A perpetual difficulty in my studies is to pinpoint precisely what I’m even trying to do.
That might sounds trivial, but as noted in a recent post, if I actually listen to what excites me and interests me and motivates me, I quickly realize that no existing concept fits the weird result, and I find myself in need of engineering a handle for it.
Let’s try something different.
I’m still working through designing this field that I’m dreaming about. Yet I also have a handful of books that are close to it in some way — they capture part of the intuition, part of the goal and approach.
So I’ll triangulate this fabled field through reviewing these books and following what resonates and what doesn’t.
There are two answers, one of which fits with what I care about, and the others which doesn’t.
First, the book is about clarifying and systematizing the method of black boxing, and its place in science. The claim is that many scientists (the main examples are Darwin, Mendel, Skinner, and Friedman[1]) use black boxing (hiding the detail of some complex mechanism of phenomena) for various productive aims, with more or less success. And Nathan wants to dig into this, and try to bring some order to this complexity.
This, I wholeheartedly agree with. What I find fascinating in science and beyond are the methods themselves, in their subtlety and complexity, and I want to systematize and explain them, why they work (or don’t), how they fit with each other, what’s the underlying structure.
And that, I honestly have no interest in. Don’t misunderstand me: I do agree that the correct approach to most philosophical debate is to dissolve them by showing how they only emerge from wrong assumptions and false dichotomies.[2] Yet this is not something I want to dig in and clarify myself. From my point of view, philosophical debates are almost always void of any relevance to understanding methods and their workings.
Case Studies: Variety and System
Turning to the case studies (which are the beating heart of any good HPS book), the pleasant surprise is their range: as mentioned above, Nathan uses examples from biology (Darwin and Mendel), psychology (Skinner), and Economics (Friedman), which is a nice change from the traditional over-reliance on physics. The broader the range of example, the more likely that the systematization will generalize and capture the core regularity.
His examples are also particularly valuable for two reasons:
They use black boxing for different goals: Darwin and Mendel black box some mechanism (heredity with variation and genes respectively) which they can’t reverse engineer, Skinner removes from consideration any mental states, and Friedman makes concepts like utility “opaque”, as a way to remove them from the reach of other fields like psychology.
They differ in the success of black boxing: Darwin and Mendel succeed, Skinner fails, and Friedman’s success is still debated.
Notably, this is essential for understanding what are the underlying conditions (epistemic regularities) which the method exploits, and requires for its success.
All of this I agree and condone, and this definitely points to core principles behind finding case studies for methodology.
Yet I feel like Nathan could have gone further.
First, he limited himself exclusively to science. I know that this is a book of HPS (History and Philosophy of Science), but I’m becoming more adamant by the day that restricting oneself to science is a mistake. Just on black boxing as interfacing, there are clear and exciting examples in software engineering (where the concept of abstraction is essential), but also simply in cooking (where you don’t need to model the ingredients themselves, only some of their core macro properties).
Similarly, I feel like there is a missing opportunity in never mentioning a recurrent aim of black boxing in sciences (especially Physics and Chemistry): precomputing a higher-level, simplified model from a complex existing one. I’ve already described this in my post on compression moves, but for example, molecular orbitals voluntarily abstract away from the details of the Schrodinger equation (which can actually be solved for very simple setups), throwing away just enough details to get a tractable combinatorial framework for how molecules react with each other.
Last but not least, there is a weird lack of control for the different goals and degrees of success of the method. That is, instead of finding examples for each goal where black boxing succeeds, fails, and is uncertain, the goal and success are tied in ways that probably confound the analysis: black boxing as a way to interface with a mechanism without having to understanding it just yet (Darwin and Mendel) is the only example of success; black boxing as a way to remove some factors from the causal explanation at all (Skinner) is the only example of clear failure; and black boxing as a way to keep a field’s concepts safe from other fields (Friedman).
Content: Interpretation and Dissertation
In terms of the actual content of the book, I’m much less convinced.
Nathan does go into decent detail about the context and content of his case studies. But then, almost all discussion of his solution, of his framing of black boxes, falter for the same reason: they are far too focused on discussing philosophy of science instead of actual methods of science.
Not only does this mean Nathan loses a lot of space integrating his view with the existing ones in the philosophical discourse (a waste of time), but even his analysis suffers from this.
Take his model of black boxes: fundamentally, what he argues is that first you characterize (mostly implicitly/contextually) what you want to explain, making it into a phenomena; then you decide which reasonable causes are considered relevant in your model (some actual causes could be neglected if they are considered obvious background assumptions); and then figure out how much detail you need to represent these relevant causes.
