Tomas Petricek on Paul Feyerabend's The Against Method

[ techno/rock music playing ]

So yeah, um, when I read the annotations
for the papers we left me top says it's

about computer science and papers. I'm
going to break all the rules.

And this is a book. Here's a real copy.
And it's not about computer science.

But I still think it's relevant. And it
sort of—

When I was in Cambridge, the PhD only
takes 3 or 4 years.

And when I was in the last year, as every
PhD student, I sort of started thinking

"Hmm, what am I actually doing here?"
[ audience laughter ]

Not in the sort of meaning of life sense,
but more in the work sense.

I found it really interesting to read some
of the fellows of your science works,

because they're doing exactly that;
they're trying to understand how science

works, what's the right way of reasoning,
how should we do things, and how does

science actually work if you look at it
from the historical perspective.

And so this is one of the books that I
found just so refreshing.

When you spend 3 years writing equations,
this is the book you should read.

And I think it could have pretty much the
same effect on people who do a lot of

computer science or programming in
general.

I'll start with just a brief context, and
it fits in this Philosophy of Science

discipline.

Philosophy of Science is sort of inspired
by the idea that if you look at how

physics has been doing over the last 300
years, it worked pretty well.

We know a lot of things about physics. And
that's because physics found this nice

method called the scientific method.

If other disciplines want to get as far as
physics got, we should understand what is

the scientific method, like why does it
actually work, and learn how to apply it.

So that's the idea that sort of inspired
some of the Philosophy of Science, and

that means we want to understand what's
the method how people actually work, how

we build sort of correct theories, how we
learn how to verify the theories, but it

also means looking at the history and
seeing what sort of scientific work

happened and what can we learn from that.

I think there's not much that has been
done on philosophy and history of computer

science specifically, just because it's
not really that old.

There's some really interesting work done
on history of computer science, but

I think there's a lot more that could be
done in that area.

So, what I'm going to do here is I'll talk
mainly about Paul Feyerabend's book, and

I'll have various side notes linking it to
ideas from computer science where I think

it makes some sense.

Audience Member: How do we know that the
scientific method works?

How do we know that the scientific method
works? That's a really good question, and

that's one of the questions that's
questioned here as well.

There's been a lot of things in that
area; I just picked some of the books

I like, so this is not like an organized
scheme, but if you want to read some nice

book, here's a bunch of them.

One of the really early ones was Karl
Popper, who came up with this idea of

falsification. So, what he said is, if you
are proposing a scientific theory, it's

only scientific if it can be falsified.

So, if you're proposing a theory, it's
only science if there is a way that it can

be invalid.

If you're saying something that can't be
invalidated, that's not science.

Then a few people reacted to this.

Thomas Kuhn, with his Scientific
Revolutions, he sort of looked at this

nice, rational way of building science,
and said,

"Well, actually, sometimes this is not how
it looks."

If you look at history, there's moments
where just kind of people start thinking

about stuff in a completely different way.

So, this is kind of what happens in
Copernican Revolution, what Newton does,

what happens with the theory of
relativity, you just have to, not really

throw away everything, but it's more like
rephrasing everything that has been done

before, and building a new ground
underneath.

Lakatos, with research programmes, follows
kind of similar idea, but he says if you

look at science overall, at any single
moment there's multiple research groups or

research programmes where they have some
basic assumptions that they're never going

to question.

And that's kind of against what Popper
was saying, where he said anything you

should be able to question, anything must
be falsifiable.

In reality, scientists don't really
question everything.

They always have some core assumptions
that's just never false.

Like, whenever if you were getting to the
dangerous area, you always find some side

condition.

I guess you could sort of say in pure
functional programming, you have a core

assumption that you can write any code
without using mutation.

If you need to write some efficient h
table, you'll just say ahh, it's not that

I should use mutation here, it's just that
the compiler must do a better optimization

[ audience laughter ]

So, that's another fun thing there.

Feyerabend goes even further, and he's—
you'll see that in the rest of the talk.

