 grossing films over each of the last
grossing films over each of the last
grossing films over each of the last three years in the United States and
three years in the United States and
three years in the United States and measure the frequency of appearance
measure the frequency of appearance
measure the frequency of appearance between women and men the result men are
between women and men the result men are
between women and men the result men are seen and heard from nearly twice as
seen and heard from nearly twice as
seen and heard from nearly twice as often as
often as
often as women okay so where are these
women okay so where are these
women okay so where are these disconnects coming from our car
disconnects coming from our car
disconnects coming from our car companies intentionally trying to hurt
companies intentionally trying to hurt
companies intentionally trying to hurt women was the decisions who color
women was the decisions who color
women was the decisions who color calibrate for light-skinned overtly
calibrate for light-skinned overtly
calibrate for light-skinned overtly racist our movie directors and producers
racist our movie directors and producers
racist our movie directors and producers and writers trying to marginalize the
and writers trying to marginalize the
and writers trying to marginalize the role of women on purpose I choose a
role of women on purpose I choose a
role of women on purpose I choose a different answer I believe that these
different answer I believe that these
different answer I believe that these were the behaviors of rational actors
were the behaviors of rational actors
were the behaviors of rational actors making what they perceived to be obvious
making what they perceived to be obvious
making what they perceived to be obvious decisions because they were in line with
decisions because they were in line with
decisions because they were in line with defaults I want to talk about one more
defaults I want to talk about one more
defaults I want to talk about one more example just to illustrate how deep this
example just to illustrate how deep this
example just to illustrate how deep this disconnect can run there was a study
disconnect can run there was a study
disconnect can run there was a study done out of Cornell University in the
done out of Cornell University in the
done out of Cornell University in the late 70s and reproduced a number of
late 70s and reproduced a number of
late 70s and reproduced a number of times since we’re participants observed
times since we’re participants observed
times since we’re participants observed the recording a video recording of an
the recording a video recording of an
the recording a video recording of an infant interacting with a number of
infant interacting with a number of
infant interacting with a number of different stimuli and then rated that
different stimuli and then rated that
different stimuli and then rated that infants emotional response
infants emotional response
infants emotional response now the catch was half the participants
now the catch was half the participants
now the catch was half the participants were told prior to watching that they
were told prior to watching that they
were told prior to watching that they were watching a girl and half that they
were watching a girl and half that they
were watching a girl and half that they are watching a boy now when the child
are watching a boy now when the child
are watching a boy now when the child interacted was something that made them
interacted was something that made them
interacted was something that made them laugh or smile people agreed across the
laugh or smile people agreed across the
laugh or smile people agreed across the board that that kids dominant emotion
board that that kids dominant emotion
board that that kids dominant emotion was joy it’s a happy kid but when the
was joy it’s a happy kid but when the
was joy it’s a happy kid but when the child interacted with something that was
child interacted with something that was
child interacted with something that was jarring like a buzzer going off or a
jarring like a buzzer going off or a
jarring like a buzzer going off or a jack-in-the-box popping out that’s where
jack-in-the-box popping out that’s where
jack-in-the-box popping out that’s where we see the split if they’d been told
we see the split if they’d been told
we see the split if they’d been told that they were watching a girl they said
that they were watching a girl they said
that they were watching a girl they said that her dominant emotion was fear and
that her dominant emotion was fear and
that her dominant emotion was fear and if they had been told that they were
if they had been told that they were
if they had been told that they were watching a boy they said his dominant
watching a boy they said his dominant
watching a boy they said his dominant emotion was anger and most of us would
emotion was anger and most of us would
emotion was anger and most of us would probably agree that it would be
probably agree that it would be
probably agree that it would be irrational behavior if we were to treat
irrational behavior if we were to treat
irrational behavior if we were to treat an angry person the same way we would
an angry person the same way we would
an angry person the same way we would treat a fearful one same child same
treat a fearful one same child same
treat a fearful one same child same response
