What Is Science?

It has, over time, become fashionable to appeal to science as an authority in all areas of public discussion. Although the technical phenomenon is driving this trend further as time goes on, it appears to anyone who asks that the greater part of people cannot answer exactly what is meant by science. Even students of a science or related field are frequently unable to give an adequate definition.

Given the prevalence of the issue and poverty of relevant material (take a look at some dictionaries' definitions for example) I think it's worthwhile to put forward my own thoughts on what science is, what makes a fact scientific and what separates a scientific theory from any other theory.

Some parts of this may look needlessly roundabout ways of expressing what may seem obvious, however, I think it's necessary to phrase it so formally to be adequately thorough.


The most immediately reasonable place to start would seem to be that a scientific fact is a proven fact, and a scientific theory is a theory proven to be true. I think that's the sort of answer you would get if you asked the man on the street, so to speak, so I think it's important to deal with it first.

First, it's important to be clear what is meant by proof and proven, because otherwise it will be impossible to actually use that definition of science. A proof is a positive conclusion reached through the application of valid inferences to axioms, and in this scenario we'd ideally like our axioms to actually be true, since science is hardly defined as either inherently false, if the axioms are false; or totally indifferent to truth and just any possible conclusion that can be reached, if the the axioms do not matter. Sans jargon, that means a proof is the end result (Socrates is mortal) after starting from a set of facts (Socrates is a man, and all men are mortal) using some set of rules (if A is a B, and all Bs are C, A is C).

This idea presents one serious problem, and another more serious problem. The serious problem is of the axioms, those initial assumptions. How do you know they're correct? If you want to try and prove them you're going to have to come up with new axioms and the same problem will apply to those, unless you want to introduce circular reasoning or infinite regress somewhere which is unsatisfactory. Ultimately in this case science will have to rest on some set of arbitrary assertions that themselves aren't scientific, meaning any scientific theory would be unscientific in the foundations.

Importantly, this means that one scientist could start with different assumptions and end with a radically different view of the world that we'd have to admit was just as scientific as our own, as long as he followed the rules in reaching that view.

Note that mathematicians and other logicians don't have this problem: a mathematical proof deals with axioms formulated by mathematicians that don't claim to necessarily represent a physical reality that exists independently of the mind, or world of forms. (They very often do represent it, but what's important is they don't have to in order to be correct.) A different mathematician could start from different axioms and arrive at radical new conclusions, and as long as it deals with quantity that's fine; it's all still maths. A scientific proof doesn't have this freedom, it's supposed to directly correspond to a separate, "outside" reality and two people can't really have very different, even opposing answers if that reality is supposed to be consistent.

But that's just the trilemma; there is a way out if we can maybe just all agree some axioms to be scientific and be done with it. The more serious problem is induction. Another quality that separates science from the world of pure logic is that a logical operation is the same every time you do it. One and five is always six, and you can't wear the numbers out. By contrast, the physical world has no rules we can directly observe. This was Hume's famous observation.

You can never be sure that some apparent relation being witnessed is causal, or just a repeating coincidence, and because of that there is no satisfactory analogue between logical operations required in proofs and scientific operations, like observation and experiment.

This isn't all some dry academic exercise in pedantry either, both of these have real-life impacts and examples. The classic example of induction is that people used to think all swans were white, until they found black ones, demonstrating how even if every swan on your entire continent is white, it's still not logically sound to generalise that fact to claim "swans are white". Note that starting from a different continent, another scientist might conclude that all swans are black, an example of the radically different conclusions reached earlier.

A more subtle example is Newton's law of gravitation. For many years it was scientific fact that gravity was a direct, attractive force with a strength proportional to the masses involved and the inverse square of the distance between them. It was experimentally observed in many cases and even today is taught in undergraduate physics classes; the problem being, it isn't always true.

