Exploration of what properly constrains the production of knowledge is a very interesting topic and ethics forms but a part our considerations of what limits the creation of knowledge. Those constrains are however far broader than what is dictated by parochial concerns about what *should* be done in terms of generating knowledge. Because the growth of knowledge is inherently unpredictable, an argument looms that perhaps the only ethical principle one requires here is: do not apply ethical prohibitions upon the creation of knowledge. Of course, practically speaking, we should not seek to discover what is the most hurtful thing we can do to make people suffer? That would be abhorent. Or what is the most dangerous risk we can take? We can play games like this and suggest that therefore we need tight restrictions on what problems people should try to solve. Such concerns are not genuine limits upon the growth of knowledge but rather silly moral-thought experiments about how values seem to conflict (on the one hand the value of knowledge production and on the other valuing personal autonomy, for example) and they are always resolvable with a little bit of critical enquiry.
So ethics, typically, is not - or should never be - the biggest constraint upon the growth of knowledge. The growth of knowledge is motivated by problems that arise. That is what the growth of knowledge is: the search for solutions to some problem situation we find ourselves in, personally or as a community or civilization.
But there are other constraints upon knowledge. From logic for example: we cannot hope to discover simultaneously that eggs are simultaneously good to eat and also deadly poison (modulo logic games like: some people are lethally allergic to eggs, or that eating 100 of them might kill a person).
Knowledge production is of course limited due to physical law, there are limitations due to time, space and energy, there are perhaps limits yet to be explored (like the so called “no go” theorems found in pure mathematics and physics - but perhaps there are more we’ve yet no notion of). David Deutsch has explained the great dichotomy when it comes to the limits of knowledge: whatever is not prohibited by physical law is possible. So the only thing preventing us from accomplishing something we want to, and which we've decided is good to do, is *knowing* how. That's an amazing thing. Resources are almost always plentiful - the universe is vast. So taking a cosmic perspective on these things, it is not matter and energy and time that is scarce (the universe provides these in abundance, as it happens) but rather it is knowledge that is always scarce. (Of course, see his books for this - or his Ted Talk).
But also, now, and in the other direction - it is not only constraints upon knowledge but also it is the availability of knowledge - which is the limiting reagent in both the universe and our lives. Knowledge itself provides the constraint that prevents us personally, as families, communities and whole civilisations from accomplishing what we want. When we lack *that* resource - knowledge - everything else (importantly progress) stagnates. Most especially, civilisations do, and so do our own personal lives.
This idea of "constraints" as some kind of theme through which to view knowledge can be a useful one. Ethics, on this view, is but one example of the constraints on knowledge and also that there are many ways the production of knowledge is constrained...and also many constraints resulting from our lack of knowledge and lack of progress in our creation of knowledge. “Constraints” might seem to be a gloomy lens through which to view a thing, but on analysis this is an uplifting lesson to learn. Creating knowledge - learning more - is typically, in our world as it now is - the only thing (or at worst the main thing) limiting each of us personally and as a civilisation from accomplishing our goals. Your choice to know more really is the way to move forward.
*Credit goes to Ric Sims (@sharpcomposer) for remarks inspiring parts of this piece.
The Search For Truth
The prevailing view of “knowledge” - handed down from Plato - is that knowledge is some kind of justified true belief. Modern incarnations, descended with mutations to fill the niche occupied by this desire for justified truth include Bayesianism (a more mathematically inclined twin of inductivism) - where the idea is that knowledge is justified as close to true by repeated confirming instances. Whether Bayesian or Inductivist, these kinds of justificationism applied to science hold that the more frequently one observes an hypothesis to work, the more confident one can be in expecting it is actually true, more true, or probably true compared to its rivals.
