AI for Good is happening.
[Music]
do you remember
what you wanted to be when you were a
kid
when i was seven i remember and
announced with a lot of pride
to my favorite teacher that i wanted to
be a nun
she rolled high broke my heart when i
was 13
i was fascinated with wildlife my heroes
were jango dol
tayan fosei i must say i was bullied at
school
so i was kind of upset with these stupid
human beings
and yes that be the plan i’ll fly to
ronda
or kenya i’ll be a vet and i spend my
entire life
protecting gorillas lions and their
habitats
today i’m working in the ai industry
and i am a big ai enthusiast ai for a
good enthusiast
how did that happen i’m going to take
you on a little journey
from my teenage aspiration to today’s
exciting perspective
growing up i eventually reconciled
myself with homoserpas
but i was still very disturbed by your
capacity to destroy a planet
and practice unfair treatments racism
sexism homophobia
so i will be a lawyer i will work in the
united nations
and i will defend human rights and
environmental rights
but in the end i didn’t become a vet
i didn’t become a lawyer no none
obviously
i started to work in the tech industry
why
i had such big ideas but i was
undeniably
attracted to this industry of doers
fearless entrepreneurs that were
changing our habits in decades
bill gates steve jobs wow
maybe we could leverage tech to do
something good in the world
so i’ve started my career in software
companies
these were super exciting time facebook
was exploding it was the beginning of
the internet of things
but yet i couldn’t reconcile
impact and tech so
i was about to move to another industry
when i got the chance
to start working in an ai startup
and i was blown away by the power of
this technology
so what is artificial intelligence three
notions to have in mind
one ai is a branch of computer science
that enables machine
to perform tasks that usually require
human intelligence
recognizing images understanding
language
solving problem make decision
to machine learning machine learning is
one approach to reach artificial
intelligence
through algorithm that analyze data
learn from them
find patterns or make predictions about
something in the world
number three deep learning
or deep neural learning is one of the
latest techniques to implement machine
learning
using algorithm designed like our human
brain
that can process an infinite number of
data
in such a short time ai
is so powerful it can solve problems
and find solutions where we human beings
will never be able to find on our own
think about alpha ford alpha ford
is a deep learning program developed by
deepmind in 2020. and basically
they’re solving one of the biggest
challenge in biology
predicting the shape of protein
did you know that each protein has a
unique shape that determines
there are literally billions of proteins
in human
living things and an infinite number of
configurations possible
alpha fod is able to predict the shape
of a protein with the same level of
accuracy than nor
dan lab in a few days versus a few
years with conventional methods opening
the door
to new treatment for diseases and to new
synthetic protein
able to digest waste produce biofuel
or make our plants more nutritious
while i was discovering the fabulous
world of ai
the united nations were announcing that
we were falling behind with regard to
achieving the sustainable development
goals in 2030
the sdgs are 17 goals adapted by the
united nations in 2015
as a universal call to protect our
planet
and poverty and make sure people live in
peace and prosperity
by 2030 that was it
i could finally reconcile tech and
impact
convinced that we could leverage ai to
accelerate progress towards the sdgs
i was so excited that i’ve decided to
turn it into my dream job
spotting and helping entrepreneurs
researchers scientists that are using ai
to help deliver unsustainable
development goals
that’s why today i want to share with
you
two areas that resonate with my teenage
questioning but also my current concerns
and where ai can help tremendously
climate change and equity and human
rights
i will show you through a pragmatic
example how
we are more informed and more empowered
thanks uai to make the right decision
and make our world a better place
let’s start with climate change ai can
help us
with the energy transition but also with
our biggest natural ally
for rest you know
that more than 70 of our carbon emission
are coming from the energy sector
that’s why we are decarbonizing it and
to do so
we’re shifting toward renewable energy
solar wind
this transition is coming with very
complex challenges
fortunately electric systems are a rash
of data
so ai can help substantially for
instance
to produce renewable energy you depend
on solar
and with conditions so operators have to
work
with standby polluting planes to avoid
shortages
which means that basically depending on
the time using energy during the day
you may be using dirty energy
by using machine learning to better
forecast
how much power is generated by renewable
sources
and how much power demand there is we
are able to reduce
dramatically the reliance understand by
polluting plants
for instance the technon profit what
time
is developing a solution named automated
emissions reduction
basically what they do is that they
forecast the grid emission
intensity in your region then they
connect with your smart device
smart freight smart car to help them
adapt the timing of the use of energy
to synchronize with clean energy and
avoid dirty energy
potentially we’re talking about a 40
reduction of carbon emission in the
coming years with this technology
greenhouse emissions are not only coming
from power plants
they’re also coming from the massive
destruction
of forest forests that do capture
carbon through photosynthesis here again
ai can help a lot
it can help to plan reforestation
and to monitor restoration program
so to plan reforestation you need to
know where to plant the trees
the causer lab from ethgeric university
carried out a study in 2019 to assess
how many trees the world could support
where they should be grown up and how
much carbon they could capture
they’ve been using ai to analyze
80 000 images from satellites
combined with viable such as topography
soil quality climate all over the globe
they came up with the three next results
one in a map a you can see the global
tree
cover which is basically the number of
tree the world can support
this equal to 4.4 billion hectares
on map b you can see the global tree
potential cover available for
reforestation which is 1 billion
hectares basically is the 4.4 billion
hectares
minus current forest minus agriculture
areas
the studies is telling us that this 1
billion
a hectare equal 1 trillion tree
and that this trillion tree is able to
capture
200 giga tons of carbon emissions
which equals to 100 years
of carbon emissions the third result is
also super interesting
half of these trillion trees are located
in the northern hemisphere
meaning that some of the richest
countries in the world usa
canada russia europe china
hold the power to reduce our carbon
emission by half
going further ai can help us with
restoration program
this is what the company pachama is
doing
they’re using deep learning combined
with images from satellite and drones
to assess from the shape of the forest
the level of carbon capture and the
level of biodiversity and wildlife
let’s take an example here in brazil and
on the left
you can see the monitoring work of
pachama and here they’re assessing
the level of carbon capture from the
project
through the years compared to the region
this kind of tool
are super important because today
only two percent of the fundings coming
from the credit card
markets are going to restoration program
so the more assessment tools we have
the more clarity and transparency we’ll
have and potentially
hopefully we’ll be able to drive more
funding to these restoration projects
now another area that is really dear to
my heart is how we can
fight against unfair treatment racism
homophobia
sexism how can we fight again these bias
racial buyers gender buyers that follow
us from
very old time and that are like kind of
hardwired in our brain
it’s interesting because we’ve never
been so talkative
and aware of this bias this day though
they’re supposed to be unconscious
because we’re so afraid to see them
perpetrated in our ai systems
let’s come back on the mechanism to do
machine learning you need data
and you need algorithm the algorithm is
the logic
applied to the data so basically we are
transferring our bias
consciously or unconsciously through
data sets and to algorithm so indeed
we are at risk that ai may reproduce
amplify and automate our bias that can
be frightening right
think about the former recruiting
algorithm of amazon
that was biased against women think
about this sentencing algorithm
in the u.s that were clearly biased
against black people
the good news is that we are aware of
this and that the ai community is
working to reduce it
i will even say that one of the most
interesting
and efficient solution to reduce this
bias
lies in ai itself because
with ai we can work at building
fair algorithm that will take over on
our unfair decision makings
and second we can use ai
to reveal and predict our level of
fairness
so we can act upon it that’s pretty
interesting because
if we rely on mentality to change it’s
gonna take
centuries the world economic forum is
telling us that it will take 130
years to close the global gender gap i
don’t know about you but i want to see
that in my lifetime
and i really think we can hack
mentalities for good with ai
so how do we do this how do we build
fair ai
three answers here one we need diversity
in engineering team that are building
algorithm today
it’s always the same kind of population
young
white male we need more ethnic
diversity we need more gender diversity
with engineers so they are able
to identify very early in the process
buyers that may be
deployed in their work second
we need to systematically use bias
auditing tools there are plenty of them
on the market
basically it’s like when you release a
car you check that it’s safe
let’s do the same with algorithm number
three
we need to make sure we have enough data
about underrepresented population in
data sets
in other words we need to be careful
when we use
historical data because this is where
the buyers are hidden
let’s take amazon again their algorithm
was basically
doing what it was supposed to do learn
from 10 years of recruitment at amazon
find a pattern
and make a recommendation on this of
course those 10 years of recruitment
were a majority of men
so obviously the algorithm will favor
men now if you train this algorithm
with performance data rather than
historical data
you may be able to reduce the risk
this is what the company pi metrics is
doing they’re developing a talent
matching platform
and they use ai to assess the soft skill
of candidates
they will assess the level of generosity
fairness
empathy and they will do the same job
with employees from the company that is
recruiting
and then they will compare the score the
candidates
whose skills are the closest to the most
performing employees
who will move forward in the process
another way to answer is how can we use
ai
to reveal and predict our level of
fairness so we can act upon it
i’m going to take here two examples in
the justice system
the stanford computational policy lab is
helping prosecutor
in san francisco to make race blind
charging decision on
incoming felony cases what does that
mean
basically they train an algorithm
to identify race related information
within policy policing electronic
records and to remove them
so basically in the example you see the
algorithm
has identified and removed the name of
the victim
in the indication related to the hair
color
skin color or neighborhood
now last example and this is one of my
favorite we can use
ai to assess the furnace of trials
trial watch is an ai powered platform
developed by microsoft and the clooney
foundation whose mission
is to build assessment tools for
fairness in trial
and protect vulnerable vulnerable groups
such as lgbt
women political opponents or journalists
whose right are at risk to be violated
during the justice process
so on this platform they’re gathering
data from trials all around the world
the ai will transform the records into
text
translate in different languages and
make it comparable so it can
identify pattern assess the trials and
grade the fairness
this tool is essential because basically
it enables us
to denounce publicly in the right time
any injustice
making the world kind of a witness into
the courtrooms
so you see there are a lot of different
ways to do ai for good
within our within our community we have
more than 200 projects
but every day we’re meeting new
entrepreneurs new projects
new brilliant ideas ai can be used to
speed up drug discovery
fight again coveted cancer balleria ai
can be used to
promote a more sustainable agriculture
ai can be used to fight against
misinformation
or terrorism of course
ai is no silver bullet we need
regulation
we need political incentive so it’s used
responsibly
and to encourage the use of it for good
but from what i’ve been witnessing in
the last year
we’re going in the right direction and
we are more and more informed
and empowered thanks to ai to make the
best decision and solve some of the
biggest challenges we face as a
civilization
and the planet and when my kid
will start to wonder what they want to
do when they’re older
i would have hundreds of heroes role
models
entrepreneurs researchers scientists
activists
and i will tell them their story and i
hope they will be
inspired and convinced they can change
the world for good thanks to artificial
intelligence
to all of you students people working in
ai
from near or far up to you to act
now i choose ai for good what will you
do
thank you
[Music]
you