[ Music ]
>> Sponsored by the
James Madison Council.
[ Music ]
>> Gal Beckerman: Well,
welcome to the 2021 Library
of Congress National
Book Festival.
I'm Gal Beckerman.
I'm an editor at the New
York Times Book Review.
And I'm thrilled to
be here today to talk
with Daniel Kahneman
and Cass Sunstein
about their new book Noise:
A Flaw in Human Judgment.
Hi, gentlemen.
>> Daniel Kahneman:
Hello, good to see you.
>> Gal Beckerman: So, so,
let's, let's get right into it,
because we don't
have a lot of time.
And I want to ask first probably
the most essential question
to understanding this
book and your work in it.
Maybe Cass, you can
take this first one.
And Daniel, you can jump in.
What is noise, and
how does it get
in the way of decision making?
And if you can give us some
of the wonderful examples
that you have in the book
to help us understand,
that would be great.
>> Cass R. Sunstein:
Thank you for that.
Noise is unwanted
variability in judgment.
So, if you go to a doctor who
says when she's, let's say,
energetic and ready to go in
the morning, let's do a ton
of tests, I think
you have a problem.
And when she's tired, let's
say at the end of the week,
at the end of the day,
she says, go home,
I don't think there's
anything to worry about.
That's not an egregious
case of noise,
but it is a case of noise.
An egregious case of noise is
when whether you get let's
say five years in jail
as a sentence, or a month
in jail as a sentence,
depends on a lottery who
is the judge who's assigned
to sentence you.
Or noise is a case where you get
hired with enthusiasm by a firm
to which you've applied,
and someone who is just
like you doesn't get hired just
because that's another person
who let's say has a
different noisy relationship
to you compared to the first.
So, unwanted variability
in judgment
within ourselves is
all over the place.
And in systems, it is rampant
and it is a pervasive source
of error and inequality.
>> Gal Beckerman: Now, is it,
and Dan, you can take this one,
is it always true that
there's going to be variability
and inconsistency
because we're human?
How do you distinguish between
the things that are systemic?
So, for example, you know,
one of your examples,
you just mentioned it, doctors,
you see doctors are more likely
to order cancer screenings
for patients they see early
in the morning than
late in the afternoon.
That seems to be
something once you figure
out that that's the
case, you can sort of,
you can balance that,
you can find a way
to solve that problem.
But then each doctor, you
know, comes to their work
with their bias and their
world view, so that seems not
so much a systemic thing, but
just a question of subjectivity.
How do you distinguish,
you know, what is the noise
that can be sort of dealt
with versus the noise
that is just human
beings being human?
Daniel, do you want
to answer that one?
>> Daniel Kahneman:
There is going
to be variability
whenever there is judgment.
And a statement on that is
that wherever there is
judgment, there is noise.
And what's most important, there
is more noise than you think,
because the reason we wrote
the book is that when you look
at the judgments
that people make,
when they're fairly
complex judgments,
not when they're elementary,
and by judgment I mean
they can be computed,
there is a reasonable answer,
but no expectation
of perfect agreement.
You find more disagreement
than expected.
And it's, you know,
whether it's human
or not, of course it's human.
But that's no excuse.
When you have a traditional
system,
you would want the judges
who are part of the system
to speak in one voice.
The insurance company
would like its underwriters
to speak in one voice.
And similarly, the hospital
wants the physician in the ER
to speak in one voice
so that the individual
who makes the judgment and the
particular statement of mind
in which that individual
is, if these are material,
if these have an effect on
the decision that is made,
that is undesirable, that's
unwanted, and this is the noise
that we're talking about.
>> Gal Beckerman: Right.
Maybe, you make a
distinction in the book
between occasion noise
and system noise.
Cass, can you explain what the
difference is between those?
>> Cass R. Sunstein: Yes.
So, occasion noise might be
that when you're very excited
and happy about something, you
will say I want to take the job,
I want to buy the house,
I want to go on the date.
But when you are not so excited,
because let's say your
favorite sports team lost
or because the weather
is terrible outside,
you'll say I shouldn't buy
that house or I shouldn't go
on that date or I
shouldn't take that job.
So, occasion noise means
that small features
in the background often, like
whether the Boston Red Sox won,
which I think everybody
should care greatly about,
can affect your judgment about
things that are unrelated
to baseball, like what you're
going to do with your life
in the next month, or
what your judgment is
about something that
really matters.