But Bayesianism, in claiming that some theory has some quantifiable (indeed calculable) and precise amount of truth we can discover, cannot explain how despite repeated “confirmations” increasing one’s confidence in the truth of a theory, nevertheless, it can still be shown utterly false by an observation that theory cannot accommodate. Indeed it cannot explain how it is that when confidence in truth is at its highest, this is when theories are typically shown false. In other words, on Bayesianism, when we have every reason to expect the theory to be true, it is shown false. So for example, every single observation that occurred prior to around 1919 was a “confirming instance” that would have granted “Bayesian credibility” to Newton’s theory of gravity. (If this date is in dispute, we need only move it back to around 1859 where Newton’s theory had never been known to produce any anomalous predictions (it was in this year that Urbain Le Verrier in “Celestial Mechanics” published data dating from around 1697 to 1842 which, when investigated carefully, appeared to reveal some anomalies with Mercury’s orbit. In principle these could reasonably, at that time, have been interpreted as consistent with Newton’s theory on the assumption the orbit was being perturbed by some other massive body (this is not unprecedented given the method of discovery of Neptune relied on something quite similar)). Whatever the case, absent any other theory, the Bayesian method of increasing confidence that a theory is true, given repeated instances that are consistent with a theory, meant that Newton’s theory of gravity was at its highest confidence right before it was shown false. At which point all of those observations that it was correct now “flowed” in some sense to its replacement: Einstein’s General Theory of Relativity. Or if they did not “flow” then the count started again and Einstein’s General Theory - being without rivals - just continues to grow and grow in truth to this day. And with each passing day we should be more confident, not less, that it is true. But nothing in Bayesianism - no matter how many confirmations there are - can rule out the possibility that Einstein’s General Theory will be ruled out by a process similar to that which Newton’s went through. Namely some observation inconsistent with Einstein’s General Theory but consistent with some other theory that does everything Einstein’s does but also accurately predicting where Einstein’s cannot work. Indeed we should expect it to be shown false because we should always expect some deeper theory to explain everything some currently accepted best theory does...and more. That is: we should admit theories are improvable and progress is always possible because knowledge continues to grow. In particular we should expect a theory in physics to be found that is deeper than both quantum theory and general relativity - one single theory that can explain why both work and which also do something new that neither is able to: perhaps explain dark matter and dark energy or something like that. Something at a deeper level. That is what we should expect. We should expect falsity to be shown and so we should expect that General Relativity is, now, strictly, false. We just don't know how it is and cannot show it is yet. One day, we will because we will have both a replacement for it, and a test to distinguish the replacement from General Relativity by comparing it against reality in some way (we call such comparisons "crucial tests" or "crucial experiments".)
To remain with Bayesianism for a moment, it is also important to note that Bayesianism alone cannot explain how an ad-hoc modification to a theory is not also “verified” to the same degree. As explained in “The Fabric of Reality” by David Deutsch the idea that the currently accepted theory of gravity is justified as true or probably true because of all the observations that people have ever made consistent with it applies also to the theory that the prevailing theory is justified as true or probably true except in cases where it is defied on those occasions when objects levitate for reasons not accounted for by the theory of gravity. The “our best theory of gravity is true except when things levitate” is justified by precisely all of those observations that justify the current accepted theory of gravity.
So it cannot be the case that theories are justified by repeated observations - no matter how many there are. If they were, the ad-hoc modification that “things sometimes also levitate” would also be justified - even if we have never (yet!) witnessed such levitation that would be inconsistent with the first theory (that the best theory of gravity always applies everywhere).
This is an argument against induction and against Bayesianism. Repeated observations are not needed. That is not how knowledge is produced. Instead theories are guessed (conjectured) and then attempts are made to refute these theories. This is the rare best case scenario: there are multiple competing theories. All these theories then get tested against reality by some means. The means - the methods of criticism - along with the subject matter itself - are what define a “discipline” or “subject area” or “domain of inquiry” or any other such synonym for fields like “Science” as compared to “Mathematics” and “Philosophy” and “History” and “Morality” and so on.
So let us recap all of this in light of the broad brush strokes that the majority of people interested in this topic of epistemology - no matter where they are on the spectrum between Plato’s JTB and Bayesianism sit.
