Conceptual Analysis is Dead - Long Live Conceptual Engineering!

Defining our problems away!

A conceptual engineer finishing his concept, just as the language police arrives!

[Philosophers] have always trusted concepts as unconditionally as they have mistrusted the senses: it never seems to have occurred to them that notions and words are our inheritance of past ages in which thinking was neither very clear nor very exact. What seems to dawn upon philosophers last of all: that they must no longer allow themselves to be presented with concepts already conceived, nor must they merely purify and polish up those concepts; but they must first make them, create them, themselves, and then present them and get people to accept them.

The Will to Power, §409, Friedrich Nietzsche

In recent years there has been somewhat of a wave of conceptual engineering - people asking, not what our concepts are, but what they ought to be. There seems to me to be two main streams in this movement, which I think might be characterized as the epistemological and the ethical streams. Or rather, they are perhaps not as much separate streams as much as they are two sides of the same coin, just with a difference in emphasis. The epistemological stream emphasizes how engineering concepts can be useful in the search for truth and knowledge, while the ethical stream emphasizes how engineering concepts can be a driver of ethical and political change. But both are unified in working with concepts as workable rather than given. In this post I will mostly focus on the epistemological form, and basically give a short introduction to why I think that philosophers should be engineering concepts rather than analyzing them. I will not get very much into the weeds and nuances of conceptual engineering. The approach I take here could probably be characterized as a sort of ideal language approach - I am not so much interested in how we actually talk, but rather in what is the most useful way to talk for our purposes. This is something which is, in my mind, very y compatible with conceptual engineering, although much of the movement (especially the ethical stream) is more concerned with our natural language.

There Are no true Conceptual Analyses

In order to understand why we should be conceptual engineers, it will be useful to understand where we are coming from. The method we are superseding is conceptual analysis. This is the practice of finding the true necessary and sufficient conditions for a given concept. Basically find the right Y for a concept X, such that “something is X IFF it is Y” (IFF is read “if and only if”).

How do we check if an analysis is correct? Well, we can't check whether it is correct, but we can check whether it is wrong. We do this by finding a case of something which is Y but not X, or X but not Y.

The Historical Failure

To get a better understanding of this method, we can take a look at a historical example. Because I strive to be as mediocre as possible in all that I do, I will take the most overused example I can think of: The justified true belief analysis of knowledge.

Basically the story goes that most philosophers since Plato thought that in order to have knowledge it was necessary and sufficient to have justified, true belief. Sadly for most philosophers, there was a man called Edmund Gettier, who was having none of it! So he asked them: “Well, what if I look at a clock and it says that it is 9 o’clock, so I believe that it is 9o’clock, and it is actually 9 o’clock. So I have justified true belief. But actually the clock was stopped, and I just happened to be lucky to look at it at the right time. So I don’t have knowledge even though I have justified true belief!”¹. Mic drop! (Literally! The guy only published 3 papers and no books in his whole career, as far as I can tell (a career which lasted some 45 years, mind you))

Words cannot describe the chaos that ensued, and the field has not recovered since - nobody has been able to come up with a counterexample-free analysis, and now no one has any idea what “knowledge” means!!! Or as Alvin Plantinga puts it:

Knowledge is justified true belief: so we thought from time immemorial. Then God said, “Let Gettier be”; not quite all was light, perhaps, but at any rate we learned we had been standing in a dark corner.

This is not an isolated case (although perhaps one of the most clear cases) - it wouldn't be far off to say that there has never been performed a successful conceptual analysis on any interesting concept.

And We Will Never Succeed

I actually think that the failure of conceptual analysis is very unsurprising once we think about how we get our concepts. We don't get our concepts by formulating necessary and sufficient conditions for the application of them built from combinations of basic concepts, and then applying them to the world (as Locke thought). Rather we look at a bunch of examples of things that fall under and fall outside the concept and (sub)consciously infer some rules that explain why such and such examples don't fall under the concept and why such and such do.

The thing is that most concepts are very complicated and there are very many things that are more or less in common between different instances. Take something like the concept of “friend”. You form this concept by looking at other people who have what they call “friends” (or if you're lucky, you might yourself have such entities). You then begin to notice commonalities between such people. Perhaps they all laugh when they tell jokes, or perhaps they spend much time in the same places, or perhaps they are all happier than you. You could also begin to see cases that expand your concept - perhaps some people have friends on the internet, and so they don't spend time in the same places, but they still intuitively feel like friends. This expands your concept, but you still have a grasp of what is meant with it.

