I cannot imagine a model that is explanatory of past observations but incapable of predicting future observations (even if just in a statistical sense).
I think the real human goal of science is to "understand reality". Of course one might push back on that depending on their foundational perspective. But it doesn't mean that the model is perfectly representative of reality. It just means the model/science provides real understanding of reality. Finding a pattern in some observations is real knowledge about reality.
Certainly a model can provide merely a very weak explanation. I've never really thought too much about what it means to "explain" though, honestly. I'll have to give it some deep thought...
Indeed, explanations mean nothing if they don't offer usable understanding, which is only confirmed by the ability to predict natural events under a given model of reality (theory). Seers, psychics, priests and astrologers (among others) have explained the world in all sorts of ways that have satisfied their followers who understood the explanation and were relieved of their anxiety. But these explanations refered to human nature, emotional states and typical behaviour of human beings, not natural events. Fortune tellers practice psychology. Their practice of placating people may be lucrative but it does nothing to explain what the natural sciences aim to explain. So you can produce explanations, but that doesn't make them science.
One is an approximation and one is a lucky guess. Repeatability is of the essence: the more different people under different conditions are able to repeat a prediction using the same method, the more confident you can be in your scientific theory. You don't rate one (1) event, you rate repeatability.
Yeah but you can't tell which is which. Deutsch is saying, if you look harder, you could find many other numerological patterns that would give the same number for the period.
The second guy can at least make sort of ok prediction of other objects in the sky whereas the first guy either can't, or has to come up with another explanation for each one.
> you could find many other numerological patterns that would give the same number for the period.
Certainly you can work backward to find all these patterns that would also have matched that one (1) event, but that's not what repeatability is. Repeatability is the reuse of the same method. If you can reuse the same "magic numbers" to predict the next few events then you may have discovered an underlying pattern of reality, and that's where you formulate it as a scientific theory. Maybe you've discovered that f = ma. Magic? Naw, just a model of how things work in reality. But if the magic numbers were only good once then what use are they? Why would you care? It's a useless "explanation". A repeatable approximation is significant. A one time only, even precise "explanation" is rather useless. Science is about the former, not the latter.
His magic numbers work for that comet. The other guys don't work, so he doesn't get the funding to continue his ideas. The first guy's method is perfectly repeatable ... for that comet.
Repeatability is not relevant in this situation. Phenomenological models can be accurate and repeatable. Actually, repeatability has more to do with the measurement, not the explanation.
Sorry, you lost me. I don't know what you're saying here. Repeating the same measurement will simply yield the same result (unless you're sloppy) so what's the point? What am I missing?
But back up. You said any theory which represents the best explanation will be an accurate predictor. This may not be the case due to measurement error and uncertainty jn parameters that go in the second guys model.
It is still the superior explanation because it has fewer parameters and its structure is not easily varied arbitrarily like the first guy's numerology ( why that kings birthday ? Why the date of that battle and not another? )
Deutsch is arguing you can do more than wait and hope. His book 'Beginning of Infinity' is great and explores this, if I remember correctly. There is also a great Ted Talk by him on this topic.
The key question is how you define 'explanatory'. If you interpret it as providing a causal explanation, than many theories are not explanatory. Newton's theory of gravity doesn't explain WHY an apple falls from the tree, it only provides a very precise model that allows you to 'predict' the process
Sure, science provides explanations. However, we value explanations precisely because they reduce uncertainty. And one of the main ways that explanations reduce uncertainty is by allowing us to accurately predict the future. And in science, this means making calculations based on mathematical models (or what we used to call "theories").
I don't think of explanation and prediction as two separate things. For me the paradigm of science is Galileo pointing his telescope at the night sky. He observes "stars" never seen before. He makes notes and tries to make sense of them. With the understanding (i.e. explanation) that he is watching satellites of Jupiter he is able to calculate the periods and radii of their orbits. The explanation leads to the calculation and predictions. Accurate predictions prove the practical/pragmatic value of the explanation (I skip over metaphysical concerns like the "truth" value of the explanation). You cannot really have one without the other.
I would say that explanation and prediction are not considered separable, except in one specific circumstance. The "instrumentalist and positivist views" that Deutsche is critical off don't exist outside of quantum mechanics (which inherited these views from charismatic Niels Bohr and co in the 1920s, when positivism was all the rage in Europe).
The distinction Deutsch emphasises really only exists in relation to quantum physicists abandoning realism and explanation in the 1920s (against the wishes of Schrödinger and Einstein). No other branch of science did this and the dilemma doesn't occur anywhere else in science. Like many physicists, Deutsch may be guilty of the assumption that science = physics (it's a while since I read TFOR).
If you want arguments for scientific pluralism, then look up Hasok Chang. His book Is Water H₂O? is fascinating (especially his account of phlogiston). Also his recent book Realism for Realistic People: A New Pragmatist Philosophy of Science looks awesome (I haven't read it yet, but soon!).
On the baleful influence of positivism on quantum mechanics, read Mara Beller's Quantum Dialogues. Adam Becker's book What is Real? (terrible title) also looks at the internal politics of quantum physics, like e.g. the pillorying and exclusion of David Bohm.
You might also want to look into the idea of "explanation". On which see
* Faye, Jan.(2007). "The Pragmatic-Rhetorical Theory of Explanation." In Rethinking Explanation. Boston Studies in the Philosophy of Science. 43-68. Edited by J. Persson and P. Yikoski. Dordrecht: Springer.
that misses the whole point. but its ok, many people miss it. an explanation gives you a few things. it involves a non statistical model which involves creating objects and their causal connections. it gives you predictions, but the key is that the same prediction for a given question, can come from different models of the underlying causality. that is why prediction by itself is not as strong as a "good" explanation (which David defines as ones that are hard to vary see text).
Explanation is fundamentally different from prediction. For example, I can explain something that happened by saying "a god willed it". But this explanation has no predictive value.
I would ask can we neatly separate explanation and prediction? To me it seems explanations are implicit predictions, and predictions are implicit explanations waiting to be validated.
Im not versed in philosophy so maybe there is some difference im not appreciating.
Ptolemaic cosmological models are a great counter here. They could and still can successfully predict the movement and location of celestial bodies, but with truly ridiculous assumptions no one believes, or ever believed, mirrored reality.
They were rejected because they lack any explanatory content. Great prediction machines, terrible at explanation.
Yeah, but this is still dependent on what perspective you take on the things a model predicts. In this example it is assumed that Ptolemaic models only predict the movement and locations of celestial bodies.
But arguably they make far more predictions having to do with the mechanisms of how these models operate and calculate these movements and locations. And it turns out these predictions are false.
I would say it is far more useful to view the totality of assumptions a model makes as part of the model. And these assumptions form falsifiable predictions. Seen in this light, prediction and explanation collapse.