Masses don't actually attract proportional to their masses and an inverse square law, although they very nearly do and because that formula is such a good approximation of what they actually do, it's hard to tell it apart from the real behaviour in all everyday situations. But it's still not the underlying truth: the general rule. The truth is that even what we have now, the theory that superseded Newton's, might not be the truth either because there is no way of knowing for sure that what we have is not "as good as" truth (which may be the same thing as truth depending on your own ideas) and thus there is no such thing as science-as-proof. There might always be some new underlying rule that is very close to the scientific theory but deviates in some manner that goes unnoticed, because all the black swans live in Australia and the non-Newtonian gravity is only noticeable far from Earth.


Another idea is that science originates in the opinions of experts, or, scientists. On further examination, however, this does not hold up.

The first place to start is to imagine asking experts if they always retained scientific knowledge or if they learned it somewhere. Clearly, nobody could honestly respond with the latter; in that case the only study or research such a scientist would have to do to produce new findings would be of their own thoughts.

That leaves the only other possibility that the scientists must have gained this scientific knowledge from somewhere, that is, it was learned. If we are to maintain that science really is the opinions of experts, we must place its ultimate origin there as well. Otherwise, we would have it that science is only contingently derived from experts, but also derives from another source, which is presumably more fundamental than scientists since it precedes them in having this knowledge.

In this case of learning, there are four possibilities as to the origin of the knowledge, which is what we're interested in: It can be self taught, come from an infinite line of teachers, come from a finite but circular line of teachers or the line of teachers can terminate in some primordial ur-scientist. The case of being self taught is equivalent to the case of being innate knowledge we disregarded earlier. The cases of the infinite, circular and terminated lines of teachers are blatantly nonphysical and nonsensical.

Since all of these options are hardly possible, never mind reasonable, a proponent of this theory is forced to admit science must originate somewhere besides the human mind. The most obvious out is that the aforementioned more fundamental source of science is nature, or empirical reality, which makes the opinions of experts most often scientific simply because they have access to the most knowledge, but defining science as knowledge of nature lands us with the first section and its problems of the nature of knowledge.


Another idea is that a scientific claim is a falsifiable claim. This is a good one, but flawed. For one, many accepted, even foundational, scientific ideas are totally unfalsifiable. For example, time/space translation invariance can never be disproved since any apparent variation can be explained as the result of unknown laws or rules [a]. We might assume everything falls at an identical rate, and find this is not the case on different planets to Earth. However, this is not an exception to some law of rates of falling, but the result of a more general rule that contradicts the hypothesis — a theory of gravity in this case — and likewise for any apparent violation of the principle: we will always just assume it is the result of a more general rule that our hypothesis has missed.

The reason that assumption will always be made is due to one of the fundamental bases of science itself, which is that the universe is rationally ordered, comprehensible and explainable. This assumption is itself not falsifiable.

A perennially favourite allegory used to argue in favour of the importance of falsifiability is Russel's teapot. If you don't know it, it goes something like this: "there is between the Earth and Mars a teapot revolving in an elliptical orbit too small to be seen by the most powerful telescopes," and the point is that this can never be proven wrong (or right), and that reason is why we can discard it as meaningless or not scientific. However, it can be noted just as validly that the claim can be modified to just "There is a teapot," and remains equally unfalsifiable. (The point here being that's true for whichever noun you care to substitute for "a teapot".) It may be objected that the second claim can be proven true (or that it's "known" to be true) but we have already shown that science is not proof and the way to investigate that empirical claim would be with science.

So maybe science is falsifiable claims plus knowledge, but that raises a big question of how we know what we know, and how we know that we know it. It clearly has to be prior to science somehow (otherwise we're saying science is falsifiable claims plus science) and this brings us back to the problems of axioms discussed in the section on proof, specifically of how the prior ideas are justified. For example, the foundational idea that the universe adheres to any rules at all is not falsifiable and we also can never really claim to know it's true either. In fact, belief in it has to be based in either faith or induction and if we follow this definition we have to either conclude the foundations of science are unscientific, or consneed that knowledge gained by faith or induction is just as scientific as the rest of scientific knowledge. Either is unacceptable.