The last one I put here, I find that also
interesting because that's more looking at

how if you look at science from the
historical perspective, or if you think

what has been done in science, and how
computers were invented, we have this very

ideal picture of how everything just
followed very nicely, linearly, towards

the ultimate truth.

This book is looking at how it actually
works when it's building.

So that's another point against this
beautiful idea of how perfect science is.

So, in Paul Freyerabend's book, he has
this phrase or slogan which is probably

the one thing to remember if you don't
want to remember it quite well, or if you

want to have slightly skewed picture of
the book then this is the thing to

remember.

But, he says, "To those who look at the
rich material provided by history—"

And I had to truncate it to fit it on the
slide, but it also says: "To those who

look at the rich material provided by
history, and who are not intent of

impoverishing it in order to please their
their lower instincts, their craving for

intellectual security in the form of
clarity, precision, objectivity and truth,

it will become clear there is only one
principle that can be defended under all

circumstances and in all stages of human
development,

it's the principle: anything goes."

I just wanted to read it because that
shows you how enjoyable the book is.

[ audience laughter ]

And he isn't really saying that there's no
order whatsoever; he had a clarification

later, where he says, "It's not really
that I'm saying that there should be no

order; I'm just saying that if you look at
the history, there's no single principle

that always applies."

And there are various principles that
work in some cases, and when scientists do

work, they follow some principles, always,
but there's just no single unifying

principle behind everything. And that's
kind of against this idea that physics

follows this beautiful scientific method
that we can just understand and copy.

Freyerabend says there's no such thing.

And he does do things in the books.

One is that he has his humanitarian
argument, where he tries to explain why

is it better for science, for mankind, and
for everyone, but he also has this

historical perspective which is mostly
around Galileo Galilei and how he managed

to convince scientists that heliocentrism
is the right way of modeling the universe.

I guess when you think of Galileo Galilei
many people think of it as this genius

scientist who was right and won over the
church, who were just following some

orthodoxy and were obviously wrong.

What's really interesting in Freyerabend's
book is that he actually tries to

understand it more from a historical
perspective, and tries to understand what

was scientific back then and how Galileo
actually in many places didn't really

follow the scientific method of the time.

So, he has again a beautiful quote. So
Galileo of course has some intellectual

reasons, there's some good motivation for
it, but he also uses propaganda and

psychological tricks to convince people of
his model.

And there's a number of really interesting
cases there.

So, something Freyerabend calls natural
interpretations, which is sort of what you

just believe in without heavy thinking.

Like, that's the background, and on top of
that, you can do science and think.

So if you were living in that time, the
first argument against the claim that the

earth is moving is that you don't feel it.

I go outside - nothing's moving with me.

I'm just standing. Like, I don't feel the
movement.

And if you were living in a world where
everyone believes that the earth doesn't

move, then that's a very valid point,
because if you get on a horse carriage and

it starts moving, you can feel it, right?

So the fact that you don't feel it is the
first sort of natural interpretation that

Galileo has to break and convince you that
actually, it can be moving even though you

don't feel it.

The other argument against movement of
Earth was the idea that if you climb on a

large tower and you throw a stone, then if
the earth was rotating then the stone

would fall somewhere further or closer to
the tower.

And, again, in the model that people
believed in back then, this is a perfectly

valid point.

And so what Galileo has to do is that he
has to invent completely new story about

how the world works, and the interesting
bit there is that he actually has to

combine two things at the same time.

So it's not like you could add one thing
and clarify how things work, and add one

more, and get closer to the right model,
you just have to convince everyone that

everything is different.

You have to come up with the idea that
every movement is relative, and if you are

on the earth, and the whole earth is
moving, you don't have to feel it.

And he also has to come up with the idea
that if the tower is rotating and the

stone is thrown from the tower, it will
continue rotating with the tower.

And this is like one place where you're
not just doing small addition to the

scientific image; you just have to change
the whole story.

So that's what Freyerabend calls natural
interpretations, and that's sort of the

background on top of which we are thinking
or doing science.