Knowledge, they sometimes argue is some kind of belief (not all Bayesians do this: some believe in knowledge that need not be about personal thoughts). But belief cannot be a property needed for knowledge as Karl Popper observed and David Deutsch has clarified in many places. Knowledge is not only something that is in minds. It is also in objects. A telescope contains the knowledge of how to focus light. A jet engine contains the knowledge of how to convert chemical energy into heat and thrust and motion. The DNA molecule contains knowledge of how to construct an organism. A book contains knowledge, as does a computer. But none of these dumb, unthinking objects have beliefs.
So knowledge is not about belief. Must it nevertheless be justified true? Justified true means “shown to be true” - but we have just seen that there is no method whereby a theory can ever be shown to be finally, once and for all, true. There is always some way it might be shown false (and we cannot rule this out). This is true in science, but even in mathematics and is basically the philosophy of "fallibilism" - the claim that error is impossible to avoid. Mathematicians make mistakes and (this is poorly understood but absolutely crucial to appreciate) proofs in mathematics are computations. Proofs are done by something. They are done by a mathematician (or a computer) using some physical object (their brain, or pen and paper or a calculator) and physical objects obey the laws of physics. And if the laws of physics say that necessarily physical processes are error prone (cannot 100% be shown to produce the same outcome every time (this is a consequence of the laws of quantum theory - our deepest physical theory)) then methods of proof will likewise not be 100% in all cases absolutely perfect. More than that - for reasons stated above about Bayesianism - we cannot even put a “close to 100%” number on it or any probability at all. My favourite example here remains Euclid’s demonstration of the obvious - clear to everyone - fact that through any two points a unique straight line can be drawn. We know this now to be false because there exist such things as curved (“non-Euclidean”) geometries and in these cases many straight lines can be drawn through any two points. For more on that, see here.
Knowledge is likewise never justified because if it could be the justifications would have to be justified. And if they could not be then our original claim would not be justified as true. But if the justifications for the justifications were true on this view, then this would only be because they were justified and so on, leading to an infinite regress. So “justification” cannot work as some kind of deep truth about how knowledge works because it rests on either an infinite regress of needing to justify justifications or stopping at some point where the justifications are unjustified meaning that “justificationism” is no kind of deep and universal truth about knowledge.
And finally “true”. When people here use “true” they seem to mean “certain”. And we cannot be certain because we can never be without doubt. And besides, certainty is just a feeling - one feels certain or not. And objective knowledge cannot be about one’s subjective feelings.
So there we have it for the moment: knowledge is not justified and it is not true and it is not about belief. Everything about Plato’s definition is wrong. Instead what is the case is that knowledge is about guessing theories (that solve some problem we have) and then criticising those theories. If we’re fortunate (because we’ve been sufficiently creative and critical and perhaps have cooperated with other similarly creatively critical people) - we manage to have many such theories. And then the critical process of experimenting (in science) or disproving (in mathematics) or trying just to argue (in all areas) and reveal weaknesses and flaws and contradictions we whittle away all the theories that fail to meet our criticisms and - again if we are fortunate - we’re left with just one theory standing. If we are not left with only one this, in science, is where we can do a crucial experiment. The experiment where the outcome is predicted to be one way given one theory but another way given another theory and that allows us to decide which is false. Whatever the case, in whatever the domain, usually we’re left with identically one theory that does what we want it to: solve our problem. And we call that The Explanation.
So we have jettisoned “justified” and “belief” in their entirety from this conception of knowledge. But what about “truth”. Is knowledge nonetheless a quest for “truth” as Popper says? Above I seemed unable to avoid the word, or its negation more than once. Of course we have seen the quest for knowledge cannot be a quest for certainty (100% infallible truth) but can it be a quest for something lesser? Well for the same reason that it cannot be a quest for 100% certain truth, it cannot be a quest for 99.99% truth or 99% truth or 50% truth.
So is truth a chimera?