response
response different perception and perception
different perception and perception
different perception and perception drives for truly every facet of machine
drives for truly every facet of machine
drives for truly every facet of machine learning but that maybe a little bit
learning but that maybe a little bit
learning but that maybe a little bit different than what you’re accustomed to
different than what you’re accustomed to
different than what you’re accustomed to hearing when the machine learning
hearing when the machine learning
hearing when the machine learning process is represented it’ll often look
process is represented it’ll often look
process is represented it’ll often look something like this a complex array of
something like this a complex array of
something like this a complex array of nodes all strung together in a
nodes all strung together in a
nodes all strung together in a sufficiently impressive looking decision
sufficiently impressive looking decision
sufficiently impressive looking decision tree but it’s important for us to unpack
tree but it’s important for us to unpack
tree but it’s important for us to unpack that a little bit and talk about what’s
that a little bit and talk about what’s
that a little bit and talk about what’s going on in the machine learning process
going on in the machine learning process
going on in the machine learning process okay so training data get collected and
okay so training data get collected and
okay so training data get collected and annotated those data get fed into models
annotated those data get fed into models
annotated those data get fed into models the models pick up on patterns in the
the models pick up on patterns in the
the models pick up on patterns in the data and as a result of finding those
data and as a result of finding those
data and as a result of finding those patterns predictions can get made it’s
patterns predictions can get made it’s
patterns predictions can get made it’s complex but it boils down to about is
complex but it boils down to about is
complex but it boils down to about is straightforward is that with those
straightforward is that with those
straightforward is that with those predictions in hand media can get ranked
predictions in hand media can get ranked
predictions in hand media can get ranked sorted generated filtered it’s easy to
sorted generated filtered it’s easy to
sorted generated filtered it’s easy to view this process as something that
view this process as something that
view this process as something that looks like an objective or a neutral
looks like an objective or a neutral
looks like an objective or a neutral pipeline the data flows in the results
pipeline the data flows in the results
pipeline the data flows in the results flow out but human judgment plays a role
flow out but human judgment plays a role
flow out but human judgment plays a role at every single step in the process we
at every single step in the process we
at every single step in the process we get to decide where the data come from
get to decide where the data come from
get to decide where the data come from and why we think they’re representative
and why we think they’re representative
and why we think they’re representative in the first place we get to design the
in the first place we get to design the
in the first place we get to design the models we get to determine what success
models we get to determine what success
models we get to determine what success looks like and we establish the
looks like and we establish the
looks like and we establish the evaluation criteria that’ll prove
evaluation criteria that’ll prove
evaluation criteria that’ll prove whether or not the models working as
whether or not the models working as
whether or not the models working as intended and we get affected by the
intended and we get affected by the
intended and we get affected by the results meaning this isn’t a pipeline
results meaning this isn’t a pipeline
results meaning this isn’t a pipeline it’s a feedback loop where the outputs
it’s a feedback loop where the outputs
it’s a feedback loop where the outputs of a system the media that we are
of a system the media that we are
of a system the media that we are exposed to affect the decisions that
exposed to affect the decisions that
exposed to affect the decisions that will make and the data that will draw
will make and the data that will draw
will make and the data that will draw from the next time we set out to build a
from the next time we set out to build a
from the next time we set out to build a model now given what we know about some
model now given what we know about some
model now given what we know about some of our gaps in perception
of our gaps in perception
of our gaps in perception our history of making less than ideal
our history of making less than ideal
our history of making less than ideal product decisions one of the conclusions
product decisions one of the conclusions
product decisions one of the conclusions people will often jump to is that we
people will often jump to is that we
people will often jump to is that we need to remove human judgment from the
need to remove human judgment from the
need to remove human judgment from the process as much as possible if we could
process as much as possible if we could
process as much as possible if we could just get rid of these pesky human
just get rid of these pesky human
