馃攳
How do smart people make smart decisions? | Gerd Gigerenzer | TEDxNorrk枚ping - YouTube
Channel: TEDx Talks
[0]
Translator: Ivana Krivoku膰a
Reviewer: Carlos Arturo Morales
[12]
How to make good decisions?
[15]
If you open a book on rational choice,
[18]
you will likely read
the following message:
[24]
look before you leap,
analyze before you act.
[28]
List all alternatives,
all the consequences,
[31]
and estimate the utilities
and do the calculation.
[35]
This is a beautiful mathematical scheme,
[37]
but it doesn't describe
how most people actually make decisions.
[42]
And not even how those who write
these books make decisions,
[46]
as the following story illustrates.
[48]
A professor from Columbia University
had an offer from a rival university,
[55]
and he could not make up his mind
whether to accept, reject, go or stay.
[62]
A colleague took him aside and said,
"What's the problem?
[66]
Just maximize your expected utility!
You always write about doing this!"
[71]
Exasperated, the professor responded,
"Come on, this is serious."
[78]
(Laughter)
[82]
The method of listing pros and cons
and doing the calculation is an old one.
[89]
Benjamin Franklin once in a letter
to his nephew recommended exactly that.
[97]
Listing all pros and cons,
and then weighting and adding.
[101]
And at the end of his letter,
he wrote the following:
[105]
"If you do not learn it, I apprehend
you will never get married."
[110]
(Laughter)
[112]
Did you choose your partner
by a calculation?
[119]
I asked my friends who teach this method
as the only method of rational choice
[126]
how they chose their partner
if they had any choice at all.
[132]
(Laughter)
[135]
All of them said, "No, no, no."
[138]
There was one exception.
[139]
And he told me that he applied
his own theory.
[143]
He explained to me
that he listed all the alternatives,
[146]
all the consequences.
[148]
For instance, will she still talk to me
after being married?
[154]
Will she take care of the kids
and let me work in peace?
[158]
And then he estimated
for each of these women
[160]
the probability
that it will actually occur.
[165]
And multiplied it by the utilities
and made the calculation.
[168]
Then he proposed to the woman
with the highest expected utility.
[172]
She accepted.
[174]
He never told her how he had chosen her.
[177]
(Laughter)
[178]
I've met him recently.
Now they are divorced.
[184]
I will talk today about two ways
of making decisions.
[190]
One is the one that is taught
at the academia;
[194]
it has many names:
[197]
Benjamin Franklin's bookkeeping method
or expected utility theory.
[202]
And this is a method
that works in a world of risk,
[207]
that is, when the probabilities,
[209]
the consequences
and alternatives are known.
[214]
Here, statistical thinking is enough.
[216]
A key example is lotteries
or if you play the casino.
[221]
Then you can calculate
how much you will lose.
[226]
In a different world,
the world of uncertainty,
[231]
calculation is not enough.
[234]
You need smart rules of thumb,
that are technically called heuristics,
[240]
and good intuitions, which are also based
on smart rules of thumb.
[246]
I will talk today
[248]
about decision making under uncertainty
[252]
and also about the dangers
[254]
of using systems that work for known risk
[260]
and applying them
to the world of uncertainty.
[264]
Choices between two jobs
or between partners
[269]
are all in the world of uncertainty.
[272]
You can't calculate everything,
you don't know the consequences,
[275]
and there will be surprises.
[280]
I will make four points
[282]
and then illustrate them
with two examples.
[286]
First, the best decision under risk
[289]
is not the best decision
under uncertainty.
[292]
Second, heuristics that you need
[296]
in order to make good decisions
in the uncertainty
[300]
are indispensable
for good decision making.
[303]
They are not, as it's often claimed,
[306]
a sign of a kind of mental retardation
[311]
or of just mental laziness.
[314]
Third, complex problems
[317]
do not always require complex solutions,
[322]
and that's again
in the world of uncertainty.
[325]
And finally, more information,
[327]
more calculation, more time
is not always better.
[330]
Less can be more.
[333]
Let's go to the first example.
[336]
This is sports.
[339]
How does an outfielder
catch a flying ball?
[343]
In baseball, in cricket,
[346]
or maybe in soccer,
where the goalie has to get it.
[350]
How does he or she know where to run?
[354]
There are two theories about that.
[356]
One is it's a complex problem,
you need complex mental processes,
[362]
and the other one is
it's a complex problem under uncertainty,
[366]
and you need to find
a simple method for that.
[368]
Let's look for the first one.
[373]
Richard Dawkins, in his famous book
"The Selfish Genes,"
[377]
proposed the complex method.
[379]
So what does the outfielder do?
He or she calculates the trajectory.
[386]
Have you ever calculated a trajectory?
[389]
(Laughter)
[390]
Okay, that's what you do?
[391]
(Laughter)
[393]
And this formula doesn't even have
wind in it or spin, so it's not enough.