Let us return to mathematics briefly. Surely it is about proving things true? What things? Well in mathematics what we assume we have are propositions (claims that are identically true, false or undecidable) and we use rules of inference to reach conclusions. But many pure mathematicians understand that because one needs to start somewhere (with axioms) that themselves must remain unproven assertions about the world, mathematics is actually not about proving things true. Rather it is just a domain of showing what necessarily follows from the axioms. Now if you assume the axioms are true then you can assume what is proved from them is true. But it is all just an assumption. If the axioms are false, well so much for your conclusion. Now because we have no method for showing that our axioms are actually true or false or undecidable - but rather that they are just assumptions, we may call what follows from them, on the assumption they are true and in the knowledge that moving from one mathematical claim (like an assumption/premise) to the next mathematical claim by following some rule of inference, we are not moving from proposition to proposition (actually, demonstrably true "meaningful sentences") we may more accurately say we are moving from statement to statement (approximations to such propositions) . So mathematics is about showing claims (that although we cannot know are true) do proceed logically (necessarily) one from another.
This works also for any domain of knowledge outside of mathematics and follows from what is called in the business “Tarski’s theory of truth” (named for Alfred Tarski). This is actually the person Popper refers to in “Objective Knowledge” (p 44 onwards) where he makes some “Remarks on Truth”. He makes the distinction there, following Tarski, that truth is “correspondence with the facts” and so it is sometimes also called the “Correspondence” theory of truth (this is the commonsense view, Popper says. I would add that this is to distinguish it from competing claims like: truth is about “Consensus” - that is, that a thing can be deemed true when some group of people agree that it is (a rather relativist notion if ever there was one. Each group, by this measure, when they disagree, has merely agreed upon contradictory “truths”) and there is also something known as the “Coherence theory of truth” where a thing is true if it coheres (agrees) with some other known true propositions. Of course how those propositions are known true is because they agree with each other and with some other “true” claims and so on. But at no point need anything need correspond with reality.
Popper begins this section on truth with the claim that “Our main concern in philosophy and science should be the search for truth…We should seek to see or discover the most urgent problems, and we should try to solve them by proposing true theories…or at any rate by proposing theories which come a little nearer to the truth than those of our predecessors.”
Is he wrong about some of that? Namely the first sentence? Should that - the search for truth - be our main concern? It would seem our main concern is solving problems. But does Popper suggest there that solving problems is to be identified with the search for truth? We cannot ask him, so I propose that this is indeed what we are doing in solving problems. We are searching for truth by eliminating error to bring us a little closer to truth. By uncovering tiny parts of it and eliminating falsehoods.
If we consider that statements are approximations to propositions (the latter what we cannot utter because those are actual truths or actual falsehoods) then the statement - being an approximation - is an approximation to truth or an approximation to falsehood. And in general terms, to correct errors is to make progress - to improve. But improvement or progress occurs in some direction. When we solve a problem it is that things get actually better. There is a direction. The direction is in bringing the approximation closer in line with reality. That is to say the statement comes to reflect that reality with increased fidelity. But this increased fidelity - this better way of capturing reality with the statement or the theory - this is an objective improvement. How is it an objective improvement? Well it solves the problem that a previous theory could not. That previous statements were unable to explain. The previous theory is shown wanting. In what way? Well the successfully criticised theory, the one refuted, cannot be the truth because it has been refuted - shown false by observation (or other criticism). Cannot be the final truth? No. Of course, as always, we may be mistaken. But having to make this caveat each time one uses the word "true" or "truth" can be cumbersome and violates Popper's injunction to "speak clearly...and avoid...complications." And regard brevity as important (p 44 "Objective Knowledge")
Theories solve problems. That is their purpose. But how can you know your problem is solved? Well - the solution has worked. That is to say that what was a problem (the position of the planet was there at point Y but you predicted point X because of theory “A”) but now you have solved that problem with your replacement theory “B” so when you do the calculation, "B" gives you the answer Y, and the old theory gave you a calculation leading to X. So the solution worked. The new theory worked. This is what “worked” means. It means it corresponded to something in the world. You compared it to something in reality. Reality matters: it is the adjudicator between your theories. Now of course you might have made a mistake. But modulo that, what do we say about theory A? It has been refuted. What does that mean? It means it cannot account for the observation that your planet was predicted to be at X but was not.