just get rid of these pesky human emotions everything be cool last year a
emotions everything be cool last year a
emotions everything be cool last year a company called North Point developed
company called North Point developed
company called North Point developed software to predict the likelihood that
software to predict the likelihood that
software to predict the likelihood that an inmate would recidivate that they
an inmate would recidivate that they
an inmate would recidivate that they would reoffended parole but if all
would reoffended parole but if all
would reoffended parole but if all criteria were equal the singular
criteria were equal the singular
criteria were equal the singular difference of that inmates race
difference of that inmates race
difference of that inmates race dramatically boosted the risk score of
dramatically boosted the risk score of
dramatically boosted the risk score of black inmates over whites often by more
black inmates over whites often by more
black inmates over whites often by more than double now the company claimed that
than double now the company claimed that
than double now the company claimed that they didn’t use race as a feature in
they didn’t use race as a feature in
they didn’t use race as a feature in training their models but unfortunately
training their models but unfortunately
training their models but unfortunately this technique of data blindness
this technique of data blindness
this technique of data blindness actually makes things worse because
actually makes things worse because
actually makes things worse because given the nature of deep learning and
given the nature of deep learning and
given the nature of deep learning and the sheer amount of information that
the sheer amount of information that
the sheer amount of information that gets exposed to a model it’s all but
gets exposed to a model it’s all but
gets exposed to a model it’s all but guaranteed that proxy variables for race
guaranteed that proxy variables for race
guaranteed that proxy variables for race will be automatically discovered and
will be automatically discovered and
will be automatically discovered and learned the example in the middle comes
learned the example in the middle comes
learned the example in the middle comes from face app this is an app that
from face app this is an app that
from face app this is an app that allowed you to take a selfie and then
allowed you to take a selfie and then
allowed you to take a selfie and then turn on a hotness filter make you look
turn on a hotness filter make you look
turn on a hotness filter make you look hotter unfortunately in many cases it
hotter unfortunately in many cases it
hotter unfortunately in many cases it just made people look whiter this is
just made people look whiter this is
just made people look whiter this is what happens when you train a model
what happens when you train a model
what happens when you train a model using skewed data save as social media
using skewed data save as social media
using skewed data save as social media behaviors of predominantly white
behaviors of predominantly white
behaviors of predominantly white countries and then there’s the example
countries and then there’s the example
countries and then there’s the example on the right it stirred up quite a bit
on the right it stirred up quite a bit
on the right it stirred up quite a bit of controversy in the research science
of controversy in the research science
of controversy in the research science community earlier this year two
community earlier this year two
community earlier this year two researchers out of China claimed to have
researchers out of China claimed to have
researchers out of China claimed to have trained a model that using just a
trained a model that using just a
trained a model that using just a headshot photo of somebody could predict
headshot photo of somebody could predict
headshot photo of somebody could predict whether or not they were a criminal with
whether or not they were a criminal with
whether or not they were a criminal with almost 90 percent accuracy now there
almost 90 percent accuracy now there
almost 90 percent accuracy now there were a number of methodological issues
were a number of methodological issues
were a number of methodological issues with the approach they took a number of
with the approach they took a number of
with the approach they took a number of examples that they were drawing from in
examples that they were drawing from in
examples that they were drawing from in their dataset pretty
their dataset pretty
their dataset pretty limited they had no way to prove for
limited they had no way to prove for
limited they had no way to prove for sure that the people labeled as non
sure that the people labeled as non
sure that the people labeled as non criminals had never actually committed a
criminals had never actually committed a
criminals had never actually committed a crime and just didn’t get caught but at
crime and just didn’t get caught but at
crime and just didn’t get caught but at a deeper level people aren’t born
a deeper level people aren’t born
a deeper level people aren’t born criminals so if there are any patterns
criminals so if there are any patterns
criminals so if there are any patterns being discovered here they’re not about
being discovered here they’re not about
being discovered here they’re not about the people being judged but instead