The predictions aren’t false, however. In fact, I read a paper once arguing that a Ptolemaic model can be used to generate any pattern whatsoever. So in terms of predictive accuracy, it’s as tight a connection as you’d like it to be.
But its explanatory content is zero. No one believes the operations in play in a Ptolemaic model are anything close to reality.
And further, even many of its practitioners historically never believed in the reality of the mechanisms. It was used strictly to make accurate predictions of celestial bodies.
High prediction with low explanation. In other words, instrumentalist.
I think you may have misunderstood my point. I wasn't arguing that the models predictions of the motion of celestial bodies were inaccurate.
I was arguing that what you call explanatory content may be considered part of the larger set of the model's predictions. This set consists of the predictions in motion of celestial bodies, but also all the other implicit assumptions involving the operations used to calculate the motion.
No, I got it. But what I’m saying is that the model’s builders themselves did not purport to be in the Explanation-Business at all, only prediction.
So yes, while it can be true that explanatory content could be considered to be contained within the model’s overall set of predictions, it is not the case here, which is one expressly about prediction IN LIGHT OF a lack of commitment to explanation.
Which is an answer to the OPs question: can explanation and prediction be done apart from one another.
I see. So would you argue that the intentions of the modeler determines whether some property of the model is considered a prediction or an explanatory property?
If so, then I suppose there is a case to be made there, yes. Still, it means the difference between prediction and explanatory content is quite thin.
The aims of the modeler should be taken in consideration, absolutely. It would be unfair and inappropriate to criticize a model for lacking a property it was not designed to manifest. The Ptolemaic modelers explicitly needed a system capable of generating accurate tables of celestial phenomena. They solved the problem of needing a system capable of refining the accuracy of predictions.
However, to your other point, about the “thin” separation between prediction and explanation: I disagree. Think about the relation to any explanatory model coming out of evolutionary theory. Highly explanatory models of inclusive fitness, for instance, are wholly incapable to predicting the evolutionary trajectory of any given species.
The explanatory content of fitness models is unquestioned, but almost zero chance of accurately predicting the state of equines in a hundred thousand years.
Explanation and prediction are two completely distinct epistemic virtues within science.
Right, this example of evolution where you have explanation but no prediction is far stronger actually.
I still have a lingering worry that maybe these evolutionary theories are still too vague and may be replaced by better theories that can explain AND predict. However, chaotic systems are a thing, so perhaps you are right that explanation and prediction should be distinguished.
Well, much of evolutionary theory is stochastic process, so there will inherently be a fuzziness and irreducible statistical nature to the problem set. I can tell you why, generally, prey species will increase in speed over time, but I cannot predict when the fastest one will run off a cliff.
I was taught that explanatory projects are attempts to answer Why-Questions. Hey dad, why do tigers have stripes, why are mammalian prey species’ eyes so far apart, why do black holes have so much mass?
In this light, prediction seems like a completely distinct sort of endeavor. Why do the payouts in Craps Games look the way they do? Ok, so pull from that the correct bet on the next throw of the dice…
An explanation of lightning will at some point involve laying out Maxwell’s Equations on electricity, but a heat map of lighting strikes meant to warn people about increased dangers of being struck will involve only probability theory plus meteorological data.
Prediction and explanation are just two distinct AIMS of scientific practice.
It’s not true that evolution has no power of prediction.
That the number of antibiotic resistant bacteria will continue to grow is a direct prediction of evolution, for example. Another would be that if I separate two populations of the same species and they get exposed to different environments, they will diverge in their development from generation to generation.
> I would ask can we neatly separate explanation and prediction?
Yes. Deutsch spends plenty of time doing this. And it boils down to counterfactuals. Deutsch uses the phrase “hard to vary”.
Consider a prediction of the seasons vs an explanation of the seasons. A prediction of the seasons is rather like a calendar. Whereas an explanation would be something like the axial tilt theory.
A calendar is easy to vary. Knowing about seasons locally — say in the UK — tells you nothing about seasons globally. If you had a predictive calendar telling you when winter and summer are, and then you found out the winter and summer were opposite in the southern hemisphere, the calendar is easy to update (not hard to vary) and provides no further counterfactual detail to explain why it must be this way nor what the world would be like if it wasn’t.
The axial tilt explanation for the season is the opposite. It says the seasons would be opposite — and if counterfactually, it turned out it wasn’t… there would be no way at all to update the theory to handle the fact that the the southern hemisphere didn’t have opposite seasons. It is hard to vary.
This is very important as science makes progress through falsification rather than successful prediction. The value of the theory can be measured in what it would eliminate from the space of possibility if it was to be falsified. The calendar (the prediction) eliminated nothing. The theory is easily modified. Well, the explanation, the axial tilt theory, eliminates in enormous space of explanation if it’s wrong.
Moreover, explanatory theories are counterfactually explanatory. A calendar tells us nothing about how Seasons would behave if the Earth were a different shape or had a different tilt. The axial tilt theory tells us this. It’s also global (and universal). It tells us about seasons on planets we’ve never studied or measured.
I will assert that an explanation tells the 'how' and 'why' of things, and a prediction tells us 'what.' Now this naturally implies that an explanation need not always lead to a prediction, and the converse is also the case.
Although there is an interesting point to discuss. Seemingly explanations lead to more explanations, which in turn explain consequences, which in turn lead to predictions. Now I will argue that the primary difference between the two lies in HOW they are evaluated. Explanations can be criticised or falsified via logical argumentation and internal inconsistency. Predictions, on the other hand are done so via observations. For example take two predictions:
An eclipse will occur tomorrow
An eclipse will not occur tomorrow
These two are contrarian to each other, but none of them are falsified; an observation falsifies them.
TL;DR: The two have overlapping areas, but they are not evaluated in the same manner.
Prediction is a well defined concept.
Explanation on the other hand less so, I don't even understand what explanation means, as to me it sounda quite subjective.
Its semantic word games. The explanation has predictive power. The explanation explains why the prediction isnt a statistical guess, but a logical certainty.
I'm not predicting that gravity causes an apple to fall, im using gravity as an explanation. Because the explanation accurately predicts what explanation fits the observations and data.
Biologist here, so this is going to be controversial. From where I stand, unexplained prediction is the starting point. Explanation comes second.
Fundamentally, this argument boils down to what the definition of the 'Explanation' is. Let me start with mine.
When a dog hears a bell ringing, a bunch of neurons connected to the sense of hearing, at specific frequencies, get activated, start growing, and secrete signalling molecules. When the dog gets fed soon after, a bunch of neurons connected to the senses of smell, taste, and vision get activated and start growing, but towards the previously activated bunch of neurons, guided there by the signaling molecules. Eventually the different bunches of neurons get connected and transmit neural signal from one to another, creating a mechanism for what we would describe in mathematics as a logical implication (a->b). The dog now expects food when the bell rings. Their brain created a simple, specific mental representation of parts of the world around them and how they relate. Many such connections create chains of thoughts, and human brains in particular are pretty good at devoting most of our neural signaling to bouncing signals back and forth inside that network of chains of thoughts in repeating patterns, creating elaborate mental representations that are very general. Our ability to create general mental representations about the world around us is behind our evolutionary success.
