Uncovering bias in daily decisions and how they impact our recruitment decisions
Bias in recruitment is understood and acknowledged. But how often is it proactively acted on?
The big 4 technology companies profess meritocracy but are they as meritocratic as they think they are?
This week, May Amr shares some of her studies on bias in every day life. As a senior member of our data science team, May spends a lot of her time identifying bias in machine learning models for recruitment so she has a lot of experience.
LinkedIn hello. Welcome back. Again.
Thank you very much for joining us a great start to this week.
Have a great weekend before.
Well, you know, one of the things that we got exposed to a lot is is by we we take on a some active steps to try and acknowledge it and confront it and deal with it as much as possible.
But we’re only human so we end up not recognizing all the devices that we have and sometimes making mistakes without even knowing we are making those mistakes.
So today I’m joined with me again who’s ahead updates and science here at such e we going to be talking about Rooting bias or at least as much as we can cover about bus, which is a huge huge huge conversation and topic.
They’re going to try and cover as much as we can and share as much knowledge as we can with you about the different types of biases how they’re relevant now that affect us in our day-to-day lives in the next hour.
I’m good and it’s too it’s so big.
So, I think we’re going to cover just few parts of it.
A lot of research has been made there and to your left or two from the beginning.
So I believe and Studies have shown that we were born by we are wired biased and this is not it shouldn’t be it’s not a very bad thing because so the human the human has acquired this mechanism of categorizing creature titillations human beings into favorable and unfavorable in the in dangerous decision in dangerous situations where the time is the essence of the visual cues that we developed or badass survival mechanism
We just run away from it.
So it is safer to make those postpositives avoiding that avoiding good things and then then then then then avoiding the bad stuff on the argument.
If I’m correct is that we are born biased that we are kind of genetically programmed in our brains to lean into bias.
And the that is something that we have evolved to do tens of thousands of years.
Fifty thousand years, whatever it is.
I don’t want to get into a fight with any anyone about how long human beings have existed.
However. So we’ve evolved over that time as a protection mechanism to be biased.
Okay, what it what evidence is that the lead you to them if we can get the video on the documentary that shows that we we infant cognition Babe Ruth known as the baby laughs parents bring their children were going to unlock the mystery of what they’re thinking the baby Whisperer is dr. Karen win who she know team of researchers believe knowing good from bad isn’t learned all actually born with those instincts.
We watch babies as they were presented with a puppet show hear a puppet struggles to open a box first.
We see a green bunny who comes along and helps to open the box good behavior.
Then we see an orange bunny slam the Box shut and run away for the bad behavior with Bad Bunny in the two more than a bunny.
And for those just three months old.
It goes to almost 90% of the difference between good and bad.
What else do they know?
He’s a very young babies very clearly like people who are similar to them.
You know, that’s so nice of you just like those who may be different.
So we’re not talking about physical differences here.
We’re talking about differences of opinion think of it like people in the same political party or rooting for the same sports team for Berkeley and Parker.
Do you like Cheerios or do you like graham crackers?
Then she watches as a grey cat and an orange cat are presented with the same Choice.
The orange cat chooses the same as Berkeley.
I don’t like graham crackers.
So that’s the idea that the Berkeley gravitates to is that something
that’s learned or is that just something we found this in baby seven months of age?
I think it’s just a natural aspect of human nature that were built to judge.
You know in what ways is this individual like me and in what ways are they different from me?
You yeah, I don’t like Grandpa rushes NOW Watch What Happens Newsies
the orange cat struggling to open a box blue puppy?
Blue puppies bad behavior Parker likes the mean one.
So it’s it’s it’s almost said he wants to punish those who don’t have
the same opinion as in individual who is actually mean to the one who
has a different opinion from themselves, which seems that we were at a
deep level built to to not like individuals who are different from
ourselves into prefer those who are similar to us.
I think it’s I think there’s a lot of reasons why we might be built to
orient towards that is who are similar to ask want to hang out with
them and conglomerate with them.
You could think of it as you know, how the how the first special
interest groups get formed Right Center babies under a year old who’s
the main guy in the babies around Parker’s age a little over
one-year-old 100% of them prefer.
The mean guy bias is inherent.
Biases were built into us.
I think it does me.
