The Philosophy of Science
A Short Introduction
If I were to ask you “What is science?”, chances are your answer would be something along the following lines:
Science is a discipline that tries to predict and explain the world using critical thinking, empirical methods and/or mathematics.
You’d be right. There are roughly 3 kinds of science — Natural (the study of natural phenomena such as physics, chemistry, biology, geology); Formal (the study of logic, mathematics); Social (the study of human behavior and societies) — and all of them try to emulate this at its core.
“Science” wasn’t always like this though. People in the medieval ages thought of cause and effect in terms of witchcraft, prayers, and Gods. Until recently, we used to think the Earth was the center of the Universe and that Humans were the center of creation. We only started studying Mathematics properly after Galileo, started searching for scientific Universal Laws after Newton, and adopted Experimentalism with Francis Bacon.
This evolution of science — the questioning of authority, the challenging of the status quo, the focus on the practical — go hand in hand with the Scientific, Cultural, Political and Industrial Revolutions of the past few centuries. But this focus on critical thinking (making claims based on sound arguments and justification) has brought its share of philosophical puzzles. For example, if we could predict everything in the universe perfectly using impersonal laws, we would predict your future course of action. Does that mean free will is only an illusion?
Philosophy is “an attempt at making rational sense of the fundamental aspects of the world and our experience” [Fermin Fulda]. We will only focus on the philosophy of science here. We can loosely define 5 views of thinking about science: Empiricism, Rationalism, Historicism, Realism, and Naturalism. We’ll explore these in-depth throughout this article.
The key relevant questions are:
1. What is science?
2. What makes science better?
3. How do you differentiate science from pseudo-science?
4. How should we do science?
5. Should we do science?
6. Is science objective?
“But why?”, you might ask. Science has worked well enough for technological advances and other practical reasons. On one hand, there were pioneers like Richard Feynman who famously said: “shut up and calculate!”. He didn’t believe that scientists should worry about philosophy. On the other hand, there are illustrious minds of the past centuries, Einstein for example, who thought the study of the history and philosophy of science was essential for any scientist. Note that the camps did not disagree that these are very important questions, they disagreed on whether they should be the ones handling it.
I think the philosophy of science is important because it lets us have more meaningful conversations about the role of science and scientific theories in our society. Science is the torch-bearer of progress and has obvious implications in policy-making too. If we do not understand the mindset and methodologies of science, we commit the fallacies we so often see in history and think, “Were these people really that stupid?” It will let you answer (or at least attempt) questions like “Why should we believe in theories that are explained using sub-atomic particles, but be wary of the ones that explain the same phenomenon using angels and demons?” In particular, thinking about the philosophy of science will let you think critically about “scientific ” claims you encounter every day — be it in international conferences or popular press. Science has shown us time and again, that our best theories are proven false at one point or the other: so the act of questioning the principles guiding a scientific claim is very scientific in itself. However, there is a fine line between a healthy dose of skepticism and conspiracy theories: the recent Flat Earth Movement, Anti-Vaxxer Movement, Alt-science Movement, etc. show how a good intention (well… at least they claim they’re searching for the truth) can spiral down to complete lunacy if people don’t understand the philosophical underpinnings of what real science looks like.
Nobody wants to be on the wrong side of history and learning about the philosophy and history of science might just give you a leg up. Seeing that you’re still here reading this, I bet you’re interested. Let’s dive in!
Empiricism and the Problem of Induction
There are two basic types of scientific inference: Deductive and Non-deductive.
Deduction starts with a few premises and uses an argument to reach a conclusion. For example; Saif is typing on a keyboard (1) and he is on Medium’s New Stories page right now (2). Therefore, he is writing an article. This is rationally persuasive. The argument is valid and since the premises are true, the argument is also sound. You can’t disagree with the argument. You may question the premises but if they are true, I’m right. My argument is infallible.
There’s a problem though: we have learned nothing new per se. We have just taken the premises and transformed them in a way to make a fact more apparent. Our circle of knowledge has not expanded. And science, as we know today, wants to discover something new and predict the future.