These elaborate general mental representations are the definition of 'the Explanation' that I will now use below to answer OP's question.
Good explanations make correct predictions. Great explanations make correct predictions at a very general level.
If an explanation makes incorrect predictions, it's not a very good explanation. Clinging onto such explanation is what we would call a mental disorder.
If the explanation cannot make predictions, then it's unfalsifiable, and thus worthless. Clinging onto such explanation is what we would call a cult or a religion.
Now, can you have a prediction and not have an explanation? There are many phenomena that we repeatedly observe but can't explain yet in a way that is coherent with our other general explanations about the world. That's usually the starting point for when the brain searches for new, general explanations.
I am a doctor so I approach this with a strong bias in favor of prediction.
If I have a drug that cures your cancer, what matters is that it does that. Take the drug, cure the cancer.
How the drug destroys cancer is of little importance to me. Does it strangle the cancer? Does it poke tiny holes in it? Does it work by altering your metabolism or perhaps your chakra?
Why does it work? If I can give you a drug that works 70% of the time and I can fully explain exactly how it works, or I have a drug that works 95% of the time and no one has any idea how it works, but the evidence that it does work 95% of the time is strong, I will never use the 70% drug again.
In medicine people who want to tell you how a drug works are almost always trying to sell you the drug. They work for the drug company and they love to explain how their drug works. They are generally attractive people. They love talking about the mechanism of action.
I want to see the clinical evidence it works. Over time I have become frustrated with the explainers.
The way these are posed as almost binary choices I’d reject.
How does an infant, toddler, or child interact with the world? They do all of experiment, predict, and attempt to explain.
Scientists also try to predict, experiment, and necessarily implicitly presuppose ontological constraints. Behaviorists/positivists can’t avoid explanation inclusive of implied ontological presuppositions no matter how hard they try.
On the other hand, explanations that provide zero predictive value easily wind up in arbitrary language games.
For a quick read on behaviorism, operationalism, and positivism see https://www.lehigh.edu/~mhb0/Operationalism.pdf Piaget is also a great reference here.
His position is more self-aggrandizing than substantive, as if he alone or he predominately is the savior of the scientific method from the barren waste-land of mere prediction.
The major arguments against prioritizing explanation over prediction include unintelligibility, being that “explanation” is ill-defined; narrative superabundance, being that explanatory adequacy can be obtained with numerous incompatible explanatory framings; and epistemic peril, insofar as the criterion for sufficiency for explanation is psychological whereas for prediction it is empirical.
> The major arguments against prioritizing explanation over prediction include unintelligibility, being that “explanation” is ill-defined;
An explanation is a conjecture about something unobserved which attempts to account for what is observed.
The key differences between a good explanation and a mere model are:
Good explanations are hard to vary while models are necessarily easy to vary
Explanations are counterfactual statements while models need not be
For example, when talking about the observation that the Earth has cyclical seasons, a mirror model of the Seasons would be something akin to a Calendar, whereas an explanation of those seasons would be entirely different such as the axial tilt theory.
The model is easy to vary. If it turns out that year is a month longer, we can simply add a month to the model. If we discovered that the southern hemisphere has opposite seasons to the north, we can simply create a southern hemisphere model to go along with the northern one.
But the axial tilt theory predicts that the southern hemisphere has opposite seasons. If instead we had found that it did not, we wouldn’t be able to modify the axial tilt theory without ruining the explanation entirely. We would have to throw the entire explanation out. And that’s a good thing because scientific theories are measured in terms of what they would rule out if they were to be falsified. If you could easily modify them, they would rule anything out except for the exact results of that exact experiment.
Second, the calendar tells us absolutely nothing about what we would need to do to change the seasons. Models are not counter factual claims. Whereas explanations are. An explanation for the earth’s seasons, such as the axial tilt theory explains that if we want different seasons, we can change the variables in the explanation such as the shape of the Earth, the angle of tilt its relationship to the sun, etc.
It even tells us about the seasons on planets we’ve never been to — or which don’t exist.
> narrative superabundance, being that explanatory adequacy can be obtained with numerous incompatible explanatory framings;
Yes that’s the idea of science. You have multiple competing conjectures and then you do experiments to differentiate them. When all available explanations work, you look to parsimony to understand which explanation is better. “Good” meaning what I said earlier: hard to vary. We can actually mathematically prove this principle of parsimony through the technique of Solomonoff induction.
I’m not sure what to make of your reply. I was addressing OP, whose study of the subject is in early days. For that reason, a conversational tone was warranted.
I’ll offer that you’re overshooting the mark in your reply to me, if not shooting off target entirely.
If that is so, then it'd be more open for non-mainstream works, opinions, thoughts and experienced to be also considered, however unfortunately the social grooming into certain positions as well as the ego and its often to be observed inability to accept things further outside of it's own narrative and experience restrict the pursuit of understanding and explanation of things scientifically thus.
Realism very wide-spread among practically every sector -- philosophers of science, scientists, and laypeople alike -- but privileging explanation over empirical power is not something I read or even hear often because it boils down to privileging metaphysics over the scientific method. You can't falsify explanations, only predictions.
I'm an instrumentalist -- I think scientific realism is a historical reflection of religion, that it misunderstands the human condition, and most importantly, that it being so widespread is something that makes people worse at science -- so I have no sympathy whatsoever for even more moderate realist positions; however, most realists at least engage with the problems that human epistemic limitations pose. Unlike me and mine, they just believe those limitations are inconsequential or surmountable.
Leaning back into metaphysics is an outright regression, though. Having scientists even more attached to the unfalsifiable components of their particular theories, more unwilling to change their minds when faced with counter-evidence, is not a way forward.
Consider this, if a theory could predict but not understand, would it serve the interests of science? IMO, no. You cannot build a deeper understanding of a system if you do not understand how all its essential causal mechanisms act and interact. So understanding is essential to science. (BTW, AI is often capable of predicting a outcome but often cannot explain the mechanisms that caused it. That's just good pattern matching, not science.)
What if a theory understands but can't predict? Is that even possible? IMO, no. If a theory can explain the mechanistic cause of an outcome, it necessarily can also explain the consequences of activating that mechanism. So prediction is the necessary consequence of understanding.
Mathematician here. I've been thinking about something equivalent: "inside our heads" and "outside our heads." Roughly models are inside our heads and measurements are outside.
Realists, transcendentalists, positivist, etc. just differ in their ratios of inside / outside.