[399]
But what else could it be?
[403]
What you see here is the idea to apply
[407]
a theory that works
if you know everything,
[411]
like under known risk,
to the world of uncertainty.
[415]
The key idea is that you say,
"Oh, he behaves 'as if' -"
[420]
and Dawkins puts in the "as if" -
[426]
the player would calculate that.
[429]
What's the alternative?
How do real players catch a ball?
[433]
That's my question.
[435]
And a number of experiments show
[437]
that real players use
a number of simple heuristics.
[441]
I'll show you one.
[443]
This one works when the ball
is already high up in the air.
[446]
It's called the gaze heuristic.
[449]
It has three steps.
[451]
First, fixate your eye
on the ball, start running,
[455]
and finally, adjust the running speed
[458]
so that the angle of gaze
remains constant.
[462]
This player here does exactly that.
[466]
He runs so that the angle of gaze
remains constant,
[471]
and that brings him there
where the ball will land.
[474]
Do you want to see it again? Here it is.
[477]
Importantly: the player
can ignore to estimate or calculate
[483]
every variable that's necessary
to estimate the trajectory.
[487]
Every one.
[489]
It's a heuristic that belongs
to a family of heuristics
[493]
that just looks at one good reason.
[497]
And then you get there.
[500]
You'll find the same heuristics
in evolutionary history,
[504]
so birds and fish,
[507]
when they hunt a prey or a mate -
[513]
which is sometimes not so different -
[516]
they just keep the optical angle constant
in three-dimensional space,
[521]
and that's enough.
[523]
No trajectory correlations.
[527]
This heuristic is used
by players unconsciously.
[531]
If you have ever interviewed a player
[535]
and asked him how is he doing this
so well, then you get "intuition."
[542]
And it's intuition,
[543]
meaning the person knows what to do,
but doesn't know why.
[548]
But how does this intuition function?
[550]
As you see here,
it functions by simple rules.
[555]
The same rule can be used deliberately.
[558]
Every rule that we studied
can be used deliberately,
[563]
and that's very different
from what you might hear
[567]
in some claims about decision making
[570]
that think that heuristics
are unconscious,
[573]
and statistical thinking is conscious.
[575]
Don't believe that.
[577]
Here's an example.
[579]
Remember the miracle of the Hudson River?
[583]
What had happened?
[584]
A plane hit shortly after takeoff
[588]
a formation of Canadian geese.
[592]
They flew in both engines
and silenced both engines.
[597]
The pilots turned around
[599]
to see whether they could get back
to La Guardia Airport,
[603]
or they would have to do something
more risky, like the Hudson River.
[610]
How did they make this decision?
[613]
Did they do calculations?
They didn't have much time.
[616]
They used the same heuristic,
now deliberately, the gaze heuristic.
[621]
How does it work in this case?
[624]
You fixate the tower
through your windshield -
[627]
that's what the pilots did -
[629]
and if the tower is slowly
moving upwards, you won't make it.
[635]
And that's exactly what Jeffrey Skiles,
the co-pilot, is saying, in other words.
[642]
Here is another instance where heuristics
can help us to make a safer world,
[648]
and the decisions are done very fast.
[652]
My second illustration
is the world of finance.
[658]
We know now that the theory of finance
[663]
is part of the problem,
[666]
or was part of the problem
of the financial crisis, not its solution.
[670]
Why? Because it's a theory
about known risk,
[674]
and it's applied
to the world of uncertainty,
[678]
and suggests certainties
that are illusiory.
[683]
Calculations of value, of risk,
and that sort of things.
[686]
I'll give you one example.
[690]
Assume you want to invest money,
[693]
and you don't want
to put everything in one basket,
[696]
but you want to diversify.
[697]
But how?
[699]
Harry Markowitz,
from the University of Chicago,
[702]
got his Nobel Prize
for finding the solution.
[707]
When Harry Markowitz
made his own investments
[710]
for the time after his retirement,
[714]
he used his Nobel Prize-winning
optimization method.
[718]
So we might think. No, he did not.
[722]
He relied on a simple heuristic
that we call 1/N.
[727]
You divide your assets equally.
[731]
For instance, if you have
just two alternatives,
[734]
you divide your money 50-50, and so on.
[738]
The interesting question
is how good is this simple heuristic
[742]
that doesn't need much calculation
in the real world of investment,
[747]
as opposed to the theory of investment?
[751]
A study by DeMiguel et al. looked at that
[756]
and gave the complex
Nobel Prize-winning method
[761]
10 years of data
to estimate its parameters,
[765]
and then to estimate what's happening
the following months.
[768]
The window wasn't shifted
until there was no data left.
[772]
What was the result?
[774]
According to common measures,
1/N made more money
[779]
than the Nobel Prize-winning
mean-variance model.
