I was watching a YouTube video of a guy named Tom Jump who summarizes his epistemology as “If a hypothesis can make new, testable predictions then that counts as evidence. Whichever hypothesis one has in mind before the prediction occurs is what the evidence should count towards. All other theories can be eliminated or should not be taken seriously.”

Of course, this may beg the question of how many correct predictions a theory would have to make that should lead you to accept a hypothesis. It would also beg the question of whether you should have partial belief in a hypothesis even if the number or strength of those predictions don’t meet that standard. However, the simplicity of this epistemology intrigued me, and I was wondering whether predictability is seemingly really the only thing that matters not just in science but in terms of justification of belief in any theory.

For example, if the Old Testament made predictions of the double helix structure of the DNA, or evolution in sufficient detail, I would bet that almost everyone in the world would be a theist. Even though the Duhem-Quine thesis allows many theories to empirically confirm this, the other theories would confirm the prediction after the fact. Since the hypothesis in the Old Testament would involve the notion of a particular kind of God before DNA was discovered, and would be the first one, it would gain evidentiary support. Any other hypothesis that may also explain it (such as a devil tricking us) would not gain any support.

This is interesting because it seems to provide a neat way of getting rid of the problem of underdetermination of evidence: One can simply ignore all the theories that come after the original hypothesis that predicts a phenomenon, no matter how ridiculous the original hypothesis initially seems. It also seems to provide a neat answer as to why, for example, one can never prefer any interpretation of quantum mechanics over another. None make any more novel, testable predictions compared to anything else. Hence, until this happens, it is forever fruitless to give even an iota of preference to something like the “Copenhagen” interpretation over the Many Worlds interpretation (unless the Copenhagen interpretation was what was fully in mind when making the original predictions of QM which I’m not sure is the case).

Lastly, I also wonder whether this is really at the basis of our intuitions behind believing in any theory when it comes to matters about the truth of the world. In the Old Testament example above, I would wager that even though philosophers could come up with an infinite number of theories that would explain the DNA prediction, almost all people would still feel psychologically compelled to believe in Yahweh. Is it true, then, that creating novel testable predictions and nothing else is really at the heart of what justifies belief in a theory?


4 Answers 4


Actually, having testable predictions may not be the main reason we come up with scientific theories.

String theory, for example, was being actively developed over decades, even though it remains a long way from making a single testable prediction.1 The heliocentric cosmology, as developed by Nicolaus Copernicus, made no testable predictions. In fact, Galileo Galilei famously defended Copernicus' theory, at the risk to his own life, against the available evidence.2 Likewise, Isaac Newton developed (and was credited for) the math framework of the universal gravity long before we could verify the very existence of the latter.

Why do we develop such theories? Why do we feel that it is imperative for us to understand, to come up with a model of how our reality (the Universe, the world) works under the hood -- even in the absence of a way to test that understanding?

I believe it's all about the process itself: We piece our understanding together like a jigsaw puzzle. It's an exciting and highly satisfying detective work. And, aside from the bragging rights, there are actually two good reasons why one wants to complete this puzzle and share it with others sooner than later:

  1. It's likely true by the virtue of all its pieces fitting together. Imagine a 1000 piece jigsaw puzzle, which, after you have completed it, reveals a picture of cat. What are the chances that you have pieced it wrong, and that it actually encodes a different picture, say that of a sofa?3
  2. The part that you have completed could well be the part that someone else is missing. And maybe with both parts completed, together we would find a way to test them -- or, indeed, make a use of them to everyone's benefit.

And, of course, finding a practical way to test/use a theory is also important. But it's not a prerequisite to developing -- and publishing! -- that theory in the first place.

UPDATE: I didn't explicitly mentioned this about the heliocentric cosmology, but it was already widely accepted (even by the Catholic Church) by the time a French physicist Léon Foucault found its first testable prediction -- the Foucault Pendulum -- around 1851.

1 Testable through any available means, that is.

2 For example, due to the primitive optics at the time, stars would appear as disks rather than point-like sources, suggesting that they were not much farther than the planets. Earth orbiting the Sun would create a parallax -- an apparent motion of some stars against the backdrop of others. Yet no star parallax could be observed. Also, Earth under our feet would have to spin around its axis so fast, the objects at its equator would be traveling faster than a bullet -- and yet we don't feel it at all! At the time, many felt that this just doesn't make sense.

3 What could happen, however, is that after completing the cat you keep going to complete picture of a cat resting on a sofa -- but even then the cat part would still be there. That's the story of classical mechanics vs relativity -- the former is not wrong, it's just incomplete.