Positivism, unlike other answers state here, was critical to modern science. Foundational in the late 1700s and 1800s. Check out Britannica's article, it's great. Electromagnetism, thermodynamics, etc. were sciences where we discovered piecemeal laws first before sticking them together into nice general differential equations. So you kind of need some kind of positivism to even start understanding pressure / volume for example.
Or maybe not. Remember that both of these sciences had very immediate applied interests around them (electrical grid and transportation), and positivism is the science of industry. Americans still follow this philosophy in general, which it's no accident we're leaders in AI at the moment, maybe the most positivistic version of knowledge ever implemented at scale. Who knows how far it will go?
An equivalent way to state inside / outside, or explanation / prediction, or model / measurement, or positivist / realist is prior / posterior from Bayesian statistics. In the end, what is the difference exactly? The more I think about it the less clear it becomes.
Prior is what we think before the experiment, it's our "bias" and the posterior is what we've "learned" after we measure. But of course, this assumes a stable reality between prior and posterior, which quantum mechanics and the uncertainty principle threatens. But even in all the other branches of science you can see this problem, we just roll that distrubance from measurement into "noise". Measuring even the temperature still cools the object, ever so slightly, for example.
In the end, I think we're still exploring the boundaries of positivism / realism. If for example, we finally grow or build a replicate of the human brain, and could map it's wiring, would that produce a positivist or realist theory of cognition? It's immediately filled with contradictions and confusions if you try to think about it.
I think the positivist / realist distinction is best thought more concretely as a continuum of model dimensionality vs predictive capability. You want low dimensional models that produce wide ranging (in space / time) predictions. In other words, lots of equations with few emperical physical constants. AI models for this reason are not immediately useful for "foundational science." But minimizing the dimensionality / predictive power ratio is also itself just partially because it's easier to think about simpler theories than complex ones. So we're back at a kind of positivism again. (There is a nice discussion between Peter Norvig and Noam Chomsky about this).
Great question! I'll be thinking about it for awhile!
Deutsch is a hardcore Everettian and I think has sometimes implied that his own insights (that were foundational to quantum computing) were only possible because of his preference for the multiverse interpretation of QM rather than collapse. But (it goes without saying) that all interpretations of QM are functionally equivalent and all of quantum computing theory can be accessed with the tools of QM regardless of what interpretation you happen to prefer. Or as a rational person might say, there is not real reason to choose one interpretation over another.
So he is most likely defending the fact that he has a preference, by attributing greater explanatory power to it - a quality which cannot be objectively measured, and he can freely assert it without anyone else being able to tell him he's wrong. Which is very much nothing at all to do with science.
I havent read this but ive seen scientists in psychology and stats draw a distinction between prediction and explanation to think about different kinds of questions we can ask, and i think this is a useful distinction.
this can be a correlation/causation point. We may observe X and Y are related, but that doesnt mean X directly causes Y, or that we can increase Y by changing X. We could say “what factors are asdociated with Y?” Is a prediction question while “what factors affect Y?” is an explanation question.
In stats, This also sounds to be like the prediction/inference distinction that comes up with regression models & machine learning. Sometimes you want a model that takes a bunch of data and does a good job predicting the future. Especially if youre like a business. But in some techniques, models are black boxes and it is difficult to parse to understand how specific factors relate to future outcomes. And people might be more interested in analyzing/doing inference with coeffiecients (individual factors or interactions) to parse out what is driving the prefictio. Sometimes using simpler models help to parse out specific factors (explaining), even with less predictive power
Also in my view experimentation is fundamentally a method for explanation, so im a bit lost when you contrast “prediction and experimentation” vs explanation
Edit: as others say, for a philosophical discussion we would need a formal definition of these distinctions
Good point, although I mean then what does Deutsch actually argue for? It seems to me that via treating science as a method, explanation and prediction are in some way steps of the method? I'm quite confused here.
I think it’s a confusion of terms. Making predictions and testing them is a core part of the scientific method. It’s a technique. An explanation is the goal. If a prediction is falsified, that just means the proposed explanation is either partially or wholly incorrect, but it doesn’t remove the need for an explanation.
I can give a real life example. Years ago, a colleague shared her patients’ clinical data with me, to discuss possible treatments strategies. I noticed that the data was not normally distributed.
Why was this? This observation demanded an explanation. To explain it we started with various predictions. The first was that it was an artefact of data. So we checked patients’ ages, other illnesses, family history, etc. Nothing there. So the next prediction was that these patients would have different clinical outcomes, which turned out to be the case.
As others have pointed out, Deutsch works mainly in quantum theory, which is “special” when it comes to science.
I have not read this work, but he is a major proponent of the many worlds interpretation of QM and Quantum computing.
The multiverse as an idea is not falsifiable, and this is a core weakness underlying any explanation that relies on the multiverse. When you understand that, his point of view makes sense.
telephantomoss | a day ago
I cannot imagine a model that is explanatory of past observations but incapable of predicting future observations (even if just in a statistical sense).
I think the real human goal of science is to "understand reality". Of course one might push back on that depending on their foundational perspective. But it doesn't mean that the model is perfectly representative of reality. It just means the model/science provides real understanding of reality. Finding a pattern in some observations is real knowledge about reality.
canbooo | 22 hours ago
Opposite is completely possible though, predicting well without explaining (well). We can predict the quantum effects but are they explained?
telephantomoss | 19 hours ago
Certainly a model can provide merely a very weak explanation. I've never really thought too much about what it means to "explain" though, honestly. I'll have to give it some deep thought...
sohcahtoa | 22 hours ago
Indeed, explanations mean nothing if they don't offer usable understanding, which is only confirmed by the ability to predict natural events under a given model of reality (theory). Seers, psychics, priests and astrologers (among others) have explained the world in all sorts of ways that have satisfied their followers who understood the explanation and were relieved of their anxiety. But these explanations refered to human nature, emotional states and typical behaviour of human beings, not natural events. Fortune tellers practice psychology. Their practice of placating people may be lucrative but it does nothing to explain what the natural sciences aim to explain. So you can produce explanations, but that doesn't make them science.
Phssthp0kThePak | 21 hours ago
Sure. Say some predicts the recurrence of a comet, but their explanation is some mystic numerology from a holy book. It works.
Now someone gives the real explanation, but based on available data, it's off by a year. How do you rate these two approaches?
sohcahtoa | 21 hours ago
One is an approximation and one is a lucky guess. Repeatability is of the essence: the more different people under different conditions are able to repeat a prediction using the same method, the more confident you can be in your scientific theory. You don't rate one (1) event, you rate repeatability.
Phssthp0kThePak | 21 hours ago
Yeah but you can't tell which is which. Deutsch is saying, if you look harder, you could find many other numerological patterns that would give the same number for the period.
The second guy can at least make sort of ok prediction of other objects in the sky whereas the first guy either can't, or has to come up with another explanation for each one.
sohcahtoa | 20 hours ago
> you could find many other numerological patterns that would give the same number for the period.