Enter, Induction. It’s a form of non-deductive scientific inference where we look at the past to predict the future. For example; Saif has been writing on medium every month for the past few months (1). Therefore, he will write on Medium next month. This is not rationally persuasive or deductively sound. You don’t really know whether I’ll be writing next month. Perhaps I’ll die, perhaps I’ll get bored with writing. But you have learned something new. Of course, this is a silly example and scientists deal with more pertinent phenomena when they use induction.
That brings me to my next point: statistics in inference lets you be confident about your inference to a certain degree. If Saif had been writing for 2 months when you wrote this argument, it’d be debatable. If Saif had been writing for 2000 years when you wrote this argument, it’d be quite convincing. That’s exactly what scientists are doing: looking at the past, to predict the future, and then waiting till the experiment is over to see whether their predictions were right. Using induction in this way, we can either make universal generalizations and/or single predictions.
A major chunk of science is trying to do Causal Inference today. We base it on the Principal of Uniformity of Nature — that nature will behave tomorrow as it did yesterday. This is where we find interesting theories. Smoking kills. Careless resource consumption is causing climate change. Vaccination doesn’t cause autism. How do we know these facts? Well, look at the data for the past 100 years. Using statistics, we can say to a high degree of confidence that these claims are right. But it is still using induction, and that’s its biggest weakness. The biggest counterexample to Induction is Newton’s Universal laws. We thought we’d figured the universe out, that Physics was done and dusted. Until Einstein came along a couple hundred years later with his theories of Relativity and proved Newton wrong.
You see, Aristotle, the ancient Greek Philosopher — one of the forefathers of “science” — believed in pure reasoning. According to him, you had to get to any conclusion solely based on axioms and deductive reasoning. Yet induction feels very strong intuitively. That’s how we make connections in our brain and survive — Don’t touch the fire because last time I did, it hurt. Could we perhaps prove that induction is okay, and that we’re all not insane?
David Hume says no. His argument, which is known as The Problem of Induction, goes like this:
Deduction disproves induction because it is non-deductive (1). However , induction has worked very well until now, so induction works well (2). Line 2 can’t prove induction because we are using induction to prove induction, which is a cyclical argument (3). Therefore, we can’t prove the validity of induction.
Does this mean that induction is a secular faith, just as medieval as witchcraft? Does this mean modern-day science is only an animal instinct based on behaviorism? Something feels wrong about this argument. Nobody has cracked this puzzle. Yet, today’s experiments in Physics predict results to dozens of digits after the decimal point. How do we explain that?
[if Hume’s problem can’t be solved]… there is no intellectual difference between sanity and insanity. — Bertrand Russell
[Induction is] the glory of science and the scandal of philosophy. — C. D. Broad
An interesting aside is that Confirmation — to say that a hypothesis is correct — takes the form of induction; and since induction isn’t rationally persuasive, we can never prove something in science. We can’t say this theory is correct because it has been working so far. Nothing stops a counterexample to the theory from being discovered tomorrow. Disconfirmation — to say that a hypothesis is wrong — takes the form of deduction. If I say there are only white ducks in the world, you can show me a counterexample and that would be deductively sound. Therefore, we can only disprove something in science.
So is science just a goose chase, where we’re continually trying to prove past science wrong? Karl Popper says yes, and we’ll see his philosophy in the next section.
Falsification and The Problem of Holism
Karl Popper, one of the greatest philosophers of science, sided with Hume’s conclusion: induction, and thus the inductivist/empiricist way of doing science, was rationally invalid.
He proposed another method — empirical falsification. Here’s how it works:
(1) You make a bold/risky hypothesis H.
(2) You make a prediction E based on H.
(3) You gather evidence to test E.
(4a) If E passes the test, then H is corroborated, but not confirmed.
(4b) If E fails the test, H is wrong and disconfirmed.
Popper was a great admirer of Einstein’s work, and you can see how his theories of relativity fell under the Popperian framework of falsification. Popper gets rid of the induction -problem by making science a purely deductive exercise of falsification (or so he thought…). This is similar to how a lot of textbook science is done nowadays, starting with a null hypothesis and using confidence intervals, statistical tests, etc. The use of statistics makes it not-so-purely Popperian, as Popper would say just one instance against hypothesis would be enough for falsification.