  • Great answer. The question was moreso about what causes acceptance of theories, not what causes us to form them. I do think understanding is crucial though Commented Feb 5 at 13:00
  • 1
    @Baby_philosopher -- thank you! I think we can all agree that acceptance is a personal choice. What was acceptable to Galileo, wasn't to the Church. What was acceptable to Niels Bohr, wasn't to Albert Einstein. As for how an individual makes this choice -- it depends, among other things, on whom they trust. In particular, it depends on how much they trust their own capacity for understanding (that is, their capacity to piece together their copy of the puzzle, to complete their own map of their reality). Commented Feb 5 at 19:38

This epistemological theory is simply random, it doesn't contribute anything to our understanding of the relation of evidence to theory, which is already captured quite well, ex. in Bayesian models of rationality and elsewhere. It is easy to recognize it as irrelevant: One can say that one had anything in mind before making the observation. It immediately becomes obvious that circumstances (in the broader, semi-Kantian sense of the term where conceptual capacities etc. are considered circumstances or conditions of possibility of experience) that make an observation count as an observation have nothing to do with such super-contingent matters which cannot be easily determined.

Also, whenever we make a false prediction, according to this theory we have in our mind our previous theory, obviously, because the epistemology prescribes us to hold onto it to be rational. Did people who recognized the inapplicability of Newtonian mechanics to the orbit of Mercury have Einstein in mind? I don't think so. This would make any progress in science unintelligible or unrational (and that's exactly what a good epistemology should account for).

No, this doesn't solve the Quinean underdetermination (which isn't a "problem" anyway - and isn't equivalent to the Duhem-Quine thesis either), which says that there is an infinity of perfectly suitable theories which fit all the evidence given no external criteria. This is true regardless of whether we have such a criterion or not.

What "solves" the underdetermination is Craig's theorem or Craig's trick (named after William Craig), which says that any recursively enumerable set of formulas of a first order language is recursively axiomatizable or, in relation to philosophy of science, that an alternative theory which encompasses only the observational conclusions of an empirically significant theory is recursively axiomatizable, which means that it is intelligible to a human. So, we could, without sacrificing intelligibility, get rid of all non-observational terms (and, so: sentences) of our theory and thus produce a minimal theory which has the same empirical conclusions as the theory. Which doesn't mean there is one unique set of axioms, of course.

I know this doesn't really answer your question directly, but I hope it was helpful nevertheless.


Keyword search: falsification criticism.

When I am ten, eleven, and twelve years old I have an amazing teacher, Ms. Wexler, as my fifth and sixth grade teacher. During reading comprehension lessons she instructs us to decode the meaning of words and sentences in context. A hypothesis or theory is only valid or invalid in some context.

When a reporter asked Einstein about the experimental proof of his theory that the mass of the Sun is sufficient to bend a ray of light, he said: One experiment would be sufficient to prove the theory wrong; and an infinite number of experiments could not prove the theory right; but the theory, as we have it, is quite correct!

Humans cannot make infinite observations in every imaginable context. I don't know if Karl Popper developed his falsification criteria before or after this quote of Einstein. But the point is that a theory is valid or not valid in some context of human perception and concepts. If it does not seem valid in any context it should be thrown out in favor of a more coherent idea.

Historically Newton's laws of motion could explain or fit the observed motions in the solar system with sufficient accuracy to be held valid in that context: except for the observed motions of Mercury. The motions of some bodies can be explained by the existence of other bodies that have not yet been confirmed to exist by actual observation. This is not what happened with Mercury. General Relativity gives sufficient accuracy to explain the motion of Mercury and it reduces to Newtonian physics when the context does not require the extra accuracy available using GR.

Theories of sociology, psychology, history, politics, etc. do not seem to be based on any measurable falsifiable observations. Statistical analysis is not sufficient to reject a hypothesis only to make it more likely or unlikely to be valid in some context based on statistical models. But the statistical model incorporates assumptions about the context as well as assumptions to make the likelihood meaningful.


Science generally assumes uniformatarianism - "... the assumption that the same natural laws and processes that operate in our present-day scientific observations have always operated in the universe in the past and apply everywhere in the universe." This is an "unprovable postulate"- by definition it cannot make any novel predictions as it is a statement of a status quo. Many sciences, such as geology, cannot make much progress without assuming it is true.

The theory of evolution by natural selection was accepted by the scientific community without unique/novel testable predictions - it was an explanation for what we already see (including the fossil record).

  • I thought that even at the time of writing Darwin predicted that we'd find intermediate forms on the path of human evolution and other lineages.
    – Dave
    Commented Feb 6 at 19:04
  • @Dave I don't think that was specifically humans, but it would be equally true of Lamarkian evolution, which predated Darwinism (i.e. it isn't novel). I don't think a prediction needs to be explicitly stated in science, a prediction that is a corollary of the theory is fine (at least from my reading of Popper) - the point is that the theory excludes some kinds of observations. Commented Feb 6 at 19:55
  • 1
    Maybe you want to update the last sentence to indicate that it did not make any unique testable predictions, if that is the point you are trying to make.
    – Dave
    Commented Feb 6 at 20:02
  • @Dave done, however I don't think the prediction was relevant to it's (eventual/reluctant) acceptance by the scientific community at the time, but that it was a very good explanation. Falsificationism is a more modern idea (that is disproportionately important in modern views of science, it isn't a full explanation of how ideas are accepted in modern science). Commented Feb 7 at 7:41

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .