Science and philosophy are not as different as they might seem. Whereas the first evokes ideas of mad geniuses and crazy experiments, and the second evokes ideas of overthinkers and tired cultural debates, these caricatures conceal a striking similarity between the modern practice of both.
Both scientists and philosophers have a common goal of coming up with a good explanation of the things they study. That is, they want to come up with a good theory. Likewise, both modern scientific and philosophical practice has moved beyond the paradigm of divine providence or rhetorical flourish as their north star. Even if speculation is still part and parcel of their respective inquiries (just ask the string theorists), scientists are constrained by empirical evidence, and philosophers are confronted by often devastating thought experiments for which they can no longer call upon a deity to justify.
Yet, although science and philosophy are both absolutely rigorous academic disciplines, the latter has gotten a bit of a bad rap. It doesn’t help that most people have vague notions of what philosophers really do, since most people’s exposure to philosophy consists of unhelpful platitudes by long-passed spiritual thinkers about how to live an ascetic life. (Or maybe they can be helpful. Run a thought experiment, or even a real experiment, to find out!)
However, that is unfortunate, for everyone, even the ordinary Joe, has engaged in some sort of philosophical inquiry. It’s an unavoidable part of life to ask questions about the most fundamental things like existence, knowledge, and morality. And although they can feel like (and still are to an extent) questions that lead to dead ends, modern philosophy is actually more conclusive than we might think, thanks to many of the same methods that have turned science into a verifiable pursuit.
Different questions, but the same method
To distill the method that both modern science and philosophy uses, I like to carve out a distinction between two modes of inquiry that I like to call thinking backwards and thinking forwards.
In both fields, you always have a bunch of data you want to explain. In science, that data is usually empirical evidence that you gather from running experiments in the physical world. In philosophy, although empirical evidence can play a role, often that data consists of thought experiments in a hypothetical possible world that may differ from the physical world in some key aspects.
Then, what I call thinking backwards is the process of taking the data and distilling it into a theory, a system that explains the data and allows you to deduce new predictions. Fundamentally, all theories consist of concepts and rules for reasoning with those concepts. For example, Newton distilled many different observations about physical motion into a theory where a couple simple concepts (such as mass and force) and mathematical rules (laws of motion) could explain those observations very damn well. Thinking backwards involves abductive reasoning, where the goal is to find the best explanation for a phenomenon. Of course, an explanation should at least be plausible, but you might favor an explanation for its other virtues as well, including practicality and simplicity.
Now, suppose you have a theory. The fortunate thing is that a theory gives you a structured system to reason about the phenomenon it explains. The unfortunate thing is that theories are almost always wrong in some way. A ‘best’ explanation is almost never a perfect explanation, but simply the best we could achieve with the data that we have. While there is often room to improve our theories even with the lack of new data, eventually it will prove to be a manacle on progress.
But how do we collect data in a way that helps us make our theories better? Much of the answer lies in actually using our theories to make new predictions, to find new directions for which our theory might be tenuous or even wrong, and that is the heart of what I call thinking forwards. For example, Einstein identified issues in Newton’s theory by first applying his theory, then noticing that it contradicted the nascent body of results from electromagnetism research. These issues do not strip the virtues in Newton’s theory of being plausible enough to serve most ordinary physical problems and of being simple enough for a bored patent office employee to casually use, but they do call for another process of thinking backwards: finding an updated theory which does take the new data into account, and that theory was special relativity. Regardless, thinking forwards involves hypothetico-deductive reasoning, where the goal is to use a theory to devise testable hypotheses to find issues with the theory. You don’t always have to run an experiment to test a hypothesis, as often just devising a hypothesis alone can point us towards data we haven’t considered, but often you do have no other choice but to run an experiment.
Thinking backwards and thinking forwards operate in a cycle. Thinking backwards from data yields us a theory, and thinking forwards from a theory yields us new (or overlooked) data which we can use for a better theory. Yet, while easy to see in science, it is not immediately apparent that philosophy rides this same cycle, so let me present a philosophical example.
Creating a theory of knowledge seems like a simple task. After all, surely we all know what ‘knowing’ is? And in philosophy, it is more common than you might think to call upon our intuitive notions of a concept when creating a first theory of it, because those notions are still data which should be taken seriously lest we say anything downright bizarre.