Instead of the above, this is what most of the book contains:
The first step, the framing stage, involves sharpening the object of explanation. Specifying the explanandum in detail presupposes a full-fledged model, which is not easy to obtain, especially in actual scientific practice. Fortunately, to get the inquiry going, it is sufficient to employ a frame, that is, a coarse-grained placeholder that stands in for patterns of behaviors in need of explanation which, in principle, could be described at a finer scale. The second step, the difference-making stage, provides a causal explanation of the target by specifying which features of the explanans make a difference to the occurrence of the explanandum. While there are various strategies for doing so, I borrowed Strevens’s effective kairetic approach. Many of these difference-makers may be left unpacked. In some cases, the micro-structure is omitted because of mere convenience, to draw the boundaries of a field, or to insulate a concept from empirical refutation. Other times, the decision is dictated by ignorance, as these details are actually unknown. The third and final step, the representation stage, determines how these difference-makers should be portrayed, that is, how much abstraction and idealization will produce the optimal explanatory “bang for the buck,” given how the explanandum was framed. This is done by constructing a model, an interpreted structure that represents bits and pieces of the world based on a relation of similarity.
And that’s in the conclusion of a chapter, when the author is trying to compress and distill his model!
This philosophical verbiage is at the expanse of digging more into what actually matters in the interpretation of methods: how they work, and what is required to make them work.
On both accounts, Nathan stay far too abstract: his model of black boxes is reasonable, but it hardly describes or explains the detailed moves that say Darwin takes when black boxing, or why these moves are good, whereas Skinner’s corresponding moves are faulty.
The best we get is this kind of reflection:
At the same time, Skinner’s construction of black boxes dismisses the relevance of mental states in the production of behavior. The result — evident in all three phases of his construction: framing, difference-making, and representation — entails that his radical behaviorism leaves out an important aspect of psychology: mentality. This eventually caused his general model to collapse. This, in a nutshell, was Skinner’s capital sin.
And instead, Nathan takes at least 4 chapters after that to discuss various philosophical debates, from emergence to progress and reductionism.
That being said, I must admit that most of his philosophical positions are reasonable. I don’t think it’s useful or valuable, but at least most of the focus is on dissolving false dichotomies and on redirecting energy to epistemology and methodology rather than the ontological reality of various abstract concepts and hidden entities.
As we shall see, the black-boxing recipe fits in quite well with the idea of science being in the business of discovering and modeling mechanisms. At the same time, the construction of black boxes, as I present it here, mitigates many of the ontological implications that characterize the contemporary landscape. For this reason, I provocatively refer to black-boxing as a “diet mechanistic philosophy” with all the epistemic flavor of traditional views, but hardly any metaphysical “calories.”
In conclusion, I stand with Nathan on the focus on methods, on the variety of case studies (although I want even more variety, going outside of science), and on caring about compressing and modeling the various shapes and forms of methods into one encompassing abstraction.
Where I disagree is that I see almost no value in thinking through the empty debates of philosophy of science, and I want more systematization in the case studies, and more predictive modelling of how the methods unfold, and where they will work or fail.
Technically, Friedman is used as a stand-in/focal point for most of the core methodological assumptions of neoclassical economics. It does seem like Friedman wrote explicit papers about these methodological choices, but the exact intuitions and details might be less specific to him than say Darwin and Mendel’s insights. (Maybe they were, I don’t know that much history of economics)
Triangulating My Interpretation of Methods: Black Boxes by Marco J. Nathan
Link post
A perpetual difficulty in my studies is to pinpoint precisely what I’m even trying to do.
That might sounds trivial, but as noted in a recent post, if I actually listen to what excites me and interests me and motivates me, I quickly realize that no existing concept fits the weird result, and I find myself in need of engineering a handle for it.
Let’s try something different.
I’m still working through designing this field that I’m dreaming about. Yet I also have a handful of books that are close to it in some way — they capture part of the intuition, part of the goal and approach.
So I’ll triangulate this fabled field through reviewing these books and following what resonates and what doesn’t.
Today, we start with Black Boxes by Marco J. Nathan.
Two Aims: Methodology and Philosophy
What is this book even about?
There are two answers, one of which fits with what I care about, and the others which doesn’t.
First, the book is about clarifying and systematizing the method of black boxing, and its place in science. The claim is that many scientists (the main examples are Darwin, Mendel, Skinner, and Friedman[1]) use black boxing (hiding the detail of some complex mechanism of phenomena) for various productive aims, with more or less success. And Nathan wants to dig into this, and try to bring some order to this complexity.
This, I wholeheartedly agree with. What I find fascinating in science and beyond are the methods themselves, in their subtlety and complexity, and I want to systematize and explain them, why they work (or don’t), how they fit with each other, what’s the underlying structure.
The other stated goal of this book is to resolve the conflict between reductionism and antireductionism, mostly by dissolving it.
And that, I honestly have no interest in. Don’t misunderstand me: I do agree that the correct approach to most philosophical debate is to dissolve them by showing how they only emerge from wrong assumptions and false dichotomies.[2] Yet this is not something I want to dig in and clarify myself. From my point of view, philosophical debates are almost always void of any relevance to understanding methods and their workings.