And I think this is kind of related to
Imre Lakatos, who has the research

programmes where you also have these basic
assumptions that you're never questioning.

And I think this is very common idea in
computer science.

I already talked about the pure functional
programming, where we are never going to

question the idea that we never need
mutation.

There's a bunch of other things, like
whenever you're modeling some computation

in functional programming, we have a
method for that, and that's called monads.

So we're going to just use monads without
really thinking about it too much.

There's one nice piece of work or
development that happened when people

started modeling dataflow computations.

There was a programming language called
Lucid, and someone came up with the

semantics for it that was built using
monads, because monads is how you model

computations.

And then later on, there was another paper
that said well, actually, this is silly.

Monads aren't helpful at at all for this.

And they came up with another model that
was ironically based on the dual, so it

was based on comonads.

But it was one of those cases where we
just follow some basic assumptions that

tell us how to do things, or how to
understand things, and it's sort of

obscuring other possible directions.

And this is something that a lot of people
in the academic programming languages

world have.

If you're talking about types and type
systems, type systems are the sound things

that prevent errors, and then you have
someone who comes up with language like

Dart, or TypeScript, that aren't doing
this at all.

And I think that's another case where the
sort of natural interpretation could be

slightly shifting, or there's people
breaking that idea, and you might think

they're just crazy and lunatics, or maybe
they're actually doing science like

everyone else.

The other interesting thing in the book is
how Galileo uses telescope to convince

people of his model.

The first thing here is that when he built
the first telescope, it was really

imperfect, and the images you would get
from it were in conflict with what you can

see with your eye, like it was stretching
some things, it wasn't accurate, there

were, like, you wouldn't see clearly, and
I guess if you think about this from our

modern perspective, we're used to the idea
that if you have imperfect lens, it will

distort the image of it.

We kind of know in what ways, like we know
it can make things shorter or longer, but

it will never hide something, or add
something to the image.

But that's something they didn't have.

So if you looked through the first
telescope and it was distorting the image,

you would be like, "Well, this is not how
it looks," right?

It could be adding things, removing
things, adding reflections, whatever.

So whenever people say Galileo built a
telescope and now he had this perfect tool

for convincing everyone that the world is
how he thinks, well, the telescope didn't

work.

We only understood how it works later when
we built more perfect versions of it

and we provided the additional theory.

So even the telescope was kind of fishy.

It's even worse with the telescope.

The other problem with the telescope is
that back then, the established scientific

fact was that celestial and terrestrial
objects are actually completely different

things; they are made from different
materials, they obey different laws.

So now-a-days, we have no problem in
believing that if you test the telescope

on the earth, on the ground and it makes
things bigger, it will also make celestial

things bigger, but that's just something
that wasn't true back then.

Because the scientific reality that was
just part of how people thought about the

world was that there's terrestrial objects
and celestial,

and they're different parts, different
things.

This is, again, where Galileo actually had
to do a lot of propaganda to convince

people that it makes sense to use the
telescope on the stars.

I think I have one more point on Galileo,
which is, again, I think really

interesting for people in computer
science.

That's when Freyerabend says that the
language in which we express our

observations may need to be changed as
well so that the new observations that

Galileo is reporting aren't sort of mixed
with the assumptions coming from the

old observations.

So when you're making observations, you're
expressing that in some terms that are

hardly based on your baseline theory.

And if you want to change the theory, you
have to come up with a new way of

describing the observations as well, like
you have to come up with new terminology

that doesn't distinguish between the
celestial and the terrestrial objects

and so on.

I think this idea of language and how you
have to carefully avoid mixing what you're

doing in the new theory with what's known
in the old theory is something that

matters in computer science as well.

There's one kind of funny but interesting
example: in F#, there's this thing called

'computation expressions,' which you can
kind of think about it as syntactic sugar

for monads, and every time someone talks
about computation expressions in F# in

front of high school familiar audience,
they get the question: "So is this

just monads?"