the people being judged but instead
the people being judged but instead maybe about those doing the judging
clearly trying to remove ourselves from
clearly trying to remove ourselves from the process and trust in the myth of
the process and trust in the myth of
the process and trust in the myth of neutral technology isn’t a viable path
neutral technology isn’t a viable path
neutral technology isn’t a viable path forward I propose an alternative I
forward I propose an alternative I
forward I propose an alternative I propose that we make machine learning
propose that we make machine learning
propose that we make machine learning intentionally human-centered and that we
intentionally human-centered and that we
intentionally human-centered and that we intervene for fairness and that means
intervene for fairness and that means
intervene for fairness and that means that there’s no silver bullet fairness
that there’s no silver bullet fairness
that there’s no silver bullet fairness is contextual and uniquely experienced
is contextual and uniquely experienced
is contextual and uniquely experienced by every individual so when we set out
by every individual so when we set out
by every individual so when we set out to build technology for humans using
to build technology for humans using
to build technology for humans using data collected about humans we have to
data collected about humans we have to
data collected about humans we have to be accountable and plan for what we’re
be accountable and plan for what we’re
be accountable and plan for what we’re gonna do when not if a bias manifests in
gonna do when not if a bias manifests in
gonna do when not if a bias manifests in some unwanted way I want to close by
some unwanted way I want to close by
some unwanted way I want to close by offering two pieces of advice for this
offering two pieces of advice for this
offering two pieces of advice for this road ahead the first is to be skeptical
road ahead the first is to be skeptical
road ahead the first is to be skeptical we joke internally when we run focus
we joke internally when we run focus
we joke internally when we run focus groups that it’s usually like a matter
groups that it’s usually like a matter
groups that it’s usually like a matter of seconds before someone mentions
of seconds before someone mentions
of seconds before someone mentions Skynet or the Terminator in the context
Skynet or the Terminator in the context
Skynet or the Terminator in the context of artificial intelligence like we’re
of artificial intelligence like we’re
of artificial intelligence like we’re all gonna go to sleep one night
all gonna go to sleep one night
all gonna go to sleep one night everything’s cool and then we wake up
everything’s cool and then we wake up
everything’s cool and then we wake up the next and the robots have legs and
the next and the robots have legs and
the next and the robots have legs and everything’s awful few things could be
everything’s awful few things could be
everything’s awful few things could be further from the truth as we’ve talked
further from the truth as we’ve talked
further from the truth as we’ve talked about today it will be human decisions
about today it will be human decisions
about today it will be human decisions yesterday today and tomorrow that shaped
yesterday today and tomorrow that shaped
yesterday today and tomorrow that shaped the future of technology we have to be
the future of technology we have to be
the future of technology we have to be skeptical of these myths that we tell
skeptical of these myths that we tell
skeptical of these myths that we tell ourselves and each other
ourselves and each other
ourselves and each other and I’m not just talking about science
and I’m not just talking about science
and I’m not just talking about science fiction we tell our so
fiction we tell our so
fiction we tell our so the myth that justice is blind and yet
the myth that justice is blind and yet
the myth that justice is blind and yet black Americans are arrested at four
black Americans are arrested at four
black Americans are arrested at four times the rate of white Americans on
times the rate of white Americans on
times the rate of white Americans on drug-related charges despite equal rates
drug-related charges despite equal rates
drug-related charges despite equal rates of use we tell ourselves the myth of the
of use we tell ourselves the myth of the
of use we tell ourselves the myth of the SATs and yet they’re not actually
SATs and yet they’re not actually
SATs and yet they’re not actually predictive of academic achievement we
predictive of academic achievement we
predictive of academic achievement we tell ourselves the myth of personalized
tell ourselves the myth of personalized
tell ourselves the myth of personalized medicine yet the vast majority of
medicine yet the vast majority of
medicine yet the vast majority of experimental participants are white and
experimental participants are white and
experimental participants are white and male and we tell ourselves about the
male and we tell ourselves about the
male and we tell ourselves about the myth of the wisdom of the crowd take for
myth of the wisdom of the crowd take for
myth of the wisdom of the crowd take for example this game that we developed
example this game that we developed
example this game that we developed recently called quick-draw out of
recently called quick-draw out of
