Video summary
In this discussion, Judea Pearl distinguishes between correlation and causation, arguing that while probability measures how often events occur together over time or across variables, true understanding requires causal logic. He posits that human intuition naturally seeks a reason for why things vary together; if two phenomena do not see each other directly, their joint variation implies an underlying cause. Pearl emphasizes that conditional probability differs fundamentally from causation because conditioning on a variable is often merely the experimenter's choice to ignore certain incidents rather than a reflection of physical reality. He illustrates this with the example of flipping uncorrelated coins where introducing a bell that rings when one coin lands tails creates an artificial correlation simply by observing only those cases where the bell rang, demonstrating how observational data can mislead if causal logic is not applied correctly. The speaker critiques what he terms "naive science," noting that much of traditional research attempts to impose causal conclusions on correlational data without sufficient justification. He highlights specific disciplines like psychology and Applied Psychology as fields often plagued by this leap from correlation to causation, particularly when dealing with complex variables such as human behavior in semi-autonomous vehicles. In these modern studies, researchers face ethical constraints that prevent controlled experiments; for instance, one cannot ethically turn off an autonomous driving system on public roads just because a driver is tired. Consequently, scientists must rely on observational data where drivers choose whether to keep the vehicle active or deactivate it when fatigued, creating uncontrolled environments where inferring causation becomes difficult and prone to error. To contextualize this historical struggle with causal inference, Pearl references an ancient experiment from approximately 2,000 years ago involving Daniel and the Babylonian king. In that story, a group of exiled individuals requested vegetarian food instead of meat due to dietary restrictions, leading their overseer to conduct a test comparing their performance against those eating the King's standard diet after one week. The results showed the vegetarians performed better, effectively serving as an early experiment on whether specific causes (diet) affect outcomes (mental ability). Pearl notes that while ancient thinkers like Democritus recognized the importance of discovering single causes to understand reality, it was not until the 1920s that mathematics finally developed enough tools to rigorously capture these causal distinctions. Ultimately, the conversation concludes with a reflection on how classical physics and algebra often fail to address causality because their equations are symmetrical; an equality sign works both ways regardless of which variable is considered the cause or effect. Pearl argues that science has historically lacked the specific mathematical framework needed to express statements like "X causes Y but Y does not cause X," a limitation that persists despite centuries of inquiry into human behavior and natural phenomena. This gap between intuitive causal reasoning and formal scientific capability remains central to modern challenges in fields ranging from ancient dietary studies to contemporary autonomous vehicle safety, underscoring the necessity for new mathematical disciplines dedicated strictly to causation rather than mere correlation.
Read the full video transcript
what is correlation what is it so
probability of something happening is
something but then there's a bunch of
things happening and sometimes they
happen together sometimes not they're
independent or not so how do you think
about correlation of things
correlational kills when two things very
together over very long time is one way
of measuring it or when you have a bunch
of variables that it was very quickly
then recalled we have a correlation here
and usually when we think about
correlation we really think cosy things
and cannot be called as unless there is
a reason for them to vary together why
should they vary together if they don't
see each other why should they vary
together so underlying it somewhere is
causation yes
hidden in our intuition D is a notion of
causation because we cannot grasp any
other logic except causation and how
does conditional probability differ from
causation so what is conditional
probability conditional probability how
things vary when one of them a stays the
same now staying the same means that I
have chosen to look only on those
incidents where the guy has the same
value as previous one it's my choice as
an experimenter so things that are not
calling it before could become
correlated like for instance if I have
two coins which are uncorrelated okay
and I choose only those flipping
experiments in which the bell rings and
bell rings when it is one of them is it
tailed then suddenly I see correlation
between the two points because I only
looked at the cases where the bell rang
you see is my design with my ignorance
essentially with my audacity to ignore
certain incident
I suddenly create any combination really
doesn't think this physically right so
that's you just outlined one of the
flaws of observing the world and and
trying to infer something from the math
about the world looking at the
correlation I don't look at the floor
the world works like that which me but
the flows comes if we try to impose
causal logic on correlation it doesn't
work too well I mean but that's exactly
what we do that's what that's has been
the majority of science is your reality
of of naive science the decisions know
it the decisions know it if you
condition on a third variable and you
can destroy or create correlations among
two other variables they know it it's in
your data right nothing surprising
that's why they all dismiss the symptom
paradox ah we know it you don't know
anything about it well there's this
disciplines like psychology where all
the variables are hard to account for
and so oftentimes there's a leap between
correlation to causation your your leap
who is trying to get causation from
correlation not you're not proving
causation but you're sort of discussing
it and implying sort of hypothesizing
without liability which discipline you
have in mind I'll tell you if they are
obsolete is they are outdated oh they're
about to get outdated
oh yes tell me which one is all
psychology you know okay what is the ACM
no no I was thinking of Applied
Psychology studying uh for example we
work with human behavior and semi
autonomous vehicles how people behave
and you have to conduct these studies of
people driving cars everything start
with the question what is the research
question what is the research question
the research question do people fall
asleep
when the car is driving itself do they
fall asleep or do they tend to fall
asleep more frequently more fickle and
the car not drive lines not driving it's
it's a good question okay
and so you measure you put people in the
car because it's real world you can't
conduct an experiment where you control
everything
why can't you can you could do my
automatic a module on and off because
it's on Road public I mean there's yes
it's a there's aspects to it it's
unethical because it's testing on public
roads so you can only use vehicle you
they have to the people the drivers
themselves have to make that choice
themselves mmm-hmm and so they regulate
that and so you just observe when they
drive it and honestly when they don't
and then maybe they turning off when
they will very tired yeah that's kind of
thing but you you don't know those there
okay so that you have now uncontrolled
uncontrolled experiment we recall it
observational study yeah and we firm the
correlation and detected that we have to
infer causal relationship and whether it
was the automatic peace
it caused them to fall asleep oh so that
is an issue that they about 120 years
old yeah I should only go a hundred
years old and well maybe it no I
actually I should say it's 2,000 years
old because we have this experiment by
Daniel but the Babylonian king that
wanted them the exiled people from
Israel that were taken in in exile to
babylon to serve the king he wanted to
serve them King's food which was meat in
Daniel as a good you couldn't eat a
non-kosher food so he asked them to eat
vegetarian food but the key
overseer says I'm sorry but if the King
see that your performance falls below
that of other kids you know he's going
to kill me then you said let's make an
experiment let's take four of us from
Jerusalem
okay us vegetarian food let's take the
other guys that to eat the King's food
in about a week's time we'll test our
performance and you know the answer of
course he did the experiment and they
were so much better than the others if
the King's nominated them to super
position in so it was a first experiment
yes so today there was a very simple
it's also the same research questions we
want to know a vegetarian food assist or
obstructing your mental ability and the
question is very old even Democritus
said if I could discover one cause of
things I would rather discuss the one
cause and be king of Persia
did they task of discovering causes
what's in the mind of ancient people
from many many years ago but the
mathematics of doing this was only
developed in the 1920s so science has
left us often okay science is not
provided that with the mathematics to
capture the idea of X causes Y and y
does not cause X because all the
question of physics are symmetrical
algebraic the Equality sign goes both
ways
you