Is the primary aim of science predicting outcomes of experiments or understanding the nature of our world?

Question

Scientist P observes a phenomenon X and develops a complicated ad hoc theory PR to describe X.

Then scientist E reads about theory PR and he realizes that theory PR can be replaced by a much simpler non ad hoc theory ER, which also describes phenomenon X.

Who is more in line with the aim of science, according to philosophers, P or E? One could argue that P is because P was the first to explain phenomenon X. But one could also argue that E is since his theory not only explains X but also does it in a simple non-ad-hoc way.

Really the essence of my question is "Is E a scientist?" or "Is E doing science?". I ask this question because I wonder whether science is about predicting outcomes of experiments or about understanding the nature of our world. If science is only about predicting outcomes of experiments, then E is not really doing science by proposing his theory ER, since it did not add anything to PR, which describes X perfectly. But if science is about understanding the nature of our world, then E is definitely doing science since his theory ER adds to our understanding of phenomenon X, even though theory ER predicts the same thing as PR. I understand that different positions are possible and would like to see arguments for both sides.

My question was inspired by the Relativity Priority Dispute. According to some historians/scientists, Poincare really deserves credit for relativity, not Einstein, and in my scenario above E could be Einstein and P could be Poincare. But I am not at all claiming that the scenario above is historically accurate.

Answer 1

There is a good explanation for why scientist P would come up with a more complicated theory than scientist E: experimental data on the bleeding edge of science is really noisy and often contains data points which shouldn't really be there. A good example of this is Hubble's original data. A study of what experimentalists do with data found that throwing away 50% of the data one has collected is common. Scientist P has to figure out what ought to count as data and then try to account for it with a model. In contrast, scientist E is exposed to much less complexity and has probably been shielded outright from a great amount of data which didn't seem to fit. It should not be surprising that E can reduce the complexity further.

It may be helpful to note that knowing how to group data together is a key scientific expertise; merely fitting the data is not necessarily the most difficult aspect. Kuhn showed that part of scientific revolutions is a change in what is even considered 'data'. We do not naturally slice reality up into the proper groups for doing excellent science. Indeed, a key advance was when humans learned to group things based on common cause instead of common appearance. But you need to have some sort of causal model in order to do so.

The question of "better" seems relevant only if we can expect scientists to regularly combine both E's and P's expertise. Otherwise both types of scientist are required. If that is the case, then which is "better" will probably be decided by simple demographics: if there are more P's than E's for whatever reason, E's will be seen as more valuable. I argue it would be rather arbitrary to make a declaration of "better" based on what is happenstance, for all we know.

Perhaps part of the problem here is a desire to connect a scientific discovery with exactly one person? That widespread practice may actually damage scientific discovery, as it incentivizes anti-collaborative behavior. But that too is merely a possibility; perhaps if we can pack enough theory and expertise into individuals, they do better science than groups.

Answer 2

I like @labreuer's answer, but I'd still like to add another point of view.

From the perspective of social epistemology, scientific research is a deeply collaborative, socially organized activity. So it can be misleading to attribute the development of theories to individual scientists. Instead, social epistemology directs our attention to the social processes and scientific institutions that produced the new theories.

Regarding relativity, note first that neither Poincaré nor Einstein were conducting experiments and gathering data. The Michelson-Morley experiments provided important anomalies (in the Kuhnian sense, roughly "phenomena that didn't fit Maxwellian ether theory"; see Collins and Pinch's The Golem for a discussion of the complex social construction of the Michelson-Morley experiments as anomalies). Second, neither Poincaré nor Einstein developed relativity ex nihilo. Numerous other theoretical physicists, mathematicians, and philosophers contributed ideas that were wrapped into relativity — Lorentz, Mach, Riemann, Leibniz, Kant — and the idea that measurements are relative to the observer goes back at least as far as Galileo. From a social epistemology perspective, relativity was developed in the scholarly interactions between these theoreticians, experimentalists, mathematicians, and philosophers; it did not spring fully-formed from the brain of one or more individual geniuses. We can more-or-less characterize the contributions of individuals to this overall achievement of developing relativity; but from this social epistemology perspective it's a mistake to say that one individual's contributions mark a strict division between relativity and its forerunners.