recently called quick-draw out of google’s AI experiments program the idea
google’s AI experiments program the idea
google’s AI experiments program the idea was that people it’ll pictures of
was that people it’ll pictures of
was that people it’ll pictures of everyday things and it trained a model
everyday things and it trained a model
everyday things and it trained a model about how to recognize them so when
about how to recognize them so when
about how to recognize them so when people were asked to draw pictures of
people were asked to draw pictures of
people were asked to draw pictures of shoes they mostly drew ones that looked
shoes they mostly drew ones that looked
shoes they mostly drew ones that looked like the one on the left so as more
like the one on the left so as more
like the one on the left so as more people interacted with the game it
people interacted with the game it
people interacted with the game it wasn’t able to recognize shoes that
wasn’t able to recognize shoes that
wasn’t able to recognize shoes that looked like the one on the right one
looked like the one on the right one
looked like the one on the right one size is almost never gonna fit all and
size is almost never gonna fit all and
size is almost never gonna fit all and my second piece of advice is to be
my second piece of advice is to be
my second piece of advice is to be curious and it starts with a question
curious and it starts with a question
curious and it starts with a question Who am I and what will my data teach me
Who am I and what will my data teach me
Who am I and what will my data teach me about everybody else
about everybody else
about everybody else I have many traits that would be
I have many traits that would be
I have many traits that would be considered defaults in the tech industry
considered defaults in the tech industry
considered defaults in the tech industry I’m white I’m a cisgendered man I have a
I’m white I’m a cisgendered man I have a
I’m white I’m a cisgendered man I have a partner of the opposite sex there are
partner of the opposite sex there are
partner of the opposite sex there are things about me that are outside of the
things about me that are outside of the
things about me that are outside of the norm – I am way too tall I’m actually
norm – I am way too tall I’m actually
norm – I am way too tall I’m actually legally blind and I was raised in a
legally blind and I was raised in a
legally blind and I was raised in a cultural minority in Judaism we’re all
cultural minority in Judaism we’re all
cultural minority in Judaism we’re all complex individuals and our traits don’t
complex individuals and our traits don’t
complex individuals and our traits don’t define us but it would be foolish to
define us but it would be foolish to
define us but it would be foolish to believe that other people don’t notice
believe that other people don’t notice
believe that other people don’t notice them by taking this uncomfortable step
them by taking this uncomfortable step
them by taking this uncomfortable step of inventory our traits physical social
of inventory our traits physical social
of inventory our traits physical social cognitive and otherwise we’re getting
cognitive and otherwise we’re getting
cognitive and otherwise we’re getting approximate to people who are reminded
approximate to people who are reminded
approximate to people who are reminded of their other nests every time they
of their other nests every time they
of their other nests every time they interact with a default on a day-to-day
interact with a default on a day-to-day
interact with a default on a day-to-day basis for many people this kind of an
basis for many people this kind of an
basis for many people this kind of an inventory isn’t optional and I hope that
inventory isn’t optional and I hope that
inventory isn’t optional and I hope that by doing this we can start to see the
by doing this we can start to see the
by doing this we can start to see the world as less binary less people like me
world as less binary less people like me
world as less binary less people like me people not like me and more like this as
people not like me and more like this as
people not like me and more like this as a world of intersectionality where we
a world of intersectionality where we
a world of intersectionality where we celebrate our differences as
celebrate our differences as
celebrate our differences as opportunities to deepen our
opportunities to deepen our
opportunities to deepen our understanding about the world and each
understanding about the world and each
understanding about the world and each other and in the process to harness
other and in the process to harness
other and in the process to harness machine learning to its full capability
machine learning to its full capability
machine learning to its full capability as a tool to help us seek out patterns
as a tool to help us seek out patterns
as a tool to help us seek out patterns in the world especially at the margins
in the world especially at the margins
in the world especially at the margins to find ways to enrich human connection
to find ways to enrich human connection
to find ways to enrich human connection and augment human capability a tool for
and augment human capability a tool for
and augment human capability a tool for explorers to help us see beyond the
explorers to help us see beyond the
explorers to help us see beyond the defaults thank you
”
Be First to Comment