So, that's occasion noise.
System noise is the noise
that goes across systems
when you may have let's say
one doctor who's really great
and orders a ton of tests
and tends to overdiagnose,
and another doctor
who's also really great,
but tends to underdiagnose
and is really worried
about putting people
through the ringer.
So, then a hospital understood
as a system will have, be noisy.
If there's a lottery, who's
the doctor [inaudible]?
Level noise is meaning a
degree of severity on the part
of one set of judges, hanging
judges, and lenient judges
who were very generous
to criminal defendants,
that can lead to
great differences
within the system
of criminal justice.
And this is pervasive in
business with respect to hiring,
promoting, recruiting,
et cetera.
And the problem of system noise
isn't what principally concerns
us, though occasion noise
is really [inaudible].
>> Gal Beckerman: I just want
to remind viewers before I go
on that you can ask
questions in the Q&A function,
and we're going to get to
that in about 15 minutes.
We'll take your questions.
You're both sort of getting
at this, at this question
that I have, but
maybe just to kind
of pin it a little bit
more clearly on the wall,
why is consistency important?
Why is it and I guess
more specifically,
what fields of our life
are consistency important?
Because I can also imagine, we
can get to this too in extreme
of consistency and uniformity.
So, in what areas of our,
of our life, and you begun
to mention the criminal
justice system, maybe education,
where do we really need
to think about sussing
out this problem of noise?
Daniel, do you want
to take that one?
>> Daniel Kahneman: Yeah.
Well, you know, the answer is
in professional judgments, most,
when it's an organization
that is making judgments,
noise is truly undesirable.
And this is true not
only traditional system
and the medical system, it's
true in the pattern system,
it's true in hiring, it's
true in personal evaluation.
It is true actually
in fingerprint reading
because there is
noise even there.
And so the answer is
wherever there is judgment,
there is noise.
Now, we do not mind diversity,
and we welcome diversity
in many situations we don't care
we want variability we don't
want all our book
critics to agree,
we don't want all the
film critics to agree,
we certainly want variability
where creativity is required.
And we want variability in the
process of reaching a decision.
So, if you have a group reaching
a judgment, a common judgment,
we would, we welcome
diversity in that discussion.
But we would want
the conclusions
to be in general not noisy.
And that is true
really when you speak
about professional judgments.
Most professional judgments,
variability system noise
is going to be [inaudible].
>> Gal Beckerman: Right, right.
So, let's get to the kind
of solution part of this.
Cass, I wonder if you
can talk us through,
what is the notion
of decision hygiene?
And give us some examples
of sort of how do we,
how do we clean up the noise
in these various fields?
>> Cass R. Sunstein: Let's
back into that, if we might.
So, the character in this
particular play we haven't
mentioned yet is biased.
And bias is a little bit of
the charismatic performer,
the Taylor Swift
of decision theory,
whereas noise is [inaudible]
very uncharismatic rock star
that you've never heard
of because they don't
get attention.
Bias is systematic departure
from what's right, as in a scale
that shows you as
consistently heavier
than you are, that's my scale.
A noisy scale is one that
is all over the place
that sometimes shows you
as lighter than you are
and sometimes as
heavier than you are.
Okay, with respect to decision
hygiene, which is the set
of remedies we have for
the problem of noise,
the good news is the decision
hygiene helps with bias too.
It tends to counteract bias
as well as [inaudible].
And let's give a
couple of examples.
Decision hygiene has a way
of kind of washing judgments
so that it combats enemies
that you may not be able
to identify in advance.
And one way to do that is to
have guidelines, guidelines.
They cut noise dramatically.
And if they're good guidelines,
they'll cut bias too.
So, when a little kid is
born in the United States,
there's something
called Apgar score.
It's pretty simple.
It's a guideline and
it cuts both noise
and bias simultaneously.
The second thing you
can do is really simple,
and it's intuitive, and in
medicine we see it, which is ask
for a second opinion
and aggregate the two.
Now, if you've got more than
two to aggregate, that's better.
So, if you've got a bunch
of independent judgments
and add them up, the majority,
the average will
be much less noise
than if you take each
one individually.
And under reasonably pervasive
circumstances, if you do that,
you're going to cut
bias as well.
>> Gal Beckerman: I was
interested about in one
of the sort of tools that you
described, Daniel, can you talk
about decomposing that decision?
What does it mean to
decompose a decision?