Certainly you can work backward to find all these patterns that would also have matched that one (1) event, but that's not what repeatability is. Repeatability is the reuse of the same method. If you can reuse the same "magic numbers" to predict the next few events then you may have discovered an underlying pattern of reality, and that's where you formulate it as a scientific theory. Maybe you've discovered that f = ma. Magic? Naw, just a model of how things work in reality. But if the magic numbers were only good once then what use are they? Why would you care? It's a useless "explanation". A repeatable approximation is significant. A one time only, even precise "explanation" is rather useless. Science is about the former, not the latter.
Phssthp0kThePak | 20 hours ago
His magic numbers work for that comet. The other guys don't work, so he doesn't get the funding to continue his ideas. The first guy's method is perfectly repeatable ... for that comet.
sohcahtoa | 20 hours ago
Again, the idea of repeatability is not to repeat the same calculation for the same single event, it is to reuse it for any number of other events.
Phssthp0kThePak | 20 hours ago
Repeatability is not relevant in this situation. Phenomenological models can be accurate and repeatable. Actually, repeatability has more to do with the measurement, not the explanation.
sohcahtoa | 20 hours ago
Sorry, you lost me. I don't know what you're saying here. Repeating the same measurement will simply yield the same result (unless you're sloppy) so what's the point? What am I missing?
Phssthp0kThePak | 20 hours ago
Measurement error.
But back up. You said any theory which represents the best explanation will be an accurate predictor. This may not be the case due to measurement error and uncertainty jn parameters that go in the second guys model.
It is still the superior explanation because it has fewer parameters and its structure is not easily varied arbitrarily like the first guy's numerology ( why that kings birthday ? Why the date of that battle and not another? )
telephantomoss | 19 hours ago
This is a classic dilemma I figure. You just keep collecting data, and hopefully one approach will emerge as better.
Phssthp0kThePak | 19 hours ago
Deutsch is arguing you can do more than wait and hope. His book 'Beginning of Infinity' is great and explores this, if I remember correctly. There is also a great Ted Talk by him on this topic.
ScatteredDandelion | 6 hours ago
The key question is how you define 'explanatory'. If you interpret it as providing a causal explanation, than many theories are not explanatory. Newton's theory of gravity doesn't explain WHY an apple falls from the tree, it only provides a very precise model that allows you to 'predict' the process
Prajnamarga | a day ago
Sure, science provides explanations. However, we value explanations precisely because they reduce uncertainty. And one of the main ways that explanations reduce uncertainty is by allowing us to accurately predict the future. And in science, this means making calculations based on mathematical models (or what we used to call "theories").
I don't think of explanation and prediction as two separate things. For me the paradigm of science is Galileo pointing his telescope at the night sky. He observes "stars" never seen before. He makes notes and tries to make sense of them. With the understanding (i.e. explanation) that he is watching satellites of Jupiter he is able to calculate the periods and radii of their orbits. The explanation leads to the calculation and predictions. Accurate predictions prove the practical/pragmatic value of the explanation (I skip over metaphysical concerns like the "truth" value of the explanation). You cannot really have one without the other.
I would say that explanation and prediction are not considered separable, except in one specific circumstance. The "instrumentalist and positivist views" that Deutsche is critical off don't exist outside of quantum mechanics (which inherited these views from charismatic Niels Bohr and co in the 1920s, when positivism was all the rage in Europe).
The distinction Deutsch emphasises really only exists in relation to quantum physicists abandoning realism and explanation in the 1920s (against the wishes of Schrödinger and Einstein). No other branch of science did this and the dilemma doesn't occur anywhere else in science. Like many physicists, Deutsch may be guilty of the assumption that science = physics (it's a while since I read TFOR).
If you want arguments for scientific pluralism, then look up Hasok Chang. His book Is Water H₂O? is fascinating (especially his account of phlogiston). Also his recent book Realism for Realistic People: A New Pragmatist Philosophy of Science looks awesome (I haven't read it yet, but soon!).
On the baleful influence of positivism on quantum mechanics, read Mara Beller's Quantum Dialogues. Adam Becker's book What is Real? (terrible title) also looks at the internal politics of quantum physics, like e.g. the pillorying and exclusion of David Bohm.
On Bohr's influence on other people see comments in Sean Carroll's interview of David Albert. (staring at 41:42).
You might also want to look into the idea of "explanation". On which see
* Faye, Jan.(2007). "The Pragmatic-Rhetorical Theory of Explanation." In Rethinking Explanation. Boston Studies in the Philosophy of Science. 43-68. Edited by J. Persson and P. Yikoski. Dordrecht: Springer.
BerkeleyYears | 23 hours ago
that misses the whole point. but its ok, many people miss it. an explanation gives you a few things. it involves a non statistical model which involves creating objects and their causal connections. it gives you predictions, but the key is that the same prediction for a given question, can come from different models of the underlying causality. that is why prediction by itself is not as strong as a "good" explanation (which David defines as ones that are hard to vary see text).
[OP] Unlucky-Prior-1838 | a day ago
Thanks for the further reading; your thoughts are interesting.
SeeBuyFly3 | 23 hours ago
Explanation is fundamentally different from prediction. For example, I can explain something that happened by saying "a god willed it". But this explanation has no predictive value.
Same with string theory, many worlds, etc.
Prajnamarga | 3 hours ago
We are talking about scientific explanations... so this example really, really does not fly.
String theory explains exactly nothing in our world.
Many worlds is speculative metaphysics rather than an explanation of anything.
SimonsToaster | a day ago
I would ask can we neatly separate explanation and prediction? To me it seems explanations are implicit predictions, and predictions are implicit explanations waiting to be validated.
Im not versed in philosophy so maybe there is some difference im not appreciating.
DumbScotus | a day ago
Yeah. I tend to agree with Deutsch… but realistically the way you evaluate a theory’s explanatory power is by testing its predictive power.
Empty_Expressionless | a day ago
I'd go further and say that statements can only be true or false if they are falsifiable.
trinfu | a day ago
Ptolemaic cosmological models are a great counter here. They could and still can successfully predict the movement and location of celestial bodies, but with truly ridiculous assumptions no one believes, or ever believed, mirrored reality.
They were rejected because they lack any explanatory content. Great prediction machines, terrible at explanation.
Semantic_Internalist | a day ago
Yeah, but this is still dependent on what perspective you take on the things a model predicts. In this example it is assumed that Ptolemaic models only predict the movement and locations of celestial bodies.
But arguably they make far more predictions having to do with the mechanisms of how these models operate and calculate these movements and locations. And it turns out these predictions are false.
I would say it is far more useful to view the totality of assumptions a model makes as part of the model. And these assumptions form falsifiable predictions. Seen in this light, prediction and explanation collapse.
trinfu | a day ago
The predictions aren’t false, however. In fact, I read a paper once arguing that a Ptolemaic model can be used to generate any pattern whatsoever. So in terms of predictive accuracy, it’s as tight a connection as you’d like it to be.