This finally gives us a way to distinguish science from pseudo-science. Think of the difference between physics, astrology, economics, and homeopathy. Which of these are sciences and which are merely “faking it”? By Popper’s method, we’d have to ask the discipline whether they’re making risky hypothesis followed by specific predictions which can be falsified. Take a moment to think about these disciplines with Popper’s framework.
This sounds great. Scientists can finally be rational again! Yet this is not always the correct approach. A recent example of how Popper falls short is the Grand Sasso Experiment, which claimed the discovery of neutrinos that broke the speed barrier of light. With no evidence, the pioneers of science today threw doubt on the experiment instead of questioning the validity of current theories. 3 years later, it was discovered that clock-drift introduced error to the calculations. The light-speed barrier was still intact.
The Problem with Falsification is that you can never test your hypothesis H in isolation — your experiment always makes a set of assumptions A. When your prediction fails, anything in the network of [H+A] could have caused the failure. This could be because of some confounding factors unaccounted for. We call this the Problem of Holism, formalized by Duhem and Quine. The duo argues that even though falsification (by deduction) and confirmation (by induction) were logically asymmetric according to Popper; they are now practically symmetric with this new problem.
Holism thus denounces logical falsification (by Popper) and adopts practical falsification. Practical falsification says you can blame/modify A (ad-hoc changes) to save H, but never do so just for the sake of saving H. It says if A is highly corroborated, keep A in lieu of H.
Popper was trying to get rid of induction from science, but he failed because of Holism. Holism tried to address the attitude of scientists towards decision making. But it fails to address the problem of induction too. Deciding to keep A instead of H because it is highly corroborated is akin to extrapolating H’s future success judging by H’s past success — induction! We’re back to the problem of Demarcation — how can you tell science apart from pseudo-science? How can we ever justify our beliefs and assumptions rationally if they all are based on inductive reasoning?
Let’s look back again on the various disciplines and try to think about them critically, even with all our logical weaknesses: Chemistry vs Alchemy, Evolutionary Theory vs Intelligent Design, Marxism, Psychoanalysis. What about String Theory, Multiverse Theory? I’ll close this section with a quote from the man himself, Karl Popper.
Science does not rest upon solid bedrock. The bold structure of its theories rises, as it were, above a swamp. It is like a building erected on piles. The piles are driven down from above into the swamp, but not down to any natural or ‘given’ base; and if we stop driving the piles deeper, it is not because we have reached firm ground. We simply stop when we are satisfied that the piles are firm enough to carry the structure, at least for the time being.
— Karl Popper
Historicism — Normal and Revolutionary Science
Until this point, we’ve had a heroic picture of what science is. The main disagreement was between Rationalists and Empiricists on the use of induction to further scientific knowledge.
An American philosopher, Thomas Kuhn, and his theory of Historicism challenged that image of the truth-seeker scientist. Kuhn argues, on empirical grounds, that the rationalism and empiricism proposed by philosophers were rosy versions of what science ought to be (prescriptive), but very far from what science is (descriptive). Kuhn claims that science is heavily influenced by social and psychological factors, rather than purely by logical or empirical factors as scientists would have you believe. Kuhn says that the scientific standards change over time and play out a cyclical pattern of Normal Science followed by Revolutionary Science.
It is no surprise that these ideas, published in Kuhn’s The Structure of Scientific Revolutions, attracted significant backlash from within the scientific community. A lot of people were distorting the writings to make this a showdown between Popper and Kuhn, and Kuhn spent the rest of his life refining and justifying the contents of his book. Let’s take a look at the summary of his argument to judge it more objectively.
Kuhn divided the timeline of science into 5 phases:
1. Pre-Paradigm — a dis-unified set of practices, theories, and methods.
2. Normal — a stable, accumulative, puzzle-solving period that adheres to an agreed-upon paradigm.
3. Crisis — loss of confidence in the paradigm because of anomalies.
4. Revolutionary — the unstable, non-accumulative period where a new paradigm overtakes the community.
5. New Normal.
Straightforward, if you understand what Kuhn means by “paradigm”. And he uses it across the book in more than a dozen ways. If we look at the paradigm in simpler terms — in a broad sense and in a narrow sense — that should suffice for this article. In the narrow sense, a paradigm is an exemplar; it is an impressive achievement that proposes new puzzles, solves some puzzles and becomes the yardstick against which you measure the rest of the work during the current normal science. In the broader sense, a paradigm is a way of doing and thinking about science — methodology, epistemology, ethics. To be a paradigm, an object can’t only have new laws, theories, and observations — it has to use these laws, theories, and observations to solve puzzles better than the paradigm it replaces. Think Copernicus, Newton, Darwin, Einstein.