We cannot jettison truth. Knowledge has something to do with truth. But what? Well knowledge creation is about solving problems and that involves correcting errors. And correction of errors brings us closer to reality such that our statements about it are approximations to the actual truth. Now what it could mean to “hit on” the actual truth (some call this the “ontological truth”) is difficult to say. Could it be possible that “triangles have 3 sides” is in some sense the actual ontological truth? No. It can always be the case that this could be improved in some way. Being unable to imagine a way is no refutation of the idea that people improve their ideas. We cannot rule out the possibility that some future civilisation will agree (because, I don’t know, (let's be fantasical for a moment) they have uploaded themselves into some some holographic higher dimensional space) where triangles, it turns out, are rough approximations to figures that, when viewed from our meagre 4-dimensional spacetime, only appear to have 3 sides and in fact, viewed from a broader and deeper perspective available only to more enlightened higher-dimensional beings, actually have more sides. This might seem bizarre but I’d say it’s no more bizarre than, having mathematically proved from the “self evidently true axioms” that triangles have an internal angle sum of 180º degrees - you then learn about geometries where this is “self evidently true” NOT to be the case. So claims in mathematics - shown true, are sometimes overturned. We cannot know that when we think we’ve got it correct, that we’re going to be moments later shown how we’ve been in error. That there’s a problem.
So is knowledge a search for truth at all? So long as we solve problems and correct our errors such that the new theory that solves the problem by correcting the errors better corresponds to reality as compared to all rival - isn’t this enough? Yes - but there is a succinct way to put this.
The new theory contains more truth. Or: the new theory is more true. The old theory is demonstrably false and we know it’s false. Do we know it is once-and-for-all certainly false? No. Do we know the new one is once and for all true? No.
Is it true at all? Yes.
Can we say one is more true that the other? Yes!
Can we say by how much more? No. It’s merely a binary distinction. But it’s convenient. One theory has more truth to it than the other.
Are we sure?
No. We never need to be.
Can we say a theory is “true”. Yes - so long as we understand “true” there is shorthand for “fallibly, provisionally true” or “pragmatically true” which we can take to mean: we act as if it is true. And why not? If the proverbial life-and-death situation is before us, we should not act any other way. The patient’s heart has stopped and the epistemologically savvy emergency doctor calls for the (external defibrillator) paddles STAT(!). Those assisting need not debate whether it’s true that the paddles will work. They act as if it’s a true claim “they work”.
“Is it true those paddles work?” someone asks our critical rational doctor later. “Yes” he says - and quite right too. To say “Well, I don’t know if it’s true they do. But I do know they work” is not only cumbersome, but it misses something important in fallible critical rationalism. And that is that the word “true” should come to be known to mean “provisionally true” - this is the default position. Someone who thinks “true” means “certainly true” is making the mistake. That’s the error. And it doesn't matter if the majority are making the error and the minority understand how epistemology actually works. After all, most people think "knowledge" means "justified true belief" but we can still use the word "know" and "knowledge" without being overly concerned about each and every time providing the caveats. When we spot the errors, we point them out and in the case of "truth" if we want to highlight or criticise that error then affix the “certainly” adjective yourself to remind people that is not what the word "true" should be thought to mean in common day-to-day usage. Why should dogmatists be able to claim the word? Let's not cede that territory.
It is quite right to say that General Relativity has more truth (corresponds closer to reality and solves more problems and corrects errors with…) Newton’s theory of gravity which itself contains more truth than some “law” of gravity like F = 2GMm/r^4 but we cannot measure the quantity of truth. Truth is not a quantity that can be measured but it is a quality that a theory possesses compared to some other. There are many things we cannot measure and yet we can make reasonable and sensible claims as to difference in kind. For example, in biology, it is a routine matter to distinguish one species from another or even one breed from another. There may be edge cases, but in general the identification that a particular organism belongs to this species and not that species is done largely on the basis of appearance of kind or type. These days we can do this with greater precision using genetic analysis. In terms of epistemology we are not there yet but there is some symmetry here (and that is no coincidence).