I mean, we’re biased to others who share our opinions or a similar to
us and keyways and I guess to try to spend the email the positive is
it weird to dislike differences were built to like similarities.
It’s it’s remembering that someone is like us in some way that gives
us the the connection to them liking good behavior and rewarding good
And we seen them Express not so positive Feelings by discriminating
So what are these studies mean in the long-term care Liberty material?
Hope that by studying is an eight feelings a baby’s we can all learn.
Where are moral beliefs come from Anderson Cooper CNN, New Haven,
Super powerful told me three were you understand from from this case
study this kind of prejudice or buy it from our early years.
So we are in favor with others who share the same opinion is 100% a
confirmation bias thing at this early stage, but there was so many all
the kinds of devices that aren’t necessarily that I’m leaning into
Being clearly defined who described as a mechanism for survival.
So we talked a lot about in in the recruitment by specifically there
were two major boss is a play you got the confirmation bias, which is
a new that person was amazing.
I said it right from the beginning looking to see if they’re amazing.
What you’re looking for is any kind of signal that you can use to
reinforce your decision say looking for the signals that confirm your
your initial kind of assessment and then the second is the halo
Sorry, this this halo effect is an interesting one in relation to what
we were describing just now in terms of kind of How It’s advice is
often can be created for as a survival mechanism what describes a
survival mechanism because the halo effect is absolutely.
I can’t see how it could be defined as a survival thing.
So for those that aren’t familiar with the halo effect is basically a
Effectively says the way it’s described is that people who look better
to sound smarter who sound better will be perceived as being better at
So are some examples of this.
Does anyone out that is what team does seen or read money bull?
It was a play in a very big way in in in in the story.
So the idea pre-money boom was the you have to look at something white
The Facebook might have to be to have to hit a certain way.
And if you did order those things will you look like you could do all
those things been the recruiters for these baseball teams were more
likely to suggest that you were really good at something really good
at baseball, but if you didn’t look,
The same way if you didn’t know the right way, they were more likely
to grade you down and this happens all the time.
So we we tend to elect leaders who we think looks likely does that
attractive to Le look small so dress better?
And I can’t see so much hell that is directly related to survival.
Yeah, there are a lot of a lot of those different price is right.
So yeah, so they’re there are different devices.
But those those factors that you just mentioned or just the evolved on
how the community perceived.
So what makes something better than another is a complete perception
people want to fit and they want to see people fitting into that they
are perceived as being better.
So this is not just your bone with a thing that you learn right?
So then how do we how do we get around these problems?
How do we find routes to be less affected by bias?
Let’s talk about was a prison about how this how would these factors
can affect affect our decisions how it can affect the the workspace
two divisions that we already have the result of the obvious ones that
did the gender by the ethnic group buys a friend if we can put put put
on screen to the eBay auction.
The other was this Thursday, we’re out everywhere to auctions on on
eBay or both of them were showing the same exact product baseball
I want to save a Facebook.
That one was that the cards were helped by African Americans or black
The other was that help with the with white hands and the ones with
the black hands got let go to 20% less than the White House.
your decision visiting and then and then sings What in workspace if we
can show a friend if we can show the if I go right sample, how did the
wick space is divided tendon out and now we are facing problems with
the gender distribution and that companies and how the house jobs are
are being assigned to two two genders Be There lyrics
Some jobs Austin lamb considered based on gender.
In the main area.
I think this is so this is stuff or anyone that’s watching the things
of this is under control.
This is still a conversation which is happening at the moment.
So cycle one of our founding team members head is currently in Silicon
Valley and she’s been with some very large companies in this list and
Been explaining the challenges of a front-facing with diversity in
gender and diversity across the board takes you from the hiring
So he knows that cold medicine over there still battling with
identifying a solution to that particular problem, and I don’t think
it’s going away anytime soon.
Is bias and now that we are training going machines to take decisions
The compass is an algorithmic.
Are you learning?
I can decide on whether a prisoner or how long a prisoner will stay in
jail is it wasn’t originally designed as a sentencing to believe it
ended up being used as one in the states and overseas.
The thing with machine learning is it lunch with inherently biased in
the base of the wolf eating it then or in-app purchases or things that
we do is a size 2 already then the machines are going to lose in this
kind of echo chamber revolver lean in the direction.