But its explanatory content is zero. No one believes the operations in play in a Ptolemaic model are anything close to reality.
And further, even many of its practitioners historically never believed in the reality of the mechanisms. It was used strictly to make accurate predictions of celestial bodies.
High prediction with low explanation. In other words, instrumentalist.
Semantic_Internalist | a day ago
I think you may have misunderstood my point. I wasn't arguing that the models predictions of the motion of celestial bodies were inaccurate.
I was arguing that what you call explanatory content may be considered part of the larger set of the model's predictions. This set consists of the predictions in motion of celestial bodies, but also all the other implicit assumptions involving the operations used to calculate the motion.
trinfu | a day ago
No, I got it. But what I’m saying is that the model’s builders themselves did not purport to be in the Explanation-Business at all, only prediction.
So yes, while it can be true that explanatory content could be considered to be contained within the model’s overall set of predictions, it is not the case here, which is one expressly about prediction IN LIGHT OF a lack of commitment to explanation.
Which is an answer to the OPs question: can explanation and prediction be done apart from one another.
Semantic_Internalist | a day ago
I see. So would you argue that the intentions of the modeler determines whether some property of the model is considered a prediction or an explanatory property?
If so, then I suppose there is a case to be made there, yes. Still, it means the difference between prediction and explanatory content is quite thin.
trinfu | a day ago
The aims of the modeler should be taken in consideration, absolutely. It would be unfair and inappropriate to criticize a model for lacking a property it was not designed to manifest. The Ptolemaic modelers explicitly needed a system capable of generating accurate tables of celestial phenomena. They solved the problem of needing a system capable of refining the accuracy of predictions.
However, to your other point, about the “thin” separation between prediction and explanation: I disagree. Think about the relation to any explanatory model coming out of evolutionary theory. Highly explanatory models of inclusive fitness, for instance, are wholly incapable to predicting the evolutionary trajectory of any given species.
The explanatory content of fitness models is unquestioned, but almost zero chance of accurately predicting the state of equines in a hundred thousand years.
Explanation and prediction are two completely distinct epistemic virtues within science.
Semantic_Internalist | a day ago
Right, this example of evolution where you have explanation but no prediction is far stronger actually.
I still have a lingering worry that maybe these evolutionary theories are still too vague and may be replaced by better theories that can explain AND predict. However, chaotic systems are a thing, so perhaps you are right that explanation and prediction should be distinguished.
trinfu | a day ago
Well, much of evolutionary theory is stochastic process, so there will inherently be a fuzziness and irreducible statistical nature to the problem set. I can tell you why, generally, prey species will increase in speed over time, but I cannot predict when the fastest one will run off a cliff.
I was taught that explanatory projects are attempts to answer Why-Questions. Hey dad, why do tigers have stripes, why are mammalian prey species’ eyes so far apart, why do black holes have so much mass?
In this light, prediction seems like a completely distinct sort of endeavor. Why do the payouts in Craps Games look the way they do? Ok, so pull from that the correct bet on the next throw of the dice…
An explanation of lightning will at some point involve laying out Maxwell’s Equations on electricity, but a heat map of lighting strikes meant to warn people about increased dangers of being struck will involve only probability theory plus meteorological data.
Prediction and explanation are just two distinct AIMS of scientific practice.
ipreuss | 22 hours ago
It’s not true that evolution has no power of prediction.
That the number of antibiotic resistant bacteria will continue to grow is a direct prediction of evolution, for example. Another would be that if I separate two populations of the same species and they get exposed to different environments, they will diverge in their development from generation to generation.
trinfu | a day ago
Regardless, thanks for the pushback and the discussion!
fox-mcleod | a day ago
> I would ask can we neatly separate explanation and prediction?
Yes. Deutsch spends plenty of time doing this. And it boils down to counterfactuals. Deutsch uses the phrase “hard to vary”.
Consider a prediction of the seasons vs an explanation of the seasons. A prediction of the seasons is rather like a calendar. Whereas an explanation would be something like the axial tilt theory.
A calendar is easy to vary. Knowing about seasons locally — say in the UK — tells you nothing about seasons globally. If you had a predictive calendar telling you when winter and summer are, and then you found out the winter and summer were opposite in the southern hemisphere, the calendar is easy to update (not hard to vary) and provides no further counterfactual detail to explain why it must be this way nor what the world would be like if it wasn’t.
The axial tilt explanation for the season is the opposite. It says the seasons would be opposite — and if counterfactually, it turned out it wasn’t… there would be no way at all to update the theory to handle the fact that the the southern hemisphere didn’t have opposite seasons. It is hard to vary.
This is very important as science makes progress through falsification rather than successful prediction. The value of the theory can be measured in what it would eliminate from the space of possibility if it was to be falsified. The calendar (the prediction) eliminated nothing. The theory is easily modified. Well, the explanation, the axial tilt theory, eliminates in enormous space of explanation if it’s wrong.
Moreover, explanatory theories are counterfactually explanatory. A calendar tells us nothing about how Seasons would behave if the Earth were a different shape or had a different tilt. The axial tilt theory tells us this. It’s also global (and universal). It tells us about seasons on planets we’ve never studied or measured.
[OP] Unlucky-Prior-1838 | a day ago
I will assert that an explanation tells the 'how' and 'why' of things, and a prediction tells us 'what.' Now this naturally implies that an explanation need not always lead to a prediction, and the converse is also the case.
Although there is an interesting point to discuss. Seemingly explanations lead to more explanations, which in turn explain consequences, which in turn lead to predictions. Now I will argue that the primary difference between the two lies in HOW they are evaluated. Explanations can be criticised or falsified via logical argumentation and internal inconsistency. Predictions, on the other hand are done so via observations. For example take two predictions:
An eclipse will occur tomorrow
An eclipse will not occur tomorrow
These two are contrarian to each other, but none of them are falsified; an observation falsifies them.
TL;DR: The two have overlapping areas, but they are not evaluated in the same manner.
knockingatthegate | a day ago
Define “how”, “why” and “what.” Alternately, read past the first chapter.
[OP] Unlucky-Prior-1838 | 23 hours ago
Sure
Vegetable_Home | a day ago
Prediction is a well defined concept. Explanation on the other hand less so, I don't even understand what explanation means, as to me it sounda quite subjective.
Riokaii | 23 hours ago
Its semantic word games. The explanation has predictive power. The explanation explains why the prediction isnt a statistical guess, but a logical certainty.
I'm not predicting that gravity causes an apple to fall, im using gravity as an explanation. Because the explanation accurately predicts what explanation fits the observations and data.
Desolsh | 22 hours ago
Biologist here, so this is going to be controversial. From where I stand, unexplained prediction is the starting point. Explanation comes second.