Case Studies: Variety and System
Turning to the case studies (which are the beating heart of any good HPS book), the pleasant surprise is their range: as mentioned above, Nathan uses examples from biology (Darwin and Mendel), psychology (Skinner), and Economics (Friedman), which is a nice change from the traditional over-reliance on physics. The broader the range of example, the more likely that the systematization will generalize and capture the core regularity.
His examples are also particularly valuable for two reasons:
They use black boxing for different goals: Darwin and Mendel black box some mechanism (heredity with variation and genes respectively) which they can’t reverse engineer, Skinner removes from consideration any mental states, and Friedman makes concepts like utility “opaque”, as a way to remove them from the reach of other fields like psychology.
They differ in the success of black boxing: Darwin and Mendel succeed, Skinner fails, and Friedman’s success is still debated.
Notably, this is essential for understanding what are the underlying conditions (epistemic regularities) which the method exploits, and requires for its success.
All of this I agree and condone, and this definitely points to core principles behind finding case studies for methodology.
Yet I feel like Nathan could have gone further.
First, he limited himself exclusively to science. I know that this is a book of HPS (History and Philosophy of Science), but I’m becoming more adamant by the day that restricting oneself to science is a mistake. Just on black boxing as interfacing, there are clear and exciting examples in software engineering (where the concept of abstraction is essential), but also simply in cooking (where you don’t need to model the ingredients themselves, only some of their core macro properties).
Similarly, I feel like there is a missing opportunity in never mentioning a recurrent aim of black boxing in sciences (especially Physics and Chemistry): precomputing a higher-level, simplified model from a complex existing one. I’ve already described this in my post on compression moves, but for example, molecular orbitals voluntarily abstract away from the details of the Schrodinger equation (which can actually be solved for very simple setups), throwing away just enough details to get a tractable combinatorial framework for how molecules react with each other.
Last but not least, there is a weird lack of control for the different goals and degrees of success of the method. That is, instead of finding examples for each goal where black boxing succeeds, fails, and is uncertain, the goal and success are tied in ways that probably confound the analysis: black boxing as a way to interface with a mechanism without having to understanding it just yet (Darwin and Mendel) is the only example of success; black boxing as a way to remove some factors from the causal explanation at all (Skinner) is the only example of clear failure; and black boxing as a way to keep a field’s concepts safe from other fields (Friedman).
Content: Interpretation and Dissertation
In terms of the actual content of the book, I’m much less convinced.
Nathan does go into decent detail about the context and content of his case studies. But then, almost all discussion of his solution, of his framing of black boxes, falter for the same reason: they are far too focused on discussing philosophy of science instead of actual methods of science.
Not only does this mean Nathan loses a lot of space integrating his view with the existing ones in the philosophical discourse (a waste of time), but even his analysis suffers from this.
Take his model of black boxes: fundamentally, what he argues is that first you characterize (mostly implicitly/contextually) what you want to explain, making it into a phenomena; then you decide which reasonable causes are considered relevant in your model (some actual causes could be neglected if they are considered obvious background assumptions); and then figure out how much detail you need to represent these relevant causes.
Instead of the above, this is what most of the book contains:
- Marco J. Nathan, Black Boxes, p.132-133, 2021
And that’s in the conclusion of a chapter, when the author is trying to compress and distill his model!
This philosophical verbiage is at the expanse of digging more into what actually matters in the interpretation of methods: how they work, and what is required to make them work.
On both accounts, Nathan stay far too abstract: his model of black boxes is reasonable, but it hardly describes or explains the detailed moves that say Darwin takes when black boxing, or why these moves are good, whereas Skinner’s corresponding moves are faulty.
The best we get is this kind of reflection:
- Marco J. Nathan, Black Boxes, p.155, 2021
And instead, Nathan takes at least 4 chapters after that to discuss various philosophical debates, from emergence to progress and reductionism.
That being said, I must admit that most of his philosophical positions are reasonable. I don’t think it’s useful or valuable, but at least most of the focus is on dissolving false dichotomies and on redirecting energy to epistemology and methodology rather than the ontological reality of various abstract concepts and hidden entities.
- Marco J. Nathan, Black Boxes, p.163, 2021
Conclusion: Removing The Philosophy
In conclusion, I stand with Nathan on the focus on methods, on the variety of case studies (although I want even more variety, going outside of science), and on caring about compressing and modeling the various shapes and forms of methods into one encompassing abstraction.
Where I disagree is that I see almost no value in thinking through the empty debates of philosophy of science, and I want more systematization in the case studies, and more predictive modelling of how the methods unfold, and where they will work or fail.
Technically, Friedman is used as a stand-in/focal point for most of the core methodological assumptions of neoclassical economics. It does seem like Friedman wrote explicit papers about these methodological choices, but the exact intuitions and details might be less specific to him than say Darwin and Mendel’s insights. (Maybe they were, I don’t know that much history of economics)
A favorite personal example is Hasok Chang’s Realism For Realistic People, which I consider the definitive world on the empty realism debate.