susu.exp wrote:I don´t know how one would go about constructing a measure of confidence based on previous observations. Let´s say we have two theories and they disagree on some predicted observations, but they are both empirically adequate, i.e. they haven´t been falsified so far. I´ve got the same confidence in both - after all they´ve passed the same muster - they agree with all known observations. We can then go and check the things they disagree one, which will lead to one of them being falsified. I.e. I´d argue that my confidence in a scientific theory does not increase with the number of supportive observations. And I can be fairly certain about two conflicting theories and simultaneously be absolutely certain that one of them is wrong. I´m with Popper on this one: There is no need for a logic of science to even mention induction.

What are the likely areas of disagreement between two competing theories, if they're not observations? Do you mean they're likely to be decided on a theoretical basis rather than from (future) predictions from experimental data?

Unknowing wrote:Thanks. I don't really understand c), is there a dumbed-down version of that!? Is this broadly speaking a constructive empiricist position — to talk about empirical adequacy vs truth?

c) is pretty much a formal logical statement of falsification. If a theory predicts something and that prediction is wrong, then so is the theory. There are formal logics in which this doesn´t hold, so for falsification to work, we need to state upfront that we use a logic that allows it.
And yes, basically it´s a constructive empiricist position. To make a claim for the truth of a theory requires more than we have at our disposal or in fact requires some stronger premises (basically the type of addition andrewk is remarking on). By not including it, we have to make a weaker claim (EA rather than truth), but in return gain greater universality. The great advantage of keeping the number of premises low is that it´s harder to construct systems that are inconsistent.

Unknowing wrote:What are the likely areas of disagreement between two competing theories, if they're not observations? Do you mean they're likely to be decided on a theoretical basis rather than from (future) predictions from experimental data?

Nope. They differ in predictions about observations. What they agree on are past observations, so neither is falsified yet. But since they disagree on some prediction, we can then go and make the type of observation where they differ in the prediction, falsifying at least one of them. I do not regard two theories that do not differ in their predictions as different theories at all, they are merely different expressions of the same theory.

susu.exp wrote:

andrewk wrote:On the first bit, if I understand you correctly, you are arguing that induction is not necessarily (and perhaps also should not be) a part of the practical methodology of science, rather than arguing that induction need not be assumed true in order to provide any epistemic justification for scientific theories (which is what I had assumed up to now you were arguing). Does that correctly describe your position?

I think that something like induction would be neccessary for an epistemic justification of scientific theories beyond the statement that they are empirically adequate. But I don´t think there is such a justification and empirical adequacy is the best we can do and for that we don´t need induction.

I haven't come across the term 'empirically adequate' before, so that's my new learning for the day . My favourite internet goto site Wikipedia, surprisingly, doesn't have an article on it. But Stanford Encyclopedia of Philosophy has one on 'constructive empiricism', which mentions empirical adequacy a lot. I didn't have time to read all of it, but the gist of it seems to be that a theory is labelled as empirically adequate if it explains a set of observed phenomena, and is not contradicted by any other observations. Is that what you intend by using the term?

The problem I see with setting empirical adequacy, so defined, as the sole goal of science, is that it only enables us to make statements about the past. For example, under that view of things, Newton's law of gravity (chosen in preference to Einstein's solely in order to avoid lots of subscripts and superscripts), when first published, could only say:


The date in this law could be moved later with each new experiment confirming the law, but it could never say anything about whether the law would apply at any time later than when the latest experiment had been done.

Only if we adopt a principle of induction/uniformity can we transform this into a statement that addresses whether the law will hold in future. And without that transformation, the law is of no use in designing rockets, bridges or buildings, or even in trying to catch a cricket ball. Hence, without an assumption of this principle, science is of academic interest, but no practical interest. In fact, science essentially becomes a branch of History.

So when I said induction is necessary to do science, I suppose that's not strictly true. Without induction we can still do science, and construct empirically adequate theories to explain what happened in the past. We just have no reason to believe that science can be used to build, plan or achieve anything. That's why I suggest that some form of principle of induction/uniformity is essential to having a rich yet consistent set of beliefs. That principle can only be adopted as an axiom, as any attempt to prove it ends in circularity. But, given that any system must have at least one axiom, a system with only one axiom seems to me to be quite elegant. Further, although the principle cannot be epistemically justified (aka 'proven'), it can be pragmatically justified, as shown by Reichenbach's very nice argument.

andrewk wrote:I haven't come across the term 'empirically adequate' before, so that's my new learning for the day . My favourite internet goto site Wikipedia, surprisingly, doesn't have an article on it. But Stanford Encyclopedia of Philosophy has one on 'constructive empiricism', which mentions empirical adequacy a lot. I didn't have time to read all of it, but the gist of it seems to be that a theory is labelled as empirically adequate if it explains a set of observed phenomena, and is not contradicted by any other observations. Is that what you intend by using the term?