Fundamentally, this argument boils down to what the definition of the 'Explanation' is. Let me start with mine.
When a dog hears a bell ringing, a bunch of neurons connected to the sense of hearing, at specific frequencies, get activated, start growing, and secrete signalling molecules. When the dog gets fed soon after, a bunch of neurons connected to the senses of smell, taste, and vision get activated and start growing, but towards the previously activated bunch of neurons, guided there by the signaling molecules. Eventually the different bunches of neurons get connected and transmit neural signal from one to another, creating a mechanism for what we would describe in mathematics as a logical implication (a->b). The dog now expects food when the bell rings. Their brain created a simple, specific mental representation of parts of the world around them and how they relate. Many such connections create chains of thoughts, and human brains in particular are pretty good at devoting most of our neural signaling to bouncing signals back and forth inside that network of chains of thoughts in repeating patterns, creating elaborate mental representations that are very general. Our ability to create general mental representations about the world around us is behind our evolutionary success.
These elaborate general mental representations are the definition of 'the Explanation' that I will now use below to answer OP's question.
Good explanations make correct predictions. Great explanations make correct predictions at a very general level.
If an explanation makes incorrect predictions, it's not a very good explanation. Clinging onto such explanation is what we would call a mental disorder.
If the explanation cannot make predictions, then it's unfalsifiable, and thus worthless. Clinging onto such explanation is what we would call a cult or a religion.
Now, can you have a prediction and not have an explanation? There are many phenomena that we repeatedly observe but can't explain yet in a way that is coherent with our other general explanations about the world. That's usually the starting point for when the brain searches for new, general explanations.
jimh12345 | 21 hours ago
I like it, personally. IMHO our goal is to create meaning, not just predict the weather.
Zestyclose-Bag8790 | 17 hours ago
I am a doctor so I approach this with a strong bias in favor of prediction.
If I have a drug that cures your cancer, what matters is that it does that. Take the drug, cure the cancer.
How the drug destroys cancer is of little importance to me. Does it strangle the cancer? Does it poke tiny holes in it? Does it work by altering your metabolism or perhaps your chakra?
Why does it work? If I can give you a drug that works 70% of the time and I can fully explain exactly how it works, or I have a drug that works 95% of the time and no one has any idea how it works, but the evidence that it does work 95% of the time is strong, I will never use the 70% drug again.
In medicine people who want to tell you how a drug works are almost always trying to sell you the drug. They work for the drug company and they love to explain how their drug works. They are generally attractive people. They love talking about the mechanism of action.
I want to see the clinical evidence it works. Over time I have become frustrated with the explainers.
Freuds-Mother | a day ago
The way these are posed as almost binary choices I’d reject.
How does an infant, toddler, or child interact with the world? They do all of experiment, predict, and attempt to explain.
Scientists also try to predict, experiment, and necessarily implicitly presuppose ontological constraints. Behaviorists/positivists can’t avoid explanation inclusive of implied ontological presuppositions no matter how hard they try.
On the other hand, explanations that provide zero predictive value easily wind up in arbitrary language games.
For a quick read on behaviorism, operationalism, and positivism see https://www.lehigh.edu/~mhb0/Operationalism.pdf Piaget is also a great reference here.
knockingatthegate | a day ago
His position is more self-aggrandizing than substantive, as if he alone or he predominately is the savior of the scientific method from the barren waste-land of mere prediction.
The major arguments against prioritizing explanation over prediction include unintelligibility, being that “explanation” is ill-defined; narrative superabundance, being that explanatory adequacy can be obtained with numerous incompatible explanatory framings; and epistemic peril, insofar as the criterion for sufficiency for explanation is psychological whereas for prediction it is empirical.
fox-mcleod | a day ago
> The major arguments against prioritizing explanation over prediction include unintelligibility, being that “explanation” is ill-defined;
An explanation is a conjecture about something unobserved which attempts to account for what is observed.
The key differences between a good explanation and a mere model are:
For example, when talking about the observation that the Earth has cyclical seasons, a mirror model of the Seasons would be something akin to a Calendar, whereas an explanation of those seasons would be entirely different such as the axial tilt theory.
The model is easy to vary. If it turns out that year is a month longer, we can simply add a month to the model. If we discovered that the southern hemisphere has opposite seasons to the north, we can simply create a southern hemisphere model to go along with the northern one.
But the axial tilt theory predicts that the southern hemisphere has opposite seasons. If instead we had found that it did not, we wouldn’t be able to modify the axial tilt theory without ruining the explanation entirely. We would have to throw the entire explanation out. And that’s a good thing because scientific theories are measured in terms of what they would rule out if they were to be falsified. If you could easily modify them, they would rule anything out except for the exact results of that exact experiment.
Second, the calendar tells us absolutely nothing about what we would need to do to change the seasons. Models are not counter factual claims. Whereas explanations are. An explanation for the earth’s seasons, such as the axial tilt theory explains that if we want different seasons, we can change the variables in the explanation such as the shape of the Earth, the angle of tilt its relationship to the sun, etc.
It even tells us about the seasons on planets we’ve never been to — or which don’t exist.
> narrative superabundance, being that explanatory adequacy can be obtained with numerous incompatible explanatory framings;
Yes that’s the idea of science. You have multiple competing conjectures and then you do experiments to differentiate them. When all available explanations work, you look to parsimony to understand which explanation is better. “Good” meaning what I said earlier: hard to vary. We can actually mathematically prove this principle of parsimony through the technique of Solomonoff induction.
knockingatthegate | a day ago
I’m not sure what to make of your reply. I was addressing OP, whose study of the subject is in early days. For that reason, a conversational tone was warranted.
I’ll offer that you’re overshooting the mark in your reply to me, if not shooting off target entirely.
CWW2022 | a day ago
Explanation is the result of attempting to reconcile predictions with observed reality. It’s downstream of prediction, and distinct.
jadedscum | a day ago
If that is so, then it'd be more open for non-mainstream works, opinions, thoughts and experienced to be also considered, however unfortunately the social grooming into certain positions as well as the ego and its often to be observed inability to accept things further outside of it's own narrative and experience restrict the pursuit of understanding and explanation of things scientifically thus.
Keikira | 21 hours ago
Realism very wide-spread among practically every sector -- philosophers of science, scientists, and laypeople alike -- but privileging explanation over empirical power is not something I read or even hear often because it boils down to privileging metaphysics over the scientific method. You can't falsify explanations, only predictions.
I'm an instrumentalist -- I think scientific realism is a historical reflection of religion, that it misunderstands the human condition, and most importantly, that it being so widespread is something that makes people worse at science -- so I have no sympathy whatsoever for even more moderate realist positions; however, most realists at least engage with the problems that human epistemic limitations pose. Unlike me and mine, they just believe those limitations are inconsequential or surmountable.