So what makes a piece of science a paradigm? Popper likes to believe it is pure rationalism. Kuhn says it is much more complicated than that. If falsification was all there was to science, wouldn’t there be disruptions in scientific fields way more often? No, the disruptions are influenced by society and psychology, says Kuhn.
Kuhn is saying that the Normal Science, which spans a large part of the cycle, is being ignored by the philosophical community because they’re engrossed by heroic/revolutionary science. That is descriptively false and during normal science, scientists gravitate towards conservatism because they’re solving puzzles instead of testing theories.
As the years of normal science unfold, anomalies build-up. At first, they’re ignored. But they become so problematic at one point, people lose confidence in the current paradigms of normal science. The conservatism wears off. A crisis emerges, and a revolution brews in the shadows. New theories compete to become the new paradigm.
On Revolutionary Science, Kuhn holds a radical, Kantian view of the world: the world is what we interpret it to be. The people who bring about the revolutions view the world differently, and as the new perspective solves more problems than the last, people see the world differently too. In this version, you can think of science as Gestalt Switches or Religious Conversions.
The concept of incommensurability explains this further. The theories that arise in the period of normal science are commensurable to one another, but the theories that arise due to crisis and revolution are incommensurable with the previous standard. Kuhn maintains that this incommensurability can be one of language since the different paradigms live in different realities (since they see the world a different way).
Slowly but surely, one of these new competing paradigms comes out on top. For Kuhn, the standard by which to judge a paradigm was predictive accuracy, plausibility, simplicity, consistency. But these are hardly neutral and specific. This gives rise to relativism — the truth of a claim depends upon extrinsic factors — politics, society, global trends. The Revolution gives us the ability to solve more puzzles, but that does not mean we’re any nearer to the “truth” since this choice of the paradigm is not purely rational. You can measure the progress of science during normal science easily, but there seems to be no neutral measure that can judge the merit of new paradigms that ultimately replace the old. During the Revolution, you need anarchy……
A new scientific truth does not triumph by convincing its opponents and making them see the light, but rather because its opponents eventually die, and a new generation grows up that is familiar with it.” — Max Planck
As you can imagine, this theory made scientists very uncomfortable and defensive. We’re back to the problem of Demarcation if Kuhn is right. How is this indoctrination and conservatism of normal science any different than pseudo-science and religion? How is this irrational paradigm choice bringing us closer to the truth?
Sir Karl has characterized the entire scientific enterprise in terms that apply only to its occasional revolutionary parts. His emphasis is natural and common: the exploits of a Copernicus or Einstein make better reading that those of Brahe or Lorentz. — Thomas Kuhn
TLDR Popper claims verification is not possible in science, that only falsification is the way to go. Kuhn claims that even falsification is not bringing us closer to the truth thanks to shifting paradigms and ever-changing standards.
All this is not to say that a bunch of scientists are as bad as a cult. The strength of science stems from ordered cooperation between communities during normal science and the possibility of de-stability during the revolution. Science may not be heroic, but it surely aspires to be.
Realism and Anti-Realism
Faced with all these seemingly intractable questions, I throw my hands up in frustration and say, “Whatever! We live in the Matrix for all I know!” And I get sucked into another debate about science, one about reality…Not again…
Realism is the belief that truth exists in a mind-independent way. This laptop I am writing on definitely exists, regardless of what you think. Anti-realism is — you guessed it — the exact opposite. Anti-realism claims this laptop exists in my reality, and it doesn’t matter whether it really exists.