As in any domain, in epistemology we want to solve a problem. The problem before us here and now is: how can we most effectively - that is to say clearly and efficiently and accurately - convey the epistemology that is critical rationalism? Should we jettison the idea that we are seeking truth? Or should we look at ways of preserving what is useful with that word and modifying what most understand the term to mean? In part this is what I have attempted above. We must be cautious we are not misunderstood as denying the possibility of truth - that may be viewed as a kind of relativism. Of course we can always be misunderstood. I return once more to Popper in “Remarks on Truth” as he says in many other places words to the same effect that and that I quoted only partially above “…aiming at simplicity and lucidity is a moral duty of all intellectuals: lack of clarity is a sin and pretentiousness is a crime. (Brevity is also important…but it is of lesser urgency, and it sometimes is incompatible with clarity).” Preserving not only the word truth, but also the idea that we are engaged in a search for it, helps with brevity and with clarity. Rather than avoiding the claim that science and reason broadly is a search for truth, we can merely correct people when they think it is about the search for certain truth, or final truth or “complete” truth (or a “complete science” as Sam Harris is fond of saying). Rather, we just correct them to “provisional truth”. Provisional truth that solves our problems.
So is it true "We aren't seeking truth"? Well is "seeking truth" synonymous with "solving problems"? Might it not be parsimonious to use these interchangeably given the facility of both terms? "I'm looking for the truth!" exclaims the exasperated scientist trying to uncover if the wobbly motion of their planet is a sign of yet another, as yet, unobserved planet. Are they wrong to do so? Should it be "I'm trying to solve this problem!".
I don't think it matters.
Do theories need to be falsifiable to be science?
That theories need to be falsifiable is a necessary but not sufficient condition for science. For example, the claims: Eating 1.00000 kg of grass cures the common cold or The world will end at 2am UTC on 2/2/22 are falsifiable theories. But they are not scientific. Without a good explanation to accompany them, they are not science. They are just “falsifiable claims”. A scientific theory should be a good explanation that also happens to be testable/falsifiable. Popper figured out that falsifiability is an improvement on verifiability of the logical positivists. It is falsifiability that better separates science from non-science. This includes separating science from pseudo-science like astrology and homeopathy as well as things like morality and philosophy broadly. But it has never been the case that all falsifiable theories are scientific theories. For example: those two claims I started this paragraph with.
But is it nevertheless necessary that scientific theories need to be falsifiable? Well the scientific theory for some phenomenon - or any theory that purports to be the scientific theory for some phenomenon must be a good (hard to vary) explanation of that phenomenon. Part of this “hard to vary” quality is that the theory is falsifiable - testable by experiment. In principle. Now it need not be in practise. But that doesn’t change its testability in principle. So, for example: many people have observed that string theory is very, very difficult to test. Some have asserted that to observe the predictions of string theory would take a particle accelerator half the size of the galaxy. Now this is impractical. So does this mean the theory is unfalsifiable? No! In practise we cannot build such a particle accelerator. But in principle it could be done. So it’s still falsifiable in principle. And perhaps there exist "natural" particle accelerators such a quasars - observations of which might rule out string theory? We do not know.
So, it’s science. It makes predictions. We need not jettison falsifiability on the basis of that. What we might do is search for better ways to test it. If it’s a claim about the physical world, then the physical world must be the adjudicator of the truth about string theory. Can we rule it out? Can we refute it? Then it’s falsifiable. But notice there are two kinds of falsifiability: in principle and in practise. In principle is a black-and-white quality of a theory that is required for science. It is just the claim that some observation of physical reality could in principle rule out the theory. But if no such observation can - that is to say no such observation exists in any possible world - then the theory is not about the physical world. There is no comparison to be made between the actual physical world where the theory holds and an imaginary fictitious physical reality where the theory does not. Or vice versa.
