“The important thing is to be able to make predictions about images on the astronomers’ photographic plates, frequencies of spectral lines, and so on, and it simply doesn’t matter whether we ascribe these predictions to the physical effects of gravitational fields on the motion of planets and photons [as in pre-Einsteinian physics] or to a curvature of space and time.”
It’s true that a key trait of good explanations is that they can be used to make accurate predictions, but I think that taking prediction to be the point of doing science is misguided in a few ways.
Firstly, on a historical basis, many of the greatest scientists were clearly aiming for explanation not prediction. Astronomers like Copernicus and Kepler knew what to expect when they looked at the sky, but spent their lives searching for the reason why it appeared that way. Darwin knew a lot about the rich diversity of life on earth, but wanted to know how it had come about. Einstein was trying to reconcile Maxwell’s equations, the Michelson-Morley experiment, and classical mechanics. Predictions are often useful to verify explanations, but they’re rarely the main motivating force for scientists. And often they’re not the main reason why a theory should be accepted, either. Consider three of the greatest theories of all time: Darwinian evolution, Newtonian mechanics and Einsteinian relativity. In all three cases, the most compelling evidence for them was their ability to cleanly explain existing observations that had previously baffled scientists.
We can further clarify the case for explanation as the end goal of science by considering a thought experiment from Deutsch’s book. Suppose we had an “experiment oracle” that could predict the result of any experiment, but couldn’t tell us why it would turn out that way. In that case, I think experimental science would probably fade away, but the theorists would flourish, because it’d be more important than ever to figure out what questions to ask! Deutsch’s take on this:
“If we gave it the design of a spaceship, and the details of a proposed test flight, it could tell us how the spaceship would perform on such a flight. But it could not design the spaceship for us in the first place. And even if it predicted that the spaceship we had designed would explode on take-off, it could not tell us how to prevent such an explosion. That would still be for us to work out. And before we could work it out, before we could even begin to improve the design in any way, we should have to understand, among other things, how the spaceship was supposed to work. Only then would we have any chance of discovering what might cause an explosion on take-off. Prediction – even perfect, universal prediction – is simply no substitute for explanation.”
The question is now: how does this focus on explanations tie in to other ideas which are emphasised in science, like falsifiability, experimentalism, academic freedom and peer review? I find it useful to think of these aspects of science less as foundational epistemological principles, and more as ways to counteract various cognitive biases which humans possess. In particular:
We are biased towards sharing the beliefs of our ingroup members, and forcing our own upon them.
We’re biased towards aesthetically beautiful theories which are simple and elegant.
Confirmation bias makes us look harder for evidence which supports than which weighs against our own beliefs.
Our observations are by default filtered through our expectations and our memories, which makes them unreliable and low-fidelity.
If we discover data which contradicts our existing theories, we find it easy to confabulate new post-hoc explanations to justify the discrepancy.
We find it psychologically very difficult to actually change our minds.
We can see that many key features of science counteract these biases:
Science has a heavy emphasis on academic freedom to pursue one’s own interests, which mitigates pressure from other academics. Nullius in verba, the motto of the Royal Society (“take nobody’s word for it”) encourages independent verification of others’ ideas.
Even the most beautiful theories cannot overrule conflicting empirical evidence.
Scientists are meant to attempt to experimentally falsify their own theories, and their attempts to do so are judged by their peers. Double-blind peer review allows scientists to feel comfortable giving harsher criticisms without personal repercussions.
Scientists should aim to collect precise and complete data about experiments.
Scientists should pre-register their predictions about experiments, so that it’s easy to tell when the outcome weighs against a theory.
Science has a culture of vigorous debate and criticism to persuade people to change their minds, and norms of admiration for those who do so in response to new evidence.
But imagine an alien species with the opposite biases:
They tend to trust the global consensus, rather than the consensus of those directly around them.
Their aesthetic views are biased towards theories which are very data-heavy and account for lots of edge cases.*
When their views diverge from the global consensus, they look harder for evidence to bring themselves back into line than for evidence which supports their current views.
Their natural senses and memories are precise, unbiased and high-resolution.
When they discover data which contradicts their theories, they find it easiest to discard those theories rather than reformulating them.
They change their minds a lot.
In this alien species, brave iconoclasts who pick an unpopular view and research it extensively are much less common than they are amongst humans. Those who try to do so end up focusing on models with (metaphorical or literal) epicycles stacked on epicycles, rather than the clean mathematical laws which have actually turned out to be more useful for conceptual progress in many domains. In formulating their detailed, pedantic models, they pay too much attention to exhaustively replaying their memories of experiments, and not enough to what concepts might underlie them. And even if some of them start heading in the right direction, a few contrary pieces of evidence would be enough to turn them back from it—for example, their heliocentrists might be thrown off track by their inability to observe stellar parallax. Actually, if you’re not yet persuaded that this alien world would see little scientific progress, you should read my summary of The Sleepwalkers. In that account of the early scientific revolution, any of the alien characteristics above would have seriously impeded key scientists like Kepler, Galileo and others (except perhaps the eidetic memories).
And so the institutions which actually end up pushing forward scientific progress on their world would likely look very different from the ones which did so on ours. Their Alien Royal Society would encourage them to form many small groups which actively reinforced each other’s idiosyncratic views and were resistant to outside feedback. They should train themselves to seek theoretical beauty rather than empirical validation—and actually, they should pay much less attention to contradictory evidence than members of their species usually do. Even when they’re tempted to change their minds and discard a theory, they should instead remind themselves of how well it post-hoc explains previous data, and put effort into adjusting it to fit the new data, despite how unnatural doing so seems to them. Those who change their minds too often when confronted with new evidence should be derided as wishy-washy and unscientific.