There are several types of realism. Common Sense Realism claims everyday objects exist because we can observe and manipulate them. Mathematical Realism claims that mathematical numbers and functions exist in reality. Moral Realism says good and bad are fundamental. Our topic of interest — Scientific Realism — claims that unobservable objects such as protons, neutrons, atoms, and black holes exist. Scientific Realism uses Common Sense Realism as a premise. (It is good to note, however, that Common Sense Realism is not unchallengeable. A big question is what entity makes this reality? What entity decides the ground truth? Radical Skepticists like Rene Descartes argued that a demon might be controlling your mind and creating an illusion of reality, The Matrix explored this concept in an epic trilogy, and recently Elon Musk said he believed we’re actually in a simulation. Yikes. Let’s not go there and focus on Scientific Realism for this article.)
To predict and explain scientific phenomena in the observable universe, scientists often postulate unobservable entities. For example, the atom was postulated by Dalton to explain the behavior of matter and Uranus was postulated to explain the orbit of Saturn.
Scientific Realists believe that:
1. observable and unobservable entities exist mind-independently.
2. our best theories give us an increasingly accurate approximation of the truth.
3. the main aim of science is to seek the truth.Scientific Anti-Realists believe that:
1. unobservable entities don’t exist mind-independently.
2. observable theories should be interpreted literally, but unobservable theories should be interpreted instrumentally (i.e. they need not be a good representation of the truth so long as they as useful for models and predictions.)
3. the main aim of science is empirical adequacy (truth about only the observable aspects)
This clash is old. The Church had no problem with Galileo using Copernicus’ Heliocentric model instrumentally to predict movements in the solar system (anti-realism), but they strongly objected to his assertion that the heliocentric model explained the solar system as it actually was (realism). You can see the same clash when reading the history of physics, atomism, quantum mechanics, genetic biology, etc.
Realists give a rational justification for inferring unobservable entities from empirical evidence, based on the concept of Abduction/Inference to the Best Explanation (IBE)/Eliminatory Inference/Explanatory Inference. Like induction, it is deductively invalid. Unlike induction, abduction is not based on statistics but on the explanatory power of theories. Given evidence E and candidate explanations H1,.., Hn, abduction infers the truth of the Hi that best explains E. Abduction does this by considering the simplicity, consistency, coherence, and scope of the competing hypotheses.
Realists put forward an abductive No Miracle Argument (NMA) in favor of their view:
1. Our best theories are extraordinarily successful.
2a. They are so successful because they’re true, OR
2b. They are so successful thanks to miracles.
3. Miracles are not really explanations, so 2a must be true.
Thus, our best scientific theories are approximately true.
There are very strong counterexamples to NMA. Perhaps the first one that comes to mind is Newtonian Mechanics. It got us to the moon, and that speaks to its practical success. But we know Einstein showed that it is strictly false, and that speaks to its empirical failure. Other counterexamples include the Classical Elements, the Phlogiston Theory, Caloric Theory, and to some extent Dalton’s atomic theory too (we’ve found smaller building blocks!). This leads to what the Anti-Realists call a Pessimistic Induction. The argument goes like this: Our best theories from the past have eventually been replaced, so our best theories from today will also be eventually replaced. Thus, our current theories are also most likely false. (Two counter-counter-arguments exist. Some people say that the premise is too pessimistic — that restricting the argument to the better theories will yield a less gloomy outcome. Others attack the outcome saying it assumes that we have not gotten better at doing science compared to the past — an assumption that just feels wrong.)
Anti-Realists believe that the NMA commits Base-Rate Fallacy, and there is no way for redemption since it is impossible to know how many of the current theories are true. (A counter-counter-argument from realists is that Base-Rate Fallacy interprets NMA as an inductive argument which makes statistical considerations when, in reality, NMA is abductive and explanatory. But I think we’re too far down the rabbit hole already. My head’s spinning, too.)
A more fundamental challenge to the NMA is one that questions abduction‘s validity in arguments, thanks to its non-deductive roots. This is the problem of underdetermination. That two hypotheses are logically inconsistent (they posit different unobservable entities) but are empirically equivalent (they make the same observable prediction) means that there is no logical reason to pick one over the other. Thus we say that our empirical evidence underdetermines which theory is true. What about a completely different theory that scientists today cannot fathom? (To all of this, realists point out “So what justification do you lot have for induction?” Ouch.)