For the most part. My position is even a bit weaker, because I distinguish between a theory - a set of predicted observations - and the expression of a theory - some system of statements that allows us to construct such a set. Theories themselves don´t contain explanations, you can take expressions as explanations, but you can express the same theory in several ways, each of which then would also be a different explanation. Because we can´t empirically distinguish between two expressions of a theory we can´t justify one explanation over another through science, we can only justify one theory over another.

andrewk wrote:The problem I see with setting empirical adequacy, so defined, as the sole goal of science, is that it only enables us to make statements about the past.

It´s the minimum goal of science. The point with the 3 statements was to give a minimum of assumptions needed for science to produce some type of knowledge. The type of knowledge you get from this minimum set is then constrained by what these assumptions allow. To get a less limited type of knowledge, you have to add more statements and then it´s not the minimal number of assumptions for science any more.

andrewk wrote:For example, under that view of things, Newton's law of gravity (chosen in preference to Einstein's solely in order to avoid lots of subscripts and superscripts), when first published, could only say:

"up to but not including the date of 5 July 1687, there was in operation a law under which there was an attractive force between any two objects of mass m and M, equal to GmM/(r*r) where the masses are given in kilograms, r is the distance between the two centres of mass in metres, and G is 6.674×10−11 N m^2 kg^−2."

The date in this law could be moved later with each new experiment confirming the law, but it could never say anything about whether the law would apply at any time later than when the latest experiment had been done.

If Newton had stated the theory this way, he would have given a useless theory. Because it only makes a finite number of predictions it is verifiable (but once verified useless). Newtons theory was empirically adequate on July 5th 1687. As stated above, it would have remained so for ever. But by stating it in a general way, containing infinitely many predictions allows us to check whether it stopped being adequate every day. 1887 Michelson and Morely observed that light didn´t follow the superposition principle and when they did so, Newtonian mechanics stopped being empirically adequate.

andrewk wrote:Only if we adopt a principle of induction/uniformity can we transform this into a statement that addresses whether the law will hold in future.

I disagree. We tenatively accept empirically adequate theories and we thus tenatively accept the predictions it makes about the future. We just don´t believe it´s true, instead we treat it as if it was true.

andrewk wrote:That's why I suggest that some form of principle of induction/uniformity is essential to having a rich yet consistent set of beliefs.

I would agree, but the question there is whether you can also generate a rich and consistent set of beliefs by rejecting uniformity. Since both are expansions that don´t follow from the 3 statements, they are both consistent. Both lead to different views of what scientific theories mean beyond empirical adequacy.

andrewk wrote:That principle can only be adopted as an axiom, as any attempt to prove it ends in circularity. But, given that any system must have at least one axiom, a system with only one axiom seems to me to be quite elegant.

Well, it would be the 4th axiom - I don´t think it dispenses with the need for multiple observers, their ability to communicate or the deductive framework needed to check whether a theory is falsified. Still a small set of axioms, but not the minimum one.

andrewk wrote:Further, although the principle cannot be epistemically justified (aka 'proven'), it can be pragmatically justified, as shown by Reichenbach's very nice argument.

I don´t think it provides a good reason to pragmatically justifying induction. If we tennatively accept EA theories which contain predictions about the future, i.e. if we treat them as if they were true, we have archieved the same thing without bringing induction in. We only need it, if we want to go further and say that we don´t just treat them as true, but believe that they are true.

CdesignProponentsist wrote:Theories are many times decided on theoretical basis when there is no experimental evidence that distinguishes one over the other. Usually the simpler more elegant of the two is favored. But that is not to say that the other theory is abandoned. There are champions of many unpopular theories that have even become mainstream again.

Thanks for that, are there examples of these?

susu.exp wrote:To make a claim for the truth of a theory requires more than we have at our disposal or in fact requires some stronger premises (basically the type of addition andrewk is remarking on). By not including it, we have to make a weaker claim (EA rather than truth), but in return gain greater universality. The great advantage of keeping the number of premises low is that it´s harder to construct systems that are inconsistent.

Unknowing wrote:What are the likely areas of disagreement between two competing theories, if they're not observations? Do you mean they're likely to be decided on a theoretical basis rather than from (future) predictions from experimental data?

Nope. They differ in predictions about observations. What they agree on are past observations, so neither is falsified yet. But since they disagree on some prediction, we can then go and make the type of observation where they differ in the prediction, falsifying at least one of them. I do not regard two theories that do not differ in their predictions as different theories at all, they are merely different expressions of the same theory.

susu, does CE take anything from pragmatism? What do you make of Quine's holism – in the context of this thread, would you say his critique is still relevant?

Thanks.