[784]
The interesting question is now
[787]
not just to show that something simpler
does something better,
[791]
but the real question is,
can we identify the world
[794]
where simplicity pays
or where the complex calculation pays?
[801]
I'll show you here
three features of this world
[804]
where 1/N outperforms mean-variance,
[808]
or at least very likely
outperforms mean-variance.
[813]
One feature is predictive uncertainty
is large - that's the case with stocks.
[818]
The second one is the number
of alternatives is large,
[823]
and you can see this
because the complex methods
[827]
need to estimate more parameters,
[829]
1/N not,
[831]
and then generate more errors.
[833]
And finally, the learning sample
is small; it was 10 years.
[837]
Now one can ask the following question:
[839]
if I would have 50 alternatives,
how many years of data do I need
[843]
so that mean-variance
actually gets better than 1/N?
[848]
What do you think?
[849]
Ten years is too little. Eleven? Twelve?
[854]
The best estimate is 500 years.
[857]
So in the year 2500,
[860]
we can switch from our intuitions, 1/N,
[865]
to doing the calculations,
[867]
provided that the same stocks
are still around in the stock market
[870]
in the first place.
[872]
Do our banks understand that? No.
[876]
I recently got a letter
from my internet bank which said:
[879]
"With Nobel Prize-winning strategy
to success in investment."
[883]
And then I read,
[885]
"Do you know Harry Markowitz?
No? You should know him."
[887]
And then a story was told
[889]
that he won the Nobel Prize
for solving the problem
[892]
and the bank has now adopted his method,
[894]
and there was a warning
about people's intuitions.
[897]
What this bank has not understood is
[901]
that they sent the letter
500 years too early.
[904]
(Laughter)
[908]
This is my second illustration
[911]
about the power of simple rules
in an uncertain world,
[917]
and also about the damage that can happen
[920]
when you rely on methods
that work very well in risk,
[924]
but apply them blindly to real world.
[928]
Ask your own bank what they use.
[932]
That this is not just a one-shot
is shown by this slide.
[937]
There are 20 studies,
and what you see here -
[941]
we have an optimization model
that's widely used multiple regression,
[946]
and we have three heuristics.
[948]
The minimalist is too simple.
It just picks something randomly.
[954]
The other two heuristics
have a different philosophy.
[957]
You already know 1/N, you just throw
the weights away and do it equally.
[963]
One good reason
[964]
is a heuristic that goes with
the first good reason that it can find,
[969]
and then it's there.
[971]
I'm not going into details,
they are all mathematically studied,
[976]
and what you see here
is something important.
[979]
When you know already all the data,
[983]
that's called fitting,
[985]
then the complex model is the best one.
[989]
So you can, you're flexible enough,
and you explain the hindsight.
[993]
When you have to predict,
then something interesting happens;
[997]
it's a crossover.
[998]
Every one of the simplifications
is more accurate,
[1003]
not just more frugal and faster.
[1007]
This condition is like hindsight.
[1009]
For instance, I hear often on the radio
[1013]
a financial adviser being asked
[1017]
why did Microsoft go down yesterday,
[1021]
and he has always an answer.
[1024]
This is hindsight.
[1026]
If you would be asked if Microsoft
is going up or down tomorrow,
[1030]
that would be prediction.
[1032]
And prediction is hard.
[1035]
That's an illustration.
[1037]
Let me get the general picture.
[1040]
What we're studying
at the Max Planck Institute
[1042]
is how do people and should people
make decisions under uncertainty,
[1049]
and the first question
is a descriptive one:
[1051]
what's in the adaptive toolbox?
[1054]
There are many more heuristics
that I can't tell you today,
[1059]
and lots of social heuristics.
[1061]
So, people trust their doctor,
[1064]
and the study of ecological
rationality asks the question:
[1068]
in what situation is this
a good idea and when not?
[1072]
If your doctor knows the medical evidence,
[1075]
has no conflicts of interest,
and doesn't do defensive decision making -
[1080]
he has fear that you
might turn into a plaintiff -
[1087]
that's a good idea.
[1088]
But that's not the case in most countries.
[1091]
Most doctors - we have studies -
don't know the evidence,
[1094]
they have conflicts of interest,
and they do defensive decision making.
[1099]
And finally, how to create situations,
environments, and also strategies
[1106]
that are intuitive and that help people
make better decisions?
[1113]
Let me finish.
[1115]
Decision making
under uncertainty is different
[1117]
from decision making under risk,
[1119]
and heuristics are not the second
best strategies, that we often hear.
[1125]
They can do better
than even optimization strategies
[1129]
in a world of uncertainty,
not in a world of risk.
[1132]
And finally, more information,
more time, and more computation
[1138]
is not always better.
[1140]
Less can be more.
[1142]
Thank you for your attention.
[1144]
(Applause)
Most Recent Videos:
You can go back to the homepage right here: Homepage