>> Daniel Kahneman: Well,
an example would be,
a very elementary example
would be when you're hiring
for a particular position, then
decomposing, if you are going
to interview someone,
decomposing would be
first constructing a list
of the attributes that
you require for the job.
And then in the interview,
you would focus on each one
of these attributes
one at a time.
And independently of the other.
And you would attempt, you
would delay your intuition,
you would delay the global
judgment about the person
until you have covered
all the attributes.
Now, more generally, when
you're facing a decision
with multiple options, you can
see any option as a candidate.
And options have attributes
that make them more
or less desirable.
And evaluating attributes
one at a time
and delaying global
evaluation are recommended steps
for options more generally as
well as for hiring candidates.
And certainly, similarly
for evaluating candidates,
or evaluating people, the
performance of people,
you would want to, you would
want the person doing this
to be thinking of separate
events, of separate attributes,
performance, and not to
jump to a global judgment.
>> Gal Beckerman: Now, I'm
wondering sort of the role,
what you see as the role of
technology in solving some
of this problem, some of
these problems with noise,
specifically when we think about
algorithmic judgment and sort
of removing, removing
the human variability.
First, let's talk about
sort of the positives.
You know, do you
actually see some hope
in technology's ability
to reduce the noise, Cass?
>> Cass R. Sunstein: I
noticed in the last period
that when one says something
positive about algorithms,
people tend not to
smile or jump for joy.
And when one says something
negative about algorithms,
people do tend to nod,
not nearly in a sense,
but in please sense.
I'm going to take a risk here
and say something
positive about algorithms.
The first thing that
algorithms do is they eliminate
the [inaudible].
They don't cut the [inaudible].
They eliminate it.
It's like a scale
that always spits out,
if not the right
answer, the same answer.
So, if you have an algorithm
to predict let's say
whether people are going
to flee the jurisdiction if
they get bail, or an algorithm
to predict whether someone
is going to stay in a job
in three years, it's going
to give the same answer
whether it's in a good mood
or in a bad mood, whether
it's Tuesday or Thursday,
it won't be in a mood.
And there won't be
any system noise.
Now, I think the human mind does
not get thrilled to hear that.
But it should be a
bit happier than it is
because cutting noise is other
things being equal are really
good thing.
It tends to cut error a lot.
Think of a scale that's noisy,
it's all over the place,
the errors add up and
cause total error in terms
of joyfully misperceiving
themselves as thin people,
and depressively misperceiving
themselves as overweight people,
those are both errors,
and they add up.
So, algorithms do
eliminate noise.
Whether and that's very good.
They can also compare to human
beings by using the word can,
not well, be less bias,
along any dimension
that you care about,
they might be less judge
so cognitive biases, let's
say like unrealistic optimism,
they might also, and this is
very much a work in progress,
be less subject to
other forms of bias,
which are very familiar
these days, let's say.
>> Gal Beckerman: Okay.
And if we could talk for a
second about the flip side,
I mean, there is, I imagine, you
can also imagine the extreme too
of the sort of dehumanizing
process that can happen
if people are just
turned into numbers.
Daniel?
>> Daniel Kahneman: Well, in
general, people are not in favor
of algorithms when they
compete with humans.
So, there was a film that
I recommend to everyone,
by the way, it's called
AlphaGo, and it's on Netflix,
and it describes a
competition between a program
and the world champion of one
of the best world players in Go.
And all of Korea was transfixed.
And there was, on game four,
the human beat the machine.
It was one of five games
where the human prevailed.
And the interesting thing is
everybody was joyous, including,
by the way, the team that
constructed the program.
They were clearly thrilled
to see the human
defeat the machine.
So, we have a very strong
prejudice against algorithms.
They are going to be resistant.
There's going to be
a lot of resistance.
And there's going to be very
tricky rearrangements to be made
where very good algorithms,
there is going
to be a temptation to replace
judgment by algorithms.
And sometimes it's going
to be a good thing to do,
and other times it's going
to be a bad thing to do.
And one thing that you know,
it's going to be
very complicated.
And this is coming, those kinds
of decisions and those kinds
of difficulties are coming.
And clearly having algorithm
replace humans is not,
is not problem free.
It's going to cause massive
disruption as it occurs.
>> Gal Beckerman: I have a
sort of a metaprocess question,
which, you know, with my book
review, my book review had
on somebody who looks
at books every day,
lots and lots of them.