Unknowing wrote:susu, does CE take anything from pragmatism? What do you make of Quine's holism – in the context of this thread, would you say his critique is still relevant?

Thanks.

I would say that CE is at odds with pragmatism - the latter holds scientific theories as true, the former merely as empirically adequate. While the truth criterion of pragmatism is weakened and one could probably say that any theiry that CE deems EA is true by the standards of pragmatism, CE by introducing empirical adequacy allows stronger definitions of truth (which I´d hold are required in other fields).
Quines holism was mainly directed at the analytic/synthetic distinction raised by logical positivists. I would say that reducing theories to their empirical content allows a similar distinction to be made which doesn´t get touched by Quines objection. Or rather, Quines objection would affect particular generators of theories, but not the theories themselves.

I had a look at CE – not sure about the observable/unobservable distinction, but it is quite interesting. From my layman's perspective, it does seem sensible to allow for a separate epistemic and pragmatic commitment to the truth claims of theories, depending on the particulars of those observable entities/unobservable entities embedded within.

One of the benefits I can see is that what once may be counted as unobservable can become observable, since the epistemic community that determines such is itself subject to development and growth.

What would a constructive empiricist make of e.g. the theoretical models of psychology, which seem to span a huge range of observables and unobservables? It looks like endorsing CE can get you into some quite nebulous areas with regards to making claims about metaphysical entities? (Not that I would consider such a bad thing myself....)

susu.exp wrote:For the most part. My position is even a bit weaker, because I distinguish between a theory - a set of predicted observations - and the expression of a theory - some system of statements that allows us to construct such a set. Theories themselves don´t contain explanations, you can take expressions as explanations, but you can express the same theory in several ways, each of which then would also be a different explanation. Because we can´t empirically distinguish between two expressions of a theory we can´t justify one explanation over another through science, we can only justify one theory over another.

This in practice can be not so simple and clear cut, as you can have theory generators which seem to make the same predictions, but having no way to construct a formal proof that they do, if either you allow them to be formulated using axiomatic systems powerful enough, or if they do not rest on an axiomatic foundation at all.

I don't understand the need to having this distinction between generators and theories at all.
I can not think of any objection to having two theories which make the same predictions. They can be tested and falsified independently of one another.

What is your reasoning behind this?

mizvekov wrote:This in practice can be not so simple and clear cut, as you can have theory generators which seem to make the same predictions, but having no way to construct a formal proof that they do, if either you allow them to be formulated using axiomatic systems powerful enough, or if they do not rest on an axiomatic foundation at all.

That´s no practical issue at all - two generators are the same theory if we can not find a prediction they differ in. This basically shifts the burden of proof - rather than having to prove two generators are the same theory (which we can do in some cases nevertheless) you have to show that they are not to state that they are different theories.

mizvekov wrote:I don't understand the need to having this distinction between generators and theories at all.
I can not think of any objection to having two theories which make the same predictions. They can be tested and falsified independently of one another.

What is your reasoning behind this?

Well theories as defined above are somewhat minimal again - they contain predictions and nothing more. There are various extensions of this, either epistemic (providing explanations) or metaphysical (various forms of realism about theoretical objects). These extensions are extensions of generators, not of the theories themselves. It´s worth noting that the extensions of two different generators of the same theory can (and often are) contradictory. Either the sun excerts a gravitic force on earth over a distance, or it generates a field earth interacts with locally, or... Both of these describe Newtons theory of gravity and make the same predictions about the motion of earth, but the explanations differ and if you were to extend them metaphysically you would make different claims about the existence of fields. All of that is outside of science - it´s untestable. And that´s the main reason to make the distinction: Theories are testable, generators aren´t. If you state that two theories make the same predictions, then of course they are falsified by the same observations. And that means they can´t be tested independently of each other.

@Unknowing: As far as I understand the observable/ubobservable distinction in CE it is rather similar to the above. Unobservable in this sense is something that can´t change - in the above case the gravity field is unobservable and there´s nothing we could do in principle to make it observable - the same predictions about anything we can observe arise in the formulation using forces in which no fields appear.

susu.exp wrote:That´s no practical issue at all - two generators are the same theory if we can not find a prediction they differ in. This basically shifts the burden of proof - rather than having to prove two generators are the same theory (which we can do in some cases nevertheless) you have to show that they are not to state that they are different theories.
...
Theories are testable, generators aren´t. If you state that two theories make the same predictions, then of course they are falsified by the same observations. And that means they can´t be tested independently of each other.

I did not mean 'independently' of each other, sorry, I meant independent of any knowledge about their relationship.
You don't need to be able to prove theories make the same predictions to do science with them, although it does have a practical value.
But the reasoning about extensions is fair enough, although it does seem to be a new terminology being introduced. Besides yourself, anyone else advocates this?