Starting a betta is biased and so our machine learning models for
getting boys, but there’s some good news.
Yeah, I think we
and it’s about the degenerate.
Denver CO so
So what where we go to from?
So we’ve got the there are certain jobs which historically have been.
Kind of all foods older than more opportunities for for women than
So you mentioned the software engineering Paul for men and that’s a
really clear example, then we’ve got other examples on the council
side to that.
So you felt like here we can see this.
I think it was a student of 2014.
Does pretty crazy right to see there was something distributions of
Joe Sutton Jose Fortune will have any distributed Two Worlds by the
And I thought I see what he missed your see what you were saying
either that buddy about the drug distribution being valid because both
taking the ethnic groups are not the same opportunities as acacian.
I agree with that.
I agree with that 50 years ago, and I don’t necessarily agree.
So I’m not sure anymore.
I filled up their educational opportunities which exists online which
are accessible to people with an internet connection mobile devices
with internet connectivity or internet connection.
No more prevalent than drinking water across the wild at the moment.
So I’m not entirely convinced the argument So Much Anymore think that
we’re getting to his stage for a gender racist set truck that they
Of a Berryessa education, which should make it less of a barrier to
I think more and more frequently at the moment though the issues we
From an AI or machine learning perspective as well as from in society.
They boil down for me.
At least they boil down to these inherent bosses that we have.
And I don’t know if we need to.
Be perfect, but do feel the buzz rude for us to improve.
And I think with machine learning are kind of like your take on.
This is with machine learning.
We have an opportunity to exponentially improve by identifying some of
these boxes and then not teaching those boxes from machine.
Yeah, that’s yeah, I agree with that to imagine a recruiter out of a
can lie about the reasons for not accepting a candidate while I’m all
done a machine learning model.
We would know if it if it’s Buy store not that once we put our hands
when we pointed fingers to the biases Factor.
We would know why this person was not hired sewing machine with
I wouldn’t fly.
So there are opportunities for changing the fact that we have right
now and for making the world a better place with the most rebuilding
if we point out those voice actors, and we we we take it into
consideration all the way while rebuilding the does the Maldives.
the point where we start collecting data and make it as inclusive as
possible and the way a refracted in the the attributes what we want to
take from this date and the features that we’re going to include in
our models and then who we include in a team who is going to be
did we have a diverse team 10 members with stores checking on the
blank photo of each other equipment out there with the mother is not
his what’s your building has four matters?
And obviously the date of matter is though I think that’s so do you
think that we got into a stage where the philosophical questions
behind the use of a lie?
I becoming more prevalent or is becoming more important to answer
those questions than we’ve ever faced before.
So far, he’s on prevention like considering who’s built.
The model is important.
No one ever said it’s important to know who belt.
I don’t know Twitter when Twitter was first created.
No one needed to know the background of the software Engineers for the
Twitter rant and I wasn’t a consideration that was just buy that
building this software that does stop.
those biases in the Maldives that we using.
Big decisions are being taken from those from those predictive model.
Like they’re like the the sentencing the compass the process model or
the recruitment process that you’re trying to introduce.
These are big decisions that we need to to flag those factors and the
north was that are already there that that can like identifying bias
in statistics Justice League vs.
You can submit your mother to be tested by Community text.
And it’s that there’s no ethnic bias in their there’s no gender bias
the that sits at waist.
Well on all the all I think that those measures are really really
valuable as there are things that I’ve always make Justin’s late.
I think it’s cool though.
They are Justice League.
I think having those things available to us a very important because
they allow us there’s so many things that we don’t recognize that
where it where bias to like, for example, you’re in my team and
therefore if anyone comes and starts to try other night creating
problem or if there’s for whatever reason you poach you away.
I’m going to I will because you were close to me because we are in
proximity with each other and because I value you and I’ve got to know
you over the lost.
However long it
I will intrinsic I will overvalue you.
Because we’re on the same thing might just like I will open by my
Will my car compared to what the market will find you as this is
sometimes we just can’t see the bosses.
So having those
exactly, right so having those opportunities to go to a third party
who can then come back to you in so I can pay you got an issue here
fixing flag out for you having that kind of oversight on my legal
ramifications to having that kind of option of having that kind of
oversight I think is an incredibly valuable in ensuring that we take
the right kind of Direction across the Escalade on a conversation,
even if it’s just satisfies you that so you can report you can report
On something that’s built and it’s not inclusive enough that the
advertisement of Justice League do that.