Now, try to give an analysis - what are the necessary and sufficient conditions for someone being your friend? That is going to be very difficult. The concept has so many moving parts and very many edge cases. Furthermore, it looks as though the concept changes through time, and across people, making it even harder to pin down. In fact, here is a wild suggestion: Perhaps there are no true necessary and sufficient conditions for what constitutes a friend - perhaps there is no actually true definition to be found. Any definition you give will have controversial cases where someone who intuitively is your friend will fall outside the category, or vice versa. In fact it is just not clear why we should expect there to be any neat necessary and sufficient conditions to our concepts at all - it just doesn't fit with the constant shaping and reshaping our concepts go through between times and individuals. Furthermore, everyone will have had different experiences from which they extrapolate a given concept, and so it would be really strange if there was some unified concept *out there* to find - one which would fit all our intuitions.

I think there might be one way to give necessary and sufficient conditions for our actual concepts. That would be to describe all possible counterfactuals about any possible case any possible person could be presented with at any time at all, and seeing whether they would say that the concept applies or not in each case. It should be obvious why this is probably worse than not giving an analysis at all. And in fact I am not even sure this would work - there will be many cases where the answer turns out to be “I don’t know” to whether the concept applies. Is the person you had five good conversations with 2 years ago and then never spoke to again your friend? It is just not clear that there is a yes or no answer. And so even that would fail to give clear necessary and sufficient conditions.

But what if we want to answer a question like “how many friends do I have”? Without a definite concept we will be unable to give a clear answer, even if we know all the empirical facts. We might have cases where it is clear whether someone is a friend or not, but there will also cases where it is not so clear, and without an analysis we will be unable to say anything meaningful about such cases. In fact we will not be able to give a clear answer to any question concerning friendship. What to do?

Introducing: Conceptual Engineering

Here is a solution: We just stipulatively definite the concept. From now on we say “X and Y are friends IFF X and Y are conscious agents, and X and Y spend at least 30 minutes a week interacting out of pleasure rather than necessity or obligation” (I know that this is a terrible definition, but it is just an example). There will of course, for any definition, be examples that intuitively count as friends but fall outside of the definition and vice versa, but now we at least have *something*, and we can now begin to answer questions like “how many friends do you have?” and “am I the only one without friends?”. In other words, we have made progress.

“Friends” is of course just an arbitrary example, but it should be clear how the above would also apply to more philosophically interesting concepts like “knowledge”, “gender”, and “Tibbles the cat”.

Concepts as Tools

We should of course not just make completely arbitrary definitions, like “X is knowledge IFF X is round and has long hair”. Rather we should think about how we can design concepts so that we can best achieve our goals (be those epistemic, ethical or whatnot).

In explaining this, it will also be worth it to address a worry that might begin to arise here. Because, if we just begin to go around stipulating whatever we feel like, will we not just end up wasting our time with wordplay rather than actually achieving stuff and finding things out about the world? I think we can answer this simply by thinking about what a concept is/does:

When we introduce a concept X, assuming that it is perfectly precisely defined (which it will really never be), we are basically dividing the whole world into two parts: One is the parts to which X applies, and the other is the parts to which X doesn't apply. By doing this we are highlighting the difference which the concept tracks. So when we say: “X and Y are friends IFF X and Y are conscious agents, and X and Y spend at least 30 minutes a week interacting out of pleasure rather than necessity or obligation” we are highlighting something about the reason why X and Y are interacting, and for how long.

By tweaking the concept, we adjust what part of reality we are highlighting, but we are still describing reality. An analogy could be photography: We can adjust which parts of the electromagnetic spectrum a camera detects, and hereby look at different features of reality.

From Wikipedia Commons

But we wouldn't say that it is more “real” or “right” to detect one part as opposed to another - rather it is just a question of what part of reality we want to highlight and investigate.

So we might think of our concepts as “probes” or “tools” for investigating reality. And we can define them such that we can best do that (or to achieve other sorts of goals). But there will still be open questions as to how reality is. Consider God defined as a necessarily existing omnipotent, omniscience and omnibenevolent being. Even if we have a fixed and well-defined concept, there will still be a question as to whether it refers to anything, and this still leaves room for further argumentation.