One theory of knowledge which passes that ‘not bizarre’ test is to define knowledge as a true belief. It sounds reasonable to say that if we believe in something that is true, we know it too, and that either a lack of truth or a lack of belief means that we do not know it. So far so good. Yet, after thinking backwards and minting a fresh new theory, the next step is not necessarily resting on our laurels but to then think forwards and subject the theory to scrutiny.
While in science thinking forwards involves making predictions to test in the physical world, in philosophy thinking forwards usually involves making predictions to test in the hypothetical world that is a thought experiment. As for the true belief theory, there is a surprisingly simple thought experiment which reveals a major problem with it.
Suppose you want to flip a coin. You believe it is going to land heads. Then, you flip it, and it lands on heads. You had a belief that the coin was going to land on heads, and that belief is true because it did land heads, so under the true belief theory, you knew that it was going to land on heads. But that is quite a stretch! Indeed, if you then tell someone that you ‘knew the coin was going to land on heads’, they would think you were either kidding (‘ha I KNEW it was going to land on heads’) or making too much of a fuss about your clairvoyant abilities.
This thought experiment gives you a new piece of data: you can believe in something that is true, and still it might be better explained by plain old luck. Given this data, you go back to the drawing board, thinking backwards to revise that theory, and rinse and repeat. In this case, some might theorize that knowledge also requires the belief to be justified as well, and indeed justification might be weak in this case since you have no better reason to believe that the coin will land heads than tails. But thinking forwards again, the justified true belief theory runs into a famous paradox called the Gettier problem, whose implications are still hotly debated today. There are other theories too, including the theory that knowledge and belief ought to be distinctive concepts since the latter is a hotbed of pernicious paradoxes.
“It sounds wrong”
Many philosophical issues are identified by identifying a deduction in a theory that just, well, plain sounds wrong. For anyone used to the neat and tidy statistics of scientific experiments, this might seem horrifying.
However, consider what both fields are trying to explain. Science constrains itself to what can be demonstrated by empirical evidence, so the data it is trying to explain are the records of events happening in the real world. The goal of a scientific theory has an analogous form: to predict the events that are going to happen in the real world.
But philosophy has a wider ambition. While philosophy can use empirical evidence, a large part of philosophy is explaining why we have the intuitions we do, so its data often includes intuitive notions. The goal of many philosophical theories hence, again, has an analogous form: to predict the intuitive notions we would have to hold given a set of assumptions. Now, unlike in science where empirical evidence is empirical evidence, in philosophy we can, of course, challenge those intuitions directly as we think forwards towards a better picture of what they entail. Still, philosophy embraces intuition as one of its evidentiary tools, for it is partly about our intuition itself, just like how science is about empirical evidence.
Nowhere is this more apparent than in the study of ethics, where elaborate theories are built upon our quite primal assessments of ‘right’ and ‘wrong’. Ethical dilemmas, even concretely specified ones, can be notoriously hard to assess because they might involve conflicts between multiple ethical considerations. For example, the famous trolley problem of choosing whether to sacrifice one to save five pits two revulsions against each other: our revulsion against unnecessary losses and our revulsion against singling out others to be harmed. To even reason about this and about similar problems, we would have to think backwards and devise a theory with which we can reason about these two revulsions accordingly. However, once we have a theory, we have an instrument for reasoning in explicit terms about moral dilemmas which we could earlier only assess at an implicit level. And by thinking forwards with a theory, we can test our ethical assumptions on new, perhaps less controversial cases, revealing where those assumptions really take us. Regardless, (orthodox) ethics research has to be constrained at least somewhat by ethical intuition. While it is fruitful to point out the myriad of contradictions and complications in folk ethical thought, the moment that an ethical theory endorses something that is so obviously, barbarously wrong is when most consumers of ethics would throw it in the bin.
A controversial example
One of the most politically inflammatory topics also turns out to be one of the most fruitful cases of why it is valuable to keep the distinction between thinking backwards and thinking forwards.