These scientific norms wouldn’t be enough to totally reverse their biases, any more than our scientific norms make us rejoice when our pet theory is falsified. But in both cases, they serve as nudges towards a central position which is less burdened by species-contingent psychological issues, and better at discovering good explanations.
* Note that this might mean the aliens have different standards for what qualifies as a good explanation than we do. But I don’t think this makes a big difference. Suppose that the elegant and beautiful theory we are striving for is a small set of simple equations which governs all motion in the solar system, and the elegant and beautiful theory they are striving for is a detailed chart which traces out the current and future positions of all objects in the solar system. It seems unlikely that they could get anywhere near the latter without using Newtonian gravitation. So a circular-epicycle model of the solar system would be a dead end even by the aliens’ own standards.
Good property of scientific theory is that it serves as a data compression. Les bits you need to explain the world around you, the better theory. This is IMO very good definition of what explanation is.
Also, the compression usually is lossy, such as Newtonian mechanics.
Agreed that this points in the right direction. I think there’s more to it than that though. Consider for example a three-body problem under Newtonian mechanics. Then there’s a sense in which specifying the initial masses and velocities of the bodies, along with Newton’s laws of motion, is the best way to compress the information about these chaotic trajectories.
But there’s still an open question here, which is why are three-body systems chaotic? Two-body systems aren’t. What makes the difference? Finding an explanation probably doesn’t allow you to compress any data any more, but it still seems important and interesting.
(This seems related to a potential modification of your data compression standard: that good explanations compress data in a way that minimises not just storage space, but also the computation required to unpack the data. I’m a little confused about this though.)
Yeah, I think you’re right. There are two types of explanations:
those which compress information
those which provides us with faster algorithms to reason about the world
The three-body systems is the example of the latter. As is lots of math and computer science.
Could you expand a bit more on how you view explanation as distinct from prediction?
(As I think about the concepts, I’m finding it tricky to draw a crisp distinction between the two.)
(Just an attempt at an answer)
Both an explanation and a prediction seek to minimize the loss of information, but the information under concern differs between the two.
For an explanation, the goal is to make it as human understandable as possible, which is to say, minimize the loss of information resulting from an expert human predicting relevant phenomena.
For a prediction, the goal is to make it as machine understandable as possible, which is to say, minimize the loss of information resulting from a machine predicting relevant phenomena.
The reason there isn’t a crisp distinction between the two is because there isn’t a crisp distinction between a human and a machine. If humans had much larger working memories and more reliable calculation abilities, then explanations and predictions would look more similar: both could involve lots of detail. But since humans have limited memory and ability to calculate, explanations look more “narrative” than predictions (or from the other perspective, predictions look more “technical” than explanations).
Note that before computers and automation, machine memory and calculation wasn’t always better than the human equivalent, which would have elided the distinction between explanation and prediction in a way that could never happen today. e.g., if all you have to work with is a compass and straight edge, then any geometric prediction is also going to look like an explanation because we humans grok the compass and straightedge in a way we’ll never, without modifications anyway, grok the more technical predictions modern geometry can make. The exceptions that prove the rule are very long geometric methods/proofs, which strain human memory and so feel more like predictions than methods/proofs that can be summarized in a picture.
As machines get more sophisticated, the distinction will grow larger, as we’ve already seen in debates about whether automated proofs with 10^8 steps are “really proofs”—this gets at the idea that if the steps are no longer grokable by humans, then it’s just a prediction and not an explanation, and we seem to want proofs to be both.
In all of your examples, the new theory allowed making predictions, either more correct than previous ones (relativity, astronomy) or in situations that were previously completely un-predictable (evolution). Scientists expected good predictions to follow from good explanations, and they were in large part motivated by it.
Wiener, on the other hand, is saying it doesn’t matter what explanation you choose if all explanations yield the same prediction, in a particular field of study or experiment. And you don’t need explanations at all if they can’t ever yield different predictions (in any possible experiment). That’s a different statement.
This seems to be just a matter of definitions. Scientists are human beings, they have a wide variety of interests and goals. You can label a more narrow subset of them “science”, and then say that some of what they’re doing “isn’t science”. Or you can label everything they tend to do as “science”, because it tends to come together. But the question “what is the real point of doing science?” is just a matter of definition.
When pointing to a name like Wiener it would be great to have the full name to be able to google who you mean. In this case Norbert Wiener seems me best guess?
When you have a deep explanation sure there are points that tell that it’s deep. However I wouldn’t exactly use the word “evidence” for that.
I think it’s pretty hard to define “mathematical cleaniness”. One is almost guaranteed to discriminate against undiscovered forms of math.
There is also the problem of where can you stop if matter of fact is not a good stopping point. That is if I have a theory that makes perfect predictions and someone comes and says “but you theory doesn’t explain the phenomena” under which kind of conditions can I say “no, it does explain the phenomena?”. I am reminded of quantum mechanics where there exist multiple formulaitons which are proven to be equivalent. Would one have to start discrimanting between these which are “explanining” formulations and which are “non-explaining” formulations? What would be the critera to raise one above others?
It would also be weird if biology was incomplete until it answered the quesition “why life?” in the “meaning of life” sense. That is other disciplines than science make use of explanation and it’s not immidiately obvious which parts of that cluster is relevant to science.