On some technical level, they're very
closely related and answer might be yes,

to some extent. It's not quite that;
there's technical differences as well,

but also, if you just use different names,
it gives you the freedom to build whole

new understanding and intuition around it.

Because monads is like very, I guess,
poisonous word, like it has this massive

amount of assumptions and basic ideas
around it that you can't say,

"This is all wrong," because it's not, but
if you want to talk about something that

has different intuition around it, you
just have to use different language.

I had another interesting case, where in
some of the F# type providers, I called

something 'union type' because it had some
shared structure with what's normally

called union types, and in one of the
reviews I got beautiful comment saying

"This is just all wrong, because this is
not what union types should be like."

Again, this was I think like, I picked the
word because I saw the similarity; they

just didn't see the same similarity I
would see there, and again, it's one of

those structures that has a lot of
assumptions around it, how it should work.

I guess what's the important thing to
learn from the book here, at least for me,

is that when we talk about things, the
language is actually really important.

Like, when I'm using some word, I'm not
just getting the technical definition or

some core of the idea, but I'm getting all
the infrastructure around it.

So that's another sort of inspiration that
I think is really well described in

the book.

What Freyerabend in the end says about the
Galileo trial is that when the Inquisition

came to see if what he's saying is
sensible or not, they declared the

doctrine to be foolish and absurd in
philosophy, which was, back then, a way of

saying, "This is unscientific."

And based just on what was the
scientifically established fact of

the time, they concluded that he was just
wrong.

This is not criticism of Galileo, like I'm
not saying he was wrong, obviously, but

I think it's a good example that shows how
sometimes what is scientific isn't

necessarily— or what's unscientific, well,
I think I have it on the next slide.

Unscientific doesn't necessarily mean
wrong.

It might be correct; it's just that the
science doesn't have the right method to

explain it, or doesn't have it yet.

It will generally just take a bit longer
until we build more tools, more theories,

more machinery, and then it will be right
and scientific.

But there's this phase where things can be
actually right but not scientific,

because even science has some limitations.

And the usual amazing Freyerabend quote is
that, "Science is much more 'sloppy' and

'irrational' than its methodological
image."

So we like to think that science is
perfect, and scientists always have to

write answers, they're getting closer to
the truth.

And I think what Freyerabend demonstrates
here with the Galileo case is that,

actually, it's not that easy.

Question?

Audience Member: Yeah, so are you implying
that scientific necessarily is right?

Am I implying that scientific necessarily
means right?

[ audience member speaks quietly ]

Will have to give you mic.

[ audience laughter ]

Man: Here you go.

Audience Member: But, by the same token,
Bruno got burnt at the stake for pretty

much the heliocentric view, so to say that
inquisition science is like a predate,

it's more, right now we don't burn people
at the stake, usually, for scientific

mistakes.

Petricek: No, we just don't give them
tenure.

[ loud audience laughter ]

Audience Member: Right, but in a certain
sense, I see as most of today's argument

is basically general science coming under
the pressure of the insights of Heisenberg

and Gödel.

Petricek: Alright, um, so I don't think
we—yeah, I definitely agree we don't burn

people anymore.

[ audience laughter ]

But I think as far as scientific
reasoning can be judged back then, there's

definitely some very valid point.

The previous thing, do I say that
scientific is right?

Audience Member: Do you value the term
'scientific'?

Petricek: Yeah, I think there's, well, I
definitely think there is some value in

the term 'scientific,' otherwise I
wouldn't be doing it for my PhD, right?

[ audience laughter ]

And I think even back then they had some
notion of correct reasoning, like correct

way of thinking, and did you think truth,
and... yeah.

So it wasn't the modern science,
obviously, but it was—and again, the

argument here isn't really whether the
Inquisition should have banned Galileo

from publishing his writings - it actually
goes the other way around.

It's more like even back then, what we now
think that that's the perfect scientific

approach, if you look a bit more carefully
it wasn't scientific according to the best

idea of what would have been scientific
back then.

Audience Member: Which was Aristotelian
methodology.

Petricek: Which was Aristotelian
methodology.