What was it like to
work collaboratively?
How did that come about?
I was sort of surprised and
thrilled when I saw three
of your names, right, you're two
of three authors of this book.
Can you explain to me I guess
maybe a little bit how it came
about, but also what the process
was of working together on this?
Cass, you can go ahead.
>> Cass R. Sunstein: I think, I
first became interested in noise
about seven years ago I think.
And then about five
years ago, [inaudible]
and I began collaborating
on something
that could become a book,
but we didn't really believe
that there would be a book.
And then, and then, let's see,
three or four years ago now,
Cass became clearly
interested and seemed
to be interested in joining us.
And we were thrilled
because as soon as we knew
that Cass was onboard, we
knew that there was going
to be a book, because
he is a force of nature,
and he was going
to make it happen.
And the collaboration was, you
know, as collaborations are,
we sometimes disagreed, we
mostly agreed, and we found,
and we found a good way
of working together,
in part by separating topics,
and in part by joining forces
and really working
cooperatively on each chapter.
And we had both forms
of cooperation.
>> Gal Beckerman: I'm curious
where the disagreements were.
I mean, I imagine the general
concepts of noise was one
that you all were onboard with.
But what's fascinating
is to imagine sort
of where you actually
found misalignment.
>> Daniel Kahneman: Well, the
disagreements were not so much
on content as they were
on the level exposition
and how many readers are
we going to lose when we go
into technicalities
and how essential it is
to have technicalities and to
speak to an academic audience,
as well as the public audience.
So, that was a balance that we
were constantly struggling with,
each of us within himself, and
frequently we disagreed on that,
and eventually we
reached consensus.
>> Gal Beckerman: Great.
Sorry, Cass, did you
want to add something?
>> Cass R. Sunstein: Yeah.
We really had fun with
the book, I should say.
And overwhelming experience
of the collaboration for me
at least was fun and
joy and laughter.
And part of the joy was Danny
is the master of seeing problems
in his own pros and analysis.
And he's almost as
good at seeing problems
in his co authors'
pros and analysis.
So, the constant 5:00 p.m.
notes saying I don't know why I,
Danny, would write and seen
until now the fundamental flaw,
and then maybe at
5:28 a.m., he'd say,
I think I might have a solution.
And that didn't happen a little.
And I'm not nearly as good at
that as he is, but I'm pleased
to say that toward the last
stages, some parts of the book
that were pervaded by detailed
analysis of law and policy,
it occurred to me late
that no one's interested
in this except for me.
And so I [inaudible].
>> Gal Beckerman: Got it.
Well, let's get to,
with our remaining time,
let's get to some
questions from the audience.
One question that I'm
interested to hear,
if you guys did research
on this specific point,
but on the question
of law enforcement,
noise in the world
of law enforcement.
Is there any insights
that you were able to come
to on that, on those questions?
>> Cass R. Sunstein: Yes.
So, that, if it's not explicitly
called out in the book,
it's implicitly there.
So, the subject of
randomness with respect
to law enforcement is very
much part of the background.
And in some places, the
foreground of the book.
Where when you think about
noise, you might think
about racial discrimination,
which isn't quite that,
but it certainly, if
understood in a certain way
in the same family, and the
fact that, and this does come
up toward the end of the book,
whether under a vague law people
have an unfriendly encounter
with the police, it's
often like a lottery,
as well as it having a racial
and class dimension,
there's a lottery.
And that's a problem.
And a big question is
what do we do about it?
>> Gal Beckerman: Right, right.
I have a question here,
someone named Julie Staple,
we are in a moment when
a significant portion
of the population is proud
to make decisions based
on no information at all.
Can your analysis speak to that?
Daniel, do you want
to take that one?
>> Daniel Kahneman: Not really,
I don't think [inaudible]
to that issue.
We are really focused on
professional judgments,
and the judgments that
professionals who do want,
we assume that they do want
to reach a great decision.
This is the assumption
we are making.
And we assume they want
to use information.
And noise happens in spite
of their intentions
to do a good job.
When people prefer to be led by
their emotions and that I'm just
as accepting of misinformation
as of valid information,
we have very little
to say about that.
>> Gal Beckerman: To say
about that, right, right.
Another question here,
noise sounds a lot
like statistical
quality control.
How does your work
differ from or go beyond?
And then the question or
mentions Deming, Juran,
I think other social scientists.