The new concepts we make should of course bear some resemblance to the concept they are trying to supercede, such that they have a similar extension and meaning (even if the original concept doesn't in fact *have* a clear extension or meaning) - if nothing else, then just to avoid unnecessary confusion. Most concepts also exist for a reason; they wouldn't have survived if they didn't have a use, and so there will often already be something useful that they are getting at. But as argued above, we should not toil endlessly to get it as close to our everyday intuitions as possible.

Of course not all concepts should be based on old ones - sometimes it will be useful to introduce entirely new concepts. Here there will be freer reins, although there are still “conceptual virtues” if you will. You will want the concept to be simple and precise so as to make the application of it easy and clear. And it should of course be useful and pick out something we are interested in. Take the example of supervenience, which arose during the 20th century: X supervenes on Y IFF a difference in Y is necessary for a difference in X. This is a very simple and clear concept, and it is very often useful in all sorts of areas.

I also think we can still make space for counterexamples as a tool of inquiry in a conceptual engineering framework. It will not necessarily be sufficient that a case doesn't intuitively fit with our definition of a concept, but if it turns out our concept excludes a case we want to include, or vice versa, then that will be a reason to revise our concept.

Suppose that we are engineering a concept of “man”. We might have a definition like this: “X is a man IFF X is born with a penis”. Then comes along a trans man, who falls outside this definition, but who identifies as a man. And suppose we have being nice people and not causing unnecessary harm for no reason as one of our goals in engineering our concept. In that case this person provides a counterexample to our definition, and we should change it to something like: “X is a man IFF X identifies as a man”. That will still have the same extension in most cases, but it better achieves our goals in other cases where the extension is changed, and so it is a better concept. We could, of course instead be interested in a concept of “male” for biological research, and in that case we should probably instead say: “X is a male IFF X is capable of producing small gametes” or something to that effect. This also shows how how we design a concept will be context-dependent. This is for example also reflected in how a tomato is generally considered a vegetable in culinary contexts, but a fruit in botanical contexts.

Where Do The Truth Conditions Derive From?

One question may begin to present itself. If the truth conditions of a term are fully determined by the truth conditions of the terms in its definition combined with the structure of the definition, then we end up in a bootstrapping problem. Either truth conditions are connected circularly, or in finite chains with beginning terms (it could also be infinite chains, but the dictionary isn't infinite, so that seems unlikely).

This is of course not only a problem with conceptual engineering. Conceptual analysis also tries to find necessary and sufficient conditions, and so runs into the same problems - tu quoque. There must be some point where our language “comes into contact” with the real world. This is really a much bigger question than what we can tackle here, and rests on questions of perception, reference, and meaning. The most I dare to say here is that this is a problem with language in general, and not just conceptual engineering.

I will also just reiterate here that I am not talking about the actual natural language we speak in our day-to-day lives. That language is, I suspect, a whole mess of loose concepts, that don’t have any clean bounds or definitions, and which serve the purpose of being good enough for communication and achieving our practical goals - I very much doubt that it has the “treelike” structure of an ideal language, with all concepts being constructed out of atomic concepts plus logical operators. But when we are doing philosophy, I think it is a good idea to at least try to make a more idealized language, which lends itself to detailed and precise investigation of the world.

Mind and Metaethics

Another problem that might begin to arise here is that conceptual engineering seems to reduce some areas of philosophy to mere questions of concept-design, where there actually seem to be substantive disagreement. I think answering this problem can also begin to give an answer to the above problem.

One area that can appear to be reduced, which really ought not, is philosophy of mind. When we ask “what is consciousness?”, it appears that a conceptual engineer can just define the question away. So we can just stipulatively say “X is conscious IFF X is a brain state of class Y”. Now we can go around in the world, pointing out things that are conscious and things that are not - the problem of other minds is solved!

The thing is, there is something specific that we want to get at, which we cannot stipulate away. You can of course take the letters “C-O-N-S-C-I-O-U-S-N-E-S-S” (Pronounced /ˈkɒn.ʃəs.nəs/) and create a language where you stipulate the above definition, or define consciousness in terms of wearing black hats. But we have an immediate access to a certain phenomenon, which we call consciousness, and we know what we mean when we say it. The best we can do for a definition is to say: “Consciousness is that”, while attending to our experience. I am of course not saying that we have to use the word “Consciousness”, we can of course also say “bevidsthed”, “Bewusstsein”, “schlorpglorp” etc. Rather the point is simply that as opposed to most concepts like “chair”, “friend” etc., here we have a brute phenomenon which we cannot be further defined, but which we are immediately aware of and can refer to directly.