Alright, so we’ve talked about the technology companies.
We talked about the eBay example to little bit about the device in
machine learning and some of the consequences of that especially with
the The Compass exam fluffiness a really strong example to avoid that
he must thank you for being so I took brother
So one of your points is if these biases have helped to survive as
human beings are as a business.
Why is it considered bad now that we have a i
do you want to stop?
I thought you can you repeat the question.
So the question is if these biases have helped to survive as human
beings or as businesses, why is it now considered as it’s affecting
it’s affecting how the the word space is now is divided and it’s not
it’s not necessarily that it said it supported us store.
It secure does it is the right thing right now is because there are
different type of why is there different type of cognitive eyes?
So it’s so we were born by yes, but that’s a some extent we need to to
What’s the what’s what we were born visit with two just to to have a
fair FFA decisions for a society.
I don’t know what the answer is.
And I think again we were just talking about the kind of philosophical
side of a i and how there is so many questions.
Now that kind of we’ve we’ve had some of the great thinkers of
History, Ponder on some of these questions without necessarily in
coming up with locking arms Subs, leaving it as a while this one of
those philosophical there’s no right answer questions and we will now
moving to a stage wet.
It’s not acceptable to not have a position weather is a right one or
wrong one remains to be seen But the self-driving call question is
suddenly from a philosophical perspective on it.
So if there’s an accident where a self car self-driving or if there’s
a situation where a self-driving cars face going straight the driver
or the owner of the vehicle of the passengers of the vehicle survive
weather for that but in front of them of pedestrians and they hit one
pedestrian killed on pedestrian should the car choose to do that or
should they like me put the passengers at Great risk and peril
Or should I do something else which is also going to bystanders and
historically those kinds of questions.
We didn’t have to have an opposite City but now because we’re kind of
pre-programming these machines to decide on an action to take now it
would kind of being presented with these very challenging for the sofa
There are some important lessons that we can learn from our biases.
We do have kind of an obligation to try and remove especially things
like a Creed Color Race language things like that.
Which room sorry, I knew that was going to come out eventually.
Sorry for so long then maybe we don’t need to don’t need to deal with
Repeat bosses actions were taken based on some bias or Prejudice,
which they sub is very obvious example something that is not rice
clear that most we can address these things this thing, which would
definitely we have an obligation as human beings.
We have an obligation as human beings who all living in the.
Of time that is great change upon us feel like we have an opportunity
on an obligation to consider how we might improve things instead of
allowing the status quo to continue the algorithms a tool that should
allow us to make change for a quickie, right?
what it’s the dumbest idea tune because of devices that can creep into
the models wild bring it while building and they are used in some
important decisions like hiring a prison sentencing or even if you’re
going to take this loan or not if you’re going to get this insurance
or not then
Yeah, I mean if we can completely without human involvement, I don’t
think anyone would suggest wearing a costume that what we’re doing is
trying to replace the role of a recruiter and it’s absolutely not
designed to be that is supposed to be like a crotch or an aide to help
recruiters Screen through more candidates milk weight fee with more
insight into those individuals Beyond just, two pages of paper
substitute to avoid the unconscious voice data could recruiter can
have white shorts his thing.
But we face this question overtime about is a I had it last night and
I had it today as well as they are going to replace humans or human
involvement or whatever.
It might be in that condo in that kind of vein.
I find it ridiculous.
I can’t imagine anyone turning around saying, you know, also you have
So now you should just completely rely on your team to do everything
and you don’t have to do anything.
That’s not no one ever says that no one ever said a team and I can
rely on them for everything.
I don’t have to do anything.
The only people that do that kids R kids are the only ones that lie on
me without any involvement of the right in the same as they were
recognized as an adult.
You don’t just rely on one specific thing or one specific person you
take matters into our own hands and you control the things that you
were able to control encourage everyone to consider that I would
encourage everyone to consider that that the role of a are especially
in the con of stage.
That’s not at the moment isn’t to replace anything to make us better
identify things make us better is posting things which would not
necessarily good at recognizing because we haven’t evolved to be
So what do you want to talk about the approaches that we take the we
have a pretty diversity of how many?