Many social categories were, more often than today, tied to basic observable properties of an individual. Okay, because I’m treading a political minefield, sigh… I’ll clarify that by ‘basic’ I am disavowing the rhetorical functions of that word and by ‘observable’ I mean in terms of phenotype. Of course, if you look at many past societies and even some current societies, a combination of cultural practice and pressure (negative feedback mechanisms) certainly yields you many behavioral observations where reasoning with such rigid categories helps you explain a lot. And I think the key here is to separate the theory that such rigid categories explain and can predict these observations, and the theory of “I simply don’t like individuals within this category” which is so often smuggled in.
It’s not just a cultural thing: generally, whether it’s about people or machines, when you have sufficiently strong negative feedback mechanisms in a system, you can approximately capture their behavior by looking at their attractors (a form of coarse-graining).
However, when you think forwards from the ‘rigid category theory’ (for the lack of a better name), you quickly encounter its biggest limitation: the many, many, many instances where an individual with observable properties belonging to a specific category simply… doesn’t perform the actions associated with that category. Now, as I had mentioned, it applies more to some cultures than others. A culture which rigidly enforces the actions associated with a category would (if the negative feedback mechanisms in the culture function) not produce nearly as many such instances. But it is a serious enough problem to cast doubt on the rigid category theory.
So that’s when you think backwards again and update the theory so it takes these cases into account. One modification is to explicitly model these cases as deviations from the behaviors that the rigid categories would predict. Another modification is to treat observable properties as only one property in a collection of other properties which altogether generate behavior. Yet another modification is to treat these properties, to some extent, as pragmatic social performances which may be advantageous in certain contexts. Without adjudicating their merits (and keeping in mind that they are not mutually exclusive), each modification can account for a different body of observations. For example, the first can account for individuals explicitly disavowing a culturally expected behavior, the second can account for the many studies that demonstrate a combination of variables to be more predictive of behavior than a single one, and the third can account for ‘campy’ performances of entrenched meanings associated to a basic observable property that nonetheless affect interactions with others.
And perhaps, if I am going to make any huge controversial claim here, that when put this way, all three modifications are by themselves less controversial than they are portrayed. Just because a theory explains an observation does not mean it necessarily judges its goodness or badness. There are clear-cut examples of individuals disavowing a culturally ‘expected’ behavior, of many factors resulting in a better fit model (albeit relative to standard statistical measures) than one, and of individuals acting in a campy manner to secure some sort of reward. Each of these modifications has its own tired political debate, and I’d at least advocate for a more careful application of ethical theories rather than quick soundbites if we are going to settle those (and that goes for both sides!), but that’s just the baggage each has to trolley around.
Not to mention that while a grand unified theory of social categories could plausibly incorporate all three of these modifications (and more) in some form, we might also go for a simpler theory when it comes to actually using it. Just because we use a simpler theory doesn’t automatically mean it is an endorsement of one ethical position over another, unless the theory explicitly elides some ethical premises or we decide to smuggle in a schema for translating descriptive premises to ethical ones.
For example, when studying a society like Samoa where certain deviations from common social categories are legally recognized, we might first look to reason principally with a deviation-based model when discussing those deviations. Whether such a model predicts behavior depends principally on the form or efficacy of the negative feedback mechanisms in Samoan society related to matters like correlations between properties (e.g., whether social categories act in a sui generis manner, or whether a social category’s expected role may be exempted by certain other properties) and performance, on which I am no expert. Regardless, we can take all these considerations into account without ever endorsing or opposing either deviations, correlations between properties, or performances.
We can also see that an advantage of carefully distinguishing the theory from the data is that it reduces the confusion as to the specific claims a theory makes. Here, we are treating these social categories as predictors of behavior, and nothing else. We could have another theory where these social categories are taken to be literally naming the observable properties to which they were historically tied, so that for example (tip-toeing my way here) a word like “tall” literally refers to height. That theory would certainly be good for, well, naming those properties, but the question of whether that theory can be extended to social or ethical prediction is a lot more controversial.
Still, the next time you see a tired political flame war – and this is an actual ‘critical thinking’ tip that isn’t just vibes – try to see if the speaker has snuck in latent premises that allow descriptive statements (like “so-and-so is of this social category [due to these phenotypical features]”) to be transferred into ethical ones (the speaker has a fetish or vendetta regarding this category = “therefore they are good/bad”).