This doesn't mean that we know everything about the thing we are referring to. It might still turn out that it is identical to brain states, or that it reduces to the functions of physical systems, or that it is an immaterial soul. If you are a type-A physicalist, you probably won't grant this, but please refer to the below image for a refutation of your view. And in any case, we still preserve there being a substantial dispute, even under a framework of conceptual engineering (in fact, by rejecting the validity of the brute concept altogether, type-A physicalists are still part of a substantive dispute, it seems - similar to non-cognitivism below).

I think a harder case it that of metaethics. It seems like we can give several definitions of “good”, like “X is good1 IFF X promotes pleasure”, “X is good2 IFF society approves of X”, or “X is good3 IFF X accords with an objective moral fact” or something like that. There doesn't seem to be anything contradictory or incoherent about accepting all of these at once. If that is the case, then it looks like metaethical disagreement will just reduce to semantic disagreement. But that doesn't look right - it looks like there is an actual disagreement between natural realists and error theorists, for example.

I think we again can introduce a brute concept of “goodness” here. So that we cannot further reduce/define that concept, only potentially identify it with other things. This might appear to beg the question in favor of realism, but I don’t think this is the case. I think we preserve the traditional positions (roughly) as follows:

• Non-cognitivism: “Goodness” is not an actual concept which can be part of propositions.

• Error theory: “Goodness” doesn't apply to any states of affairs.

• Relativism: “Goodness” applies to some states of affairs, depending on the attitudes of people/groups of people.

• Naturalistic Realism: “Goodness” is identical/reducible to some natural concept(s).

• Non-naturalistic Realism: “Goodness” objectively applies to some states of affairs and is not identical/reducible to some natural concept(s).

I am not quite as sure with this case as with consciousness, but it seems like at least a possible solution for now. I think that we might have to repeat this procedure for quite a few areas. For example something like “material object” and “true” might turn out to be brute in this same way.

It may feel like we are beginning to slip back into the conceptual analysis paradigm again; there are some real concepts out there which we are trying to grasp. But I don’t think this is the case. Remember, every time we introduce a concept like “goodness” or “consciousness”, we are introducing an irreducible concept; we can identify it with or put it in relation with other things, but it is in a sense sui generis and undefinable. So we should really avoid doing so as much as possible, and only introduce as many of such concepts as is necessary. Most concepts we deal with and use will instead be defined in terms of more basic concepts in a way we find useful, and thus be engineered.

Conclusion

Herman Cappelen characterizes conceptual engineering as broadly consisting in the following three steps:

(I) The assessment of representational devices, (ii) reflections on and proposal for how to improve representational devices, and (iii) efforts to implement the proposed improvements.

In this post I have mostly focused on the first two steps. These are the armchair steps, and steps which don't require very much theoretical baggage to accept. The question about how to implement conceptual changes will, for example, depend on whether you are an internalist or externalist about meaning. It will also be a complicated and messy question depending on a lot of empirical contingencies, how we best bring about wider conceptual changes in society. I am quite lazy, so I leave this mess to other people, or possibly even future me (if I decide to write about it).

1. Actually he didn't use that example, rather he used some much more contrived examples like:

Let us suppose that Smith has strong evidence for the following proposition:

(f) Jones owns a Ford.

Smith's evidence might be that Jones has at all times in the past within Smith's memory owned a car, and always a Ford, and that Jones has just offered Smith a ride while driving a Ford. Let us imagine, now, that Smith has another friend, Brown, of whose whereabouts he is totally ignorant. Smith selects three place names quite at random and constructs the following three propositions:

(g) Either Jones owns a Ford, or Brown is in Boston.

(h) Either Jones owns a Ford, or Brown is in Barcelona.

(i) Either Jones owns a Ford, or Brown is in Brest-Litovsk.

Each of these propositions is entailed by (f). Imagine that Smith realizes the entailment of each of these propositions he has constructed by (0, and proceeds to accept (g), (h), and (i) on the basis of (f). Smith has correctly inferred (g), (h), and (i) from a proposition for which he has strong evidence. Smith is therefore completely justified in believing each of these three propositions. Smith, of course, has no idea where Brown is.

But for pedagogical purposes, I used the more enlightening clock example.