We have a lot in Rio 16.
Mike 66 lots and lots of different nationalities and Jen display leans
more towards females in those tomatoes so that we all think it’s
pretty close to 5050 is on 5545 or something.
But still this diversity thing continues regardless of the fact that
on paper we have great gender split great diversity on age as well.
We’ve got 20 year olds the 20-something year olds through to 40
something year olds and people from all over the walls.
All these different countries with different upbringings or blah blah
blah the mindset differences, but you’ll start playing to how we’re
trying to tackle.
This is kind of approach to how we’re trying to position our approach
to to buy some machine learning to Stephanie on the on the people side
with definitely tackling it head-on from that perspective by with the
with people love.
Different cultures in the collecting this data from people all around
So our data is pretty diverse as well.
I think we have a nicely diverse Stacey sent people from different
parts of the world that are labeling people from different parts of
the world as well.
So this is a personal multiple people labeling information.
And then we’re trying to average out.
So is the Ri Ra juice any issues and tons of pain outliers.
I also think that’s quite an effect of steps.
I’ve taken 2
What else do we do?
We got people who don’t know anything to talk to us.
Are we have a lot of people who are involved in some of the testing
process that no, nothing.
They know nothing about any of the complexity that I can bury
I’m gathering feedback from those people that helps is to ensure that
we know what kind of overthinking or shooting in a very weird
I think getting that feedback is this is very important very important
to identify if we’ve missed something.
Personality traits which has a worm in Walton.
We are enabling our data for gender at 4 for gender ethnic group
Not sure not to not for any reason just to stand correctly sample or
data while we’re building a wood mall is just to make sure that our
data is inclusive of all those type of devices that can that can girls
live in the features that we’re extracting from the videos or tell me
We are not we are not taking into consideration the words that are
being used just the tone of voice.
The frequency of the voice as is what is taken into consideration of
the energy in the voice how a person is enthusiastic while while while
talking and while performing the interview,
Oh and then going into the vision was tied extracting the action units
and these are the main features that are taking into consideration
while building the personality traits along with the with the with the
So excluding the living the word 6qt or we have this Dolan you’re not
convinced with kind of talking about it.
And we do have this goal with the van is described as looking pasta
Something is still early days and times of getting us to that point,
but I think it’s it.
It’s a very Noble goal to have so they all I did the concept of
looking pasta pixel is it it doesn’t really matter took the color of
the density or whatever right now.
I think it’s very Noble concept.
But we need to we need to continue exploring and we need to allow that
I think we need to do a lot of that as we continue going for Words
About how are the main points that are in top of my head right now?
How do I extract a lot of feedback from from clients which house has
helped you end up with the situation where I have a client kind of
performs some analysis on their own team and they have an opinion of
that team which is based off of very little interaction with them.
They don’t really know them and then you could have an opinion which
isn’t based in reality.
It wasn’t grounded in reality or they could have an opinion which is
led by the person behaving or something when way when they were around
that team leader.
And then sometimes we have issues with our analysis conflict completes
quite aggressively with with what the perception is and is a challenge
to be able to to Conner.
people to look at it slightly differently and to consider why they
might be wrong.
Our goal is to be significantly better than people.
All right now being able to do one specific to approaches identified
from from short periods of a video in an information
What is a challenge on the list to be able to communicate to somebody
that they might not necessarily be grasping all of the relevant
information from the person the best speaking with although they think
they know I’ve got some some cool stuff happening.
Next week is going to be back.
She’s in the US of the moment in Silicon Valley which she seems to be
enjoying a little bit too much.
She’s she’s going to meet
Snoop Doggy Dogg in the next couple of days thinks might be today.
I’m not sure that’s cool.
Right Snoop Dogg, and then she will be back next week and on Tuesday.
She’s going to be I believe with a very cool lady who saw ramadani,
and they’re going to take you through the stuff that we’re doing.
How are video interview questions asked what’s and some of the insights that we can
provide to you?
Which might help you either with the recruitment process or
understanding a little bit more about yourself as an individual as
So shooting next week.
If you’re interested in that session.
I think that it will be very very valuable for people to understand
kind of how this stuff works and what inside she can get from it.
So thank you very much and we will see you next time.