Let us take an even more extreme case than string theory (which I argue is science - but for reasons I will come to is not necessarily “good” or “optimal” science) - and that is the theory that there exist other universes where the very laws of physics are themselves different. So universes outside our own, but where the laws are different. Now it was once thought that such universes are in principle unobservable and so therefore not testable and this makes them unfalsifiable and not science. After all: another universe? Outside our own? How can we access that? Well as it turns out - in principle - we could see such a universe. A universe where the laws are different will have different physical constants and as far back as 1999 physicists claimed to have observed a changing fine structure constant. This would be evidence of a region of space where the laws were different. Another universe (by some definitions). It turned out they were wrong (see that very same link above) - but it is this kind of observation that, in principle, could allow us to observe other universes beyond our own. (Or force us to change what we mean by “universe”).
But this “falsifiable in principle” (necessary as it is) as a criterion to demarcate science from metaphysics (for example) is also not sufficient to make something a good, hard to vary, explanation. Let us return to string theory. What we’re interested in is solving problems in physics and string theory is an attempt to unify quantum mechanics (a physics of discrete entities like particles and energy) with general relativity (a physics of continuous entities like space and time). As we have already seen, string theory could in principle be tested with a particle accelerator half the size of the galaxy. That's the worst case scenario - likely things are not that grim. But say they were. There is probably not enough matter for several lightyears to construct such an experiment. It’s impractical. So “in practice” we would have to say it’s not falsifiable. It’s “not falsifiable in practice”. But this is not a black-and-white all-or-nothing thing. In practice means something like “we lack the wealth to do so” - we cannot actually perform the physical transformation of the matter to do this. Actually we do not even know how to gather enough matter - using the technology we have presently - to build one. So knowledge is also a problem here.
The fact that string theory makes this assertion of itself (as being practically not testable right now), as things currently stand, makes it an “easy to vary” theory even though it’s testable in principle. This is because minor modifications of the theory making similar predictions cannot be distinguished by experiment. And many such varieties of string theory exist. So “untestable” in practice is a weakness. This does not make string theory unscientific - it just makes it a poor explanation. For now. Maybe someone will think up a better test. Maybe someone will make a prediction that operates at lower energies requiring a smaller particle accelerator. Or - and this is key - maybe someone will come up with a theory that makes all similar predictions string theory can but which itself can be tested by some routine means available here on Earth - making that new theory a good explanation and worthy replacement for both quantum theory and the general theory of relativity. Such a theory - testable in practice as well as principle - would be a very good explanation. And string theory would then not be.
What is an example of a good theory within science that is unfalsifiable in principle? I do not know of one. Why is it important to distinguish between science and other subjects or disciplines anyway? It is largely a matter of convenience but also important to distinguish efficiently and effectively between pseudoscience and scientistic arguments (so arguments that claim something like: science can tell us what we ought to do - that there can be a science of morality or a science of economics or politics). Knowledge is some kind of unified whole, it is true. But "falsifiability" is a useful necessary criterion for science. And it is useful to know that demanding, say, moral theories are testable would be a terrible error. This would mean requiring the conducting of experiments on people (say) in order to determine if an even more pure version of Communism than anything China or North Korea has ever tried would be a good idea because - science! No, we do not need to experiment. We begin instead with moral explanations about people and rule out the "a falsification is required here before we can properly reject this theory". Morality is not science and we should not require it to be. But science is a place where experiments - conducted on the physical world are necessary. I think it's necessary we preserve this distinction.
A note on evolution
Quite rightly I was altered to and corrected upon a misconception I had about a particular kind of exception to this strict requirement for falsification in science. For reasons we shall see this does not undermine the central idea that scientific theories must be falsifiable. Now in the case of some (large number!) of theories they are not practically testable because there are no viable alternatives. This means we need to split the meanings of "falsifiable" and "testable in practice". Because there are no viable alternatives to Neo-Darwin "Evolution by natural selection" it cannot be "tested" - because to be testable it needs to be tested against something. And there is nothing. As David Deutsch observes in The Beginning of Infinity: if we observed something inconsistent with the prevailing theory of evolution by natural selection, nothing could be said except the test we used to find the inconsistency was faulty. It is often said, following Haldane that "Rabbits in the pre-Cambrian" would refute evolution by natural selection. But they would not. They would be a problem - but they could be explained by being a rare complex organism that somehow got there earlier than anything else (unlikely) or that a mistake was made by our geologist or paleontologist or evidence of a prankster. Many things would need to be ruled out (and how?) before we ruled out evolution on the basis of rabbits in the pre-Cambrian. But this untestability does not mean unfalsifiable in principle. These are different things.