Answer 3

In you example, if both theories explained the experiment equally and predicted (or not) the outcome of other experiments equally then they would be considered equivalent. Both science, both equally valid. If one or the other had more explanatory power then that would be considered "better" but it doesn't follow that the other is considered less scientific. Just less powerful. Also, priority isn't really a concern wrt the science. For example, Newtonian mechanics is no less scientific just because it's shown to be inaccurate at extrema. Also, Lagrangian mechanics is known to be equivalent to Newtonian mechanics, and not considered lesser because it was developed later.

Now, that's the theoretical bit. In practice, these theories may be treated very differently. Trivially, we honour (arguably fetishise) priority so the first theory is likely to get more notice, be up for the Nobel etc etc.

More usefully, scientists favour utility. So if one or other of these theories makes it easier to make predictions then it will likely be used more. Feynman's approach to QED is a good example. There are two other, more or less contemporaneous approaches, all 3 won the Nobel, but Feynman's is by far the most used. Not because it's better scientifically (it was proven by Dyson not to be) but because it's much easier to obtain results.

Also, scientists have a predilection for Occam's razor. They will tend to favour a simpler explanation. This is a sensible heuristic (historically it's proved worthwhile) but it isn't scientific per se. For example, the search for a Theory of Everything has some historical validity (Einstein's main achievements came from reconciling phenomena) but is still just based on belief in its existence.

Finally, it's a false dichotomy to ask if science is there to explain experiments or the real world. Experiments are part of the real world. A theory that explains one tiny thing is still valid but less useful than a theory that predicts a huge range of phenomena.

Answer 4

Your answer is 'mu' -- either answer is equally bad.

Science has a composite 'primary aim'. It wishes to model the universe, but it insists on doing so in a logically traceable, rationally accountable way. However thoroughly you did the former, if it included none of the latter it would not be science. Likewise, just pushing around existing explanations to see what you can get out of them, without explaining anything more is merely technology and not science.

This combined critierion is captured well in Popper's notion of falsifiability. From Popper's point of view, if, in proposing a new explanation, you haven't risked bringing down this accountability on yourself when some test fails, you are not doing science.

If you can't stage a test, you haven't risked the test failing. So it would be silly to label the former alone as the primary aim science.

If you have only handled the known material, you are only repeating existing tests, and you know they won't fail. So it would be silly to label the latter alone as the primary aim of science.

In those terms, there is no good reason to think of either P or E as better. Increasing the coverage of the domain, even if it is with a very weak theory, has value. Refining the theory has value. (Pointless labeling is waste. Can't we all just get along?)

P, a discoverer, just stating his data in scientific terms implies it can be repeated under ideal conditions. The theory, nature and limits of some expected reliability standard are build into the definitions of the terms a scientific discipline uses to describe its observations. So just recording data carefully implies something falsifiable. He needs not even have a theory, in order to be doing science. (So it does not matter if his stated theory is crap.)

As to E's contribution, any 'less ad-hoc' theory automatically predicts a wider range of outcomes. It is automatically more falsifiable and more 'normalizing', because the theory into which it embeds the explanation has many more dimensions than a more direct, non-embedded explanation. So he is doing science.

When manipulation does not simplify assumptions or tie in other science, but just picks up a theory from its base, makes it prettier and sets it back down on exactly the same base, you can argue this is only the preparation for future science, and not science itself. String theory has been accused of this. Many folks have said that until someone makes a prediction based specifically on the nature of Strings themselves, the whole thing is not yet science.

But that is not what you are describing here. He is making a different theory that is in some way more general or more deeply based in existing theories, or you would not describe it as being any less ad-hoc.