“The method of ‘postulating’ what we want has many advantages; they are the same as the advantages of theft over honest toil.” — Bertrand Russell
There is a discontinuity between the theories regarding observable and unobservable in the mind of anti-realists. In contrast, realists believe that there is a continuum since the line between the observable and unobservable is also rather blurred. We say that entities we observe using an MRI Scanner, an X-Ray scan, a microscope, or a telescope are “real”. How about an electron microscope that sees electrons or a world-sized telescope that clicks a picture of a black hole? After all this debate, the question “How should we interpret the relationship between scientific theories and reality?” remains unsolved.
Naturalism
Now that we know a bit about the mindsets of scientists and methodologies used in science, we can move onto one of the most heated debates in history: Science vs Religion. To understand the two stances scientists hold in this debate, we must first understand naturalism.
There are mainly two kinds of naturalism — methodological naturalism and metaphysical naturalism. Methodological naturalism argues that science is the study of “nature” — the objective reality we all interact with — but implies nothing about the existence (or lack) of the supernatural. This means that “The Canon of Science” is a subset of “The Canon of Reality”. Metaphysical naturalism is a lot more philosophical in its claim that there is no supernatural reality out there. Something is in the realm of science if and only if it exists in reality — “The Canon of Science” equals to “The Canon of Reality”.
There are counterarguments to both the claims, as in all things philosophical. The problem with methodological naturalism is that the definitions and boundaries are very vague. How do we categorize what is natural and what is supernatural? Is it the case that the supernatural never interacts with the natural? What if they do? Would we push the natural into the supernatural, or pull the supernatural into the natural and then study it? Some things are easy to categorize: God is on the supernatural side of things, while water is on the natural side of things. But what about Truth, Morality, Beauty, Consciousness, the Mind, Free Will? We may get away by saying that we keep religion apart from science to get rid of the conflict but what happens when religious scriptures/gurus make claims about our objective reality? What about the problem between Creationism/Intelligent Design and Darwin’s Evolutionary Theory?
The motive of methodological naturalists seems to be avoiding a conflict between science and religion — after all, there exist scientists who are also religious. Another motive for the distinction is to get rid of supernatural explanations from science — something rampant before modern science came about. The problem with supernatural explanations is that they contribute nothing to our objective understanding of reality, however far from the truth they may be. If a scientist is faced with a big-but-surmountable problem she’s been trying to solve for years, and she has the license to evoke the supernatural, it is certainly the path to least resistance. Supernatural explanations slow down and hinder science. The other problem is that it is impossible to know when you’ve hit the barrier between the natural and the supernatural because (1) there is no guarantee that science can explain everything observable and (2) there is no clear definition of the “natural”.
This is not to say metaphysical naturalism — which is often touted as a scientific conclusion — is right either, precisely because it is philosophical and has no scientific backing. It can not have a scientific basis by the very definition of “The Canon of Science”. There is a strong desire among many people to perform this reduction from methodological naturalism to metaphysical naturalism because there is a certain comfort in knowing that we can eventually figure everything out.
“Science is wonderful at destroying metaphysical answers, but incapable of providing substitute ones. Science takes away foundations without providing a replacement. Whether we want to be there or not, science has put us in the position of having to live without foundations.” — Hilary Putnam
The debate between intelligent designers and evolutionists is very constructive in understanding the arguments for and against both types of naturalism. I’ll refer you to Aquinas Arguments For God, Paley’s Watch, Darwin’s Theory of Evolution, Evolutionary Psychology, the Theory of the Mind, etc. if you’re curious — because all these are long-fought battles and beyond the scope of this article. Just make sure you look at both sides of the argument.
I hope you can now look back at the questions I posed at the beginning of this article and at least know where to begin. I hope you can now look at science and know that we haven’t figured a lot out yet, as with everything man-made. I also hope that we, as a society, can celebrate science — even with its flaws — as it’s the only tool we have right now to get better and better approximations of reality.
I’ve had a lot of fun taking a detour from the common Computer Engineering things I do and delving a bit into Philosophy. Unfortunately, for now, this is where I leave you to do your own research. Bon Voyage!
This piece was inspired by the course HPS120: How to Think About Science, taken at the University of Toronto during Fall 2019, under Fermin Fulda.