If an organism (or many organisms, many different species) we found to undergo only or mainly favorable mutations then this would be better explained by Lamarckism and would rule out Darwinism. But then there are all those organisms we already know of that would refute Lamarckism. But the point would be Darwinism would be refuted as a universal (applies to all cases, everywhere) explanation for the evolution of life. It would just be a special case - presumably of some deeper explanation that accounted for why both Lamarckism and Darwinism worked within their less-than-universal domains. So testability and falsifiability are not synonyms. While the latter is needed (and Darwinism is that) the former is about the practical ability of performing some test (experiment) and having somewhere else to "jump to". Some viable alternative theory to test our theory against.
Not everything in science, it should also be noted, is falsifiable. Some eminently scientific claims are unfalsifiable. In physics we say "Work is a form of energy". That's a scientific claim. It's also unfalsifiable because it's essentially a definition. One will never calculate the physical work done (by using a classical formula for work like Work = Force x Distance) and discover that it is not a form of energy. These are just words and terms and though scientific, untestable and unfalsifiable. So some things in science are unfalsifiable. But they are not explanatory theories as such but more like frameworks within which we do science. In chemistry the scientific claim that "The 6th element on the periodic table is carbon". Or "The element with 6 protons is carbon" is a scientific claim. But it too is unfalsifiable. No one can possibly ever, in any world, discover an atom containing only 6 protons and conclude it is not an atom of carbon. No one will find some element which, upon analysis is carbon but contains 7 protons in every nucleus (because that would be nitrogen). And no one will find an element that contains only 5.5 protons in the nucleus bumping carbon up one position on the periodic table. These things are ruled out by the definitions of words like "element" and "atom" and "proton" and "carbon". So unfalsifiable claims in science are common. But the explanatory theories in which these definitions are used and themselves explained make predictions that can turn out to be false. It would not falsify the definitions - but the theories. In particular all existential claims of the form "X exists" are unfalsifiable. So the claim "gravity exists" is not falsifiable. But the claim "Gravity is a force" is, and was falsified. Gravity still existed - it just turned out not to be a force but rather, as Einstein showed, was the manifestation of space being warped by energy and matter. "Gravity" is a word used to describe some phenomenon that exists. The concept of gravity cannot be "falsified" - only what it appears to be, or what it is claimed to be, can be. In an extreme case, the idea "Matter exists" cannot be falsified, though matter may not be the most fundamental thing, in the final analysis. Maybe it is true that there is something deeper - a Platonic realm of sorts from which the appearance of matter arises. But that would just be to explain that matter is an emergent feature. The appearance of it - which is to say its measurable qualities - would still be real emergent things.
So falsifiablility is a necessary quality for scientific theories to possess. But not all claims in science are falsifiable. And falsifiability is not the same as testability. In particular the theory of evolution is not obviously testable in practise. Though it is in principle falsifiable. What we call science in the final analysis is an open question. It is a domain of study focussed on discovering how the physical world works - the patterns in nature, their beauty and their dangers. In part so we can control our environment and use it to our advantage. We guess what's true and compare our guess against that physical reality in some way. So long as we are making progress and solving our problems, that is what matters. But if progress is slow, that can be when these debates can be extra useful to understand.
This post has been motivated by some inspiring Tweets by Lulie Tannet (@reasonisfun) which then resulted in a subsequent exchange of ideas with David Deutsch (@daviddeutschoxf) and others. As always, "The Beginning of Infinity" and "The Fabric of Reality" underpin much of what I say - but errors are my own and nothing I say should be seen as an endorsement by David Deutsch. You can (and should!) buy both books here: https://www.daviddeutsch.org.uk/books/the-beginning-of-infinity/