False Claims

A worthy digression is the weaker problem of falsifiable and false or likely false claims. Methods like homeopathy or claims like "Somewhere, out there in deep space at some set of specific co-ordinates in range of our telescopes, there lies an otherwise identical replica of my guitar in front of me now," are manifestly falsifiable — and also falsified, false, to the best of our abilities at least. (At least, I assume for any co-ordinates I could pick and check, that second one is false.) But actual truth or falsity don't determine science; Newton's laws have already been discussed, and along with examples such as Bohr's model of the atom or Aristotle's cosmology, form the body of past scientific theories since discredited - and yet current knowledge (from Newton, Bohr or Aristotle's perspectives, future knowledge) does not retroactively 'unscience' those theories in their own time. Today, yes; yesterday and yester-century, no.

It is prudent to assume a significant portion of current 'knowledge' will turn out the same way and forms a body of current, false, scientific theories and claims. Clearly, even if it's not a death blow to the idea, there is a difference between our interstellar chordophone and Newton's laws which science as the conjoining of knowledge and falsifiable claims does not appreciate.


Enough with shooting down every other idea; that isn't worth much if I can't put forward a science of my own. It's plain to anyone that while the previous sections may deal with incorrect assumptions of science, they nevertheless all deal with something close to the truth.

Science may not be proof, but we want scientific knowledge to at least be our best attempt at the truth; it cannot be based in induction, and yet experiment is the foundation of scientific research; it is not defined by the opinions of experts, but in practice scientists really do know the most about it; and while its limits are not circumscribed by falsifiability, it's obvious that science often deals with testing its own claims.

A common thread in all definitions of science is the centrality of experimentation — if a theory conforms to experiments better than all others, it is retained; if not, rejected. That is to say, a scientific theory above all else desires to predict the outcome of the experiment, and the better it predicts the result, the greater the scientific value placed on it, to the expense of others. It would make sense, then, that a scientific fact (so far as there is such a thing) is what the most predictive theory predicts; a best guess, so to speak.

This allows for all the quirks noted in previous sections. For example, Newton or Aristotle's theories of the cosmos were scientific in their time since they lacked any better competition just as our theories will likely be superseded by even more predictively valid ones in the future. Induction may not be valid proof but if assuming all electrons have an identical negative charge is more predictive than assuming any electron can have a positive or negative charge of arbitrary magnitude and only happen to hold that negative charge through sheer chance, then it's scientific to assume the former.

In fact, if assuming the next swan I see blocking the canal towpath will be white is always a good idea, then scientifically, until I find out about Australia, swans are as good as always white. Those fundamental ideals of science, that the universe is rationally explainable and not arbitrary, are scientific (so far) since they continue to be able to predict phenomena, even though ultimately they are taken as articles of faith.

So science isn't actually about truth — not directly. It's only about truth to the extent that truth is the most predictive thing; obviously if you had perfect knowledge (i.e. your knowledge and the truth were identical) you could make perfect predictions every time. It's important, though, to recognise science's path to truth is only indirect and proceeds imperfectly through jumps and starts as new theories upset the old.

Justified Prediction

It's tempting to add a caveat to that definition that the predictive theory must be justified, and that in the same way that knowledge must be justified true belief, science must be justified predictive theory. However, that runs into a pitfall (similar to one that theory of knowledge would) of how we know a justification is valid and what makes one justification better than another. Most theories will be justified with repeated observation in line with the scientific method, but we've already decided induction is not a firm foundation for knowledge, and it's not hard to see that the next line of justification is that the repetition signals the theory is predictive - which is completely circular since we want to justify the prediction in the first place.

It also presents another problem as the base of science is the assumption of cosmic order, which cannot be justified by anything other than its continuing predictive validity in areas it works, and as an article of faith in the areas where it (currently) fails that it will prove true in time, and that the fault is with us for having poor theories.


a: The principle of spatial translation invariance states that the laws of physics are the same everywhere in space, that is, if I perform an experiment in some lab L, move L by some arbitrary amount in space and then perform the same experiment, I will obtain the same result. Likewise for time translation invariance, if I perform an otherwise identical experiment today and tomorrow I ought to obtain the same result.