Leaning back into metaphysics is an outright regression, though. Having scientists even more attached to the unfalsifiable components of their particular theories, more unwilling to change their minds when faced with counter-evidence, is not a way forward.
randcraw | 21 hours ago
Consider this, if a theory could predict but not understand, would it serve the interests of science? IMO, no. You cannot build a deeper understanding of a system if you do not understand how all its essential causal mechanisms act and interact. So understanding is essential to science. (BTW, AI is often capable of predicting a outcome but often cannot explain the mechanisms that caused it. That's just good pattern matching, not science.)
What if a theory understands but can't predict? Is that even possible? IMO, no. If a theory can explain the mechanistic cause of an outcome, it necessarily can also explain the consequences of activating that mechanism. So prediction is the necessary consequence of understanding.
DrStrangelove0000 | 20 hours ago
Mathematician here. I've been thinking about something equivalent: "inside our heads" and "outside our heads." Roughly models are inside our heads and measurements are outside.
Realists, transcendentalists, positivist, etc. just differ in their ratios of inside / outside.
Positivism, unlike other answers state here, was critical to modern science. Foundational in the late 1700s and 1800s. Check out Britannica's article, it's great. Electromagnetism, thermodynamics, etc. were sciences where we discovered piecemeal laws first before sticking them together into nice general differential equations. So you kind of need some kind of positivism to even start understanding pressure / volume for example.
Or maybe not. Remember that both of these sciences had very immediate applied interests around them (electrical grid and transportation), and positivism is the science of industry. Americans still follow this philosophy in general, which it's no accident we're leaders in AI at the moment, maybe the most positivistic version of knowledge ever implemented at scale. Who knows how far it will go?
An equivalent way to state inside / outside, or explanation / prediction, or model / measurement, or positivist / realist is prior / posterior from Bayesian statistics. In the end, what is the difference exactly? The more I think about it the less clear it becomes.
Prior is what we think before the experiment, it's our "bias" and the posterior is what we've "learned" after we measure. But of course, this assumes a stable reality between prior and posterior, which quantum mechanics and the uncertainty principle threatens. But even in all the other branches of science you can see this problem, we just roll that distrubance from measurement into "noise". Measuring even the temperature still cools the object, ever so slightly, for example.
In the end, I think we're still exploring the boundaries of positivism / realism. If for example, we finally grow or build a replicate of the human brain, and could map it's wiring, would that produce a positivist or realist theory of cognition? It's immediately filled with contradictions and confusions if you try to think about it.
I think the positivist / realist distinction is best thought more concretely as a continuum of model dimensionality vs predictive capability. You want low dimensional models that produce wide ranging (in space / time) predictions. In other words, lots of equations with few emperical physical constants. AI models for this reason are not immediately useful for "foundational science." But minimizing the dimensionality / predictive power ratio is also itself just partially because it's easier to think about simpler theories than complex ones. So we're back at a kind of positivism again. (There is a nice discussion between Peter Norvig and Noam Chomsky about this).
Great question! I'll be thinking about it for awhile!
PressureBeautiful515 | 17 hours ago
Deutsch is a hardcore Everettian and I think has sometimes implied that his own insights (that were foundational to quantum computing) were only possible because of his preference for the multiverse interpretation of QM rather than collapse. But (it goes without saying) that all interpretations of QM are functionally equivalent and all of quantum computing theory can be accessed with the tools of QM regardless of what interpretation you happen to prefer. Or as a rational person might say, there is not real reason to choose one interpretation over another.
So he is most likely defending the fact that he has a preference, by attributing greater explanatory power to it - a quality which cannot be objectively measured, and he can freely assert it without anyone else being able to tell him he's wrong. Which is very much nothing at all to do with science.
baydew | 14 hours ago
I havent read this but ive seen scientists in psychology and stats draw a distinction between prediction and explanation to think about different kinds of questions we can ask, and i think this is a useful distinction.
this can be a correlation/causation point. We may observe X and Y are related, but that doesnt mean X directly causes Y, or that we can increase Y by changing X. We could say “what factors are asdociated with Y?” Is a prediction question while “what factors affect Y?” is an explanation question.
In stats, This also sounds to be like the prediction/inference distinction that comes up with regression models & machine learning. Sometimes you want a model that takes a bunch of data and does a good job predicting the future. Especially if youre like a business. But in some techniques, models are black boxes and it is difficult to parse to understand how specific factors relate to future outcomes. And people might be more interested in analyzing/doing inference with coeffiecients (individual factors or interactions) to parse out what is driving the prefictio. Sometimes using simpler models help to parse out specific factors (explaining), even with less predictive power
baydew | 14 hours ago
Also in my view experimentation is fundamentally a method for explanation, so im a bit lost when you contrast “prediction and experimentation” vs explanation
Edit: as others say, for a philosophical discussion we would need a formal definition of these distinctions
Oposasa | 9 hours ago
Science doesn’t have a primary goal. Science is a method. You could apply the method of Science towards any goal.
[OP] Unlucky-Prior-1838 | 8 hours ago
Good point, although I mean then what does Deutsch actually argue for? It seems to me that via treating science as a method, explanation and prediction are in some way steps of the method? I'm quite confused here.
knockingatthegate | 6 hours ago
He’s arguing for book sales.
After_Network_6401 | 7 hours ago
I think it’s a confusion of terms. Making predictions and testing them is a core part of the scientific method. It’s a technique. An explanation is the goal. If a prediction is falsified, that just means the proposed explanation is either partially or wholly incorrect, but it doesn’t remove the need for an explanation.
I can give a real life example. Years ago, a colleague shared her patients’ clinical data with me, to discuss possible treatments strategies. I noticed that the data was not normally distributed.
Why was this? This observation demanded an explanation. To explain it we started with various predictions. The first was that it was an artefact of data. So we checked patients’ ages, other illnesses, family history, etc. Nothing there. So the next prediction was that these patients would have different clinical outcomes, which turned out to be the case.
Then how that happened, needed an explanation …
CyberpunkAesthetics | 6 hours ago
Scientific method is about prediction.
The psychology that drives science, involves the desire for explanation.
Without the desire for explanation, we would not attempt science.
SmallCap3544 | 5 hours ago
As others have pointed out, Deutsch works mainly in quantum theory, which is “special” when it comes to science.
I have not read this work, but he is a major proponent of the many worlds interpretation of QM and Quantum computing.
The multiverse as an idea is not falsifiable, and this is a core weakness underlying any explanation that relies on the multiverse. When you understand that, his point of view makes sense.
MattAtPublicThink | an hour ago
Does the why matter if we can predict the outcome? Maybe. Maybe not. Depends on the circumstance.
Cybtroll | a day ago
It depends on the specific current of philosophy of science, but overall if you are a bayesianist Deutsch assumptions are somehow inevitable.