Is that, is there something
to that that you're looking
at things that are adjacent to
statistical quality control?
Daniel, you can answer that.
>> Daniel Kahneman: Yes.
>> Gal Beckerman: Yes?
>> Daniel Kahneman: The
answer is absolutely.
We deal with is effectively
quality control.
That is, and the basic idea
of a certain [inaudible] is
to achieve discipline thinking
and to discipline judgment,
which is the essence of
that is quality control.
And it's the reduction of noise,
by the way, in quality control,
because when you are limited
noise, you can see bias.
And then you can correct bias.
But eliminating noise
by achieving some uniformity
is actually an early step
in quality control
and an essential one.
>> Gal Beckerman:
We have a question
about whether suggestions for
readings, for those interested
in reducing and identifying
and reducing noise,
do either of you have any good
reading suggestions for people
who want to dig deeper
into this topic?
>> Cass R. Sunstein: Well,
there's a book called Noise
that has [inaudible]
of producing noise.
And that book, I'm
pleased to be able to say,
has a lot of footnotes that
deal with particular areas.
So, if you're interested
in recruitment and hiring,
or if you're interested
in medicine,
something that we got
[inaudible] interested in,
or if you're interested in law
and policy, we have something
like a bibliography that can
be extracted from footnotes.
>> Gal Beckerman: Let
me ask a question of,
another question of my own here.
I'm curious about bias
training and sort of the whole,
this whole world of, which, and
a lot of institutions have sort
of taken on over the last year
with a lot more intensity.
How do you see that sort
of converge or diverge
from this world of
identifying noise?
I mean, that seems fairly
subjective as opposed to kind
of the systemic kind of
thing that you're looking at.
But is there, is it
a good development
that institutions are making
us more aware of things
like our own bias with
regards to this big question
that you're looking at?
>> Daniel Kahneman: Certainly.
But the bias that we talk
about in this context is
a particular kind of bias.
It's a bias against social
groups, enemy groups, or gender.
And those are not
the cognitive biases.
When we talk of controlling
noise, we're talking mostly
of the cognitive side, not
about the emotional side,
which is the one that is being
tackled in bias training.
>> Gal Beckerman: Right, right.
Another question here,
what happens when one
of the noisy decisions is
actually the correct judgment?
So, and they give us the
example, the 2008 stock crash,
where the majority expert
opinions were wrong.
Can you speak to that, Cass?
>> Cass R. Sunstein:
Well, if you have a scale
and you weigh yourself
a thousand times,
and one of the times it's
right, that's a positive thing.
But you'd be lucky to be
able to figure it out.
Maybe you have an accurate
scale that can tell you
about the noisy scale.
The problem with the noisy
system or a noisy individual,
like an astrologer,
who sometimes is right,
is that it's very hard
to identify in real time
which of the assortment of
judgments is the accurate one.
And that's why decision
hygiene is so crucial
as a noise dampener that
can typically get you closer
to what's accurate.
>> Gal Beckerman: Okay, I'll
just, I'll ask one more just
to kind of close us out here.
Which, which, you know, as you
look out on the various fields
that you are examining
where noise is a factor,
I guess in which, where is it
most consequential you think
in our world today
that we figure out how
to eliminate noise, or just
most a question of sort
of if not life or
death, but, you know,
really could make an impact
on the way, on the way people,
on the way people live?
And you can, Daniel,
you can start.
>> Daniel Kahneman:
Well, I mean,
clearly there is noise
in policy making.
So, for example, you can see
the difference jurisdictions,
different countries, different
regions have different rules
about COVID, and very
centrally the same situation.
And that's noise, and it's
not necessarily a good thing,
although in this case
perhaps there is something
to be learned from it.
I would say the life and death
is clearly medicine and law,
obviously urgent problems.
And in both of those, we would
want to make rapid progress.
>> Gal Beckerman: Right.
Cass, do you want to
add anything to that?
>> Cass R. Sunstein: I
say H S E. Health, safety,
and the environment,
taking them as a unit,
would be the area number
one for noise reduction
and decision [inaudible].
>> Gal Beckerman: Okay,
well, thank you both.
It's a fascinating book.
I enjoyed reading it.
And I think there are a
lot of important lessons
for how our society can
function better than it is.
I appreciate you joining us.
>> Daniel Kahneman: Thank you.
>> Gal Beckerman: Thank
you all for your questions.
>> Cass R. Sunstein: Thank you.
[ Music ]