‘Conceptual engineering’ builds crooked thoughts

Illustrations

Imagine this little scene:

You: “May I have the payment?”
Them: “Certainly, here it is …”
You: “What? You have given me an IOU!”
Them: “No, it is perfectly OK: the concept ‘cheque’ has been redefined to include IOUs.”

What is going on here? You cannot help feeling that some strange funny business is being played on you. Is it really possible to ‘redefine’ something? …

——

Or let us observe this another way, with you playing the role of the ‘seller’.

Maybe you have commendable environmental concerns, and you have been trained with this philosophical technique that offers a means of help. Everyone wants to buy cars, but we really need them to switch to bicycles, so you adopt this special method: you redefine the category toward your ethical requirements, you redefine ‘cars’ to include ‘bicycles’!

But there appear to be some problems … On the sales web-site, what pictures will you have for items of this new concept for ‘cars’? If people see bikes, won't they just choose those other four-wheeled boxes with engines instead? And if you sell someone a car, and then they receive a bike, won't they complain that they did not get what they wanted? Are you going to reply that you ‘changed the concept of car’, and think that will satisfy?

——

What you have just met is a philosophical technique, called ‘conceptual engineering’ (or ‘ameliorative analysis’) and it even has some popularity in certain quarters. Yet even superficially, it looks like it cannot work properly.

Background

Let us review what philosophy thinks ‘conceptual engineering’ is:

From Chalmers:

“The more detailed definition […] is: assessing and improving our deficient representational devices. I'd say that's part of conceptual engineering. Suppose we said that civil engineering, say bridge engineering, was the project of fixing and improving our deficient bridges, or that software engineering was fixing and improving our deficient programs. That's part of bridge engineering and software engineering. But it's not the only part […]. There's also the whole project of building new bridges and building new software.”

‒ §4 p5 ¶1, "What is conceptual engineering, and what should it be?"; Chalmers; 2020 / https://doi.org/10.1080/0020174X.2020.1817141 / article.

And from Burgess and Plunkett:

“Which concepts should we use to think and talk about the world and to do all of the other things that mental and linguistic representation facilitates?”

“The first point here is that our conceptual repertoire determines not only what we can think and say but also, as a result, what we can do and who we can be.”

‒ §0 (abstract), "Conceptual ethics I, II"; Burgess, Plunkett; 2013 / article.

From Cappelen:

“1: if word W has meaning M, there are similar meanings that W could have.
2: There is no reason to suppose that M is the best meaning W could have.
3: It is important to ensure that our words have the best meanings possible.
4: Corollary: when doing philosophy, we should try to find the best meanings for philosophical terms.
5: So philosophers should do conceptual engineering.”

‒ "Fixing Language. An Essay in Conceptual Engineering"; Cappelen; 2018.

From Haslanger:

“My priority in this inquiry is not to capture what we do mean, but how we might usefully revise what we mean for certain theoretical and political purposes.”

“I'm asking us to use an old term in a new way.”

‒ §1 ¶6, §6 ¶6, "Gender and race: (What) are they? (What) do we want them to be?"; Haslanger; 2000 / article.

——

Here are two good summaries:

• "What is conceptual engineering, and what should it be?"; Chalmers; 2020 / https://doi.org/10.1080/0020174X.2020.1817141 / article.
  • article: https://philpapers.org/rec/CHAWIC-3
  • video: https://www.youtube.com/watch?v=6yLKd9_mpRc
  • transcript: http://consc.net/papers/engineering.pdf
• "Conceptual Engineering"; KaneB; 2020 / video.
  • video: https://www.youtube.com/watch?v=5APi7E09OBQ

Preview

Now, most importantly, one can see that there are two kinds of conceptual engineering:

• creating new concepts,
• changing existing concepts.

(‘Ameliorative analysis/inquiry’ being of the second type.)

The first ‒ creating ‒ is OK, though certainly not a new activity. That one is not what this article is about. But the second ‒ changing, or redefining ‒ is specious, nonsensical, and unworkable. If it is supposed to be ‘engineering’, it would not be licensable, it would be malpractice! Strawson's criticism ‒ ‘Is this not just changing the subject?’ ‒ is on target, but only a hint. Following is a precis of how that will herewith be pursued …

First, we face the ‘changeableness is meaninglessness’ trade-off problem. How much could you redefine something? You cannot change something and still have it fit the same uses in the world, unless that change is not meaningful to that usage. So the degree of choice is the degree of meaninglessness of the difference it makes.

Second, there will be the ‘self-obfuscation or subject-change’ conflict problem. A plausible change would be into something closely related. But the closeness of that relation implies that you could not understand one without the other. And so you could not redefine one with the other, since that is to delete and replace one with the other. Because if it is not an overwriting, it is only changing the subject.

And third, there is the ‘functionless or deceptive’ implementation problem. We must already know the old concept, else the proposal for us to change it is unintelligible. But the new definition must be different, else the proposal is functionless. So in any material instance ‒ where we bear the consequences ‒ we must know the truth, and the proposal must deny it; and the redefinition can only be realised by deception and ultimately force.

Eventually, the ‘superclass abstraction’ idea suggests a diagnosis and direction to treatment. Something cannot be the thing that changes, and the point from which the change is measured. So for a thing to be described as changed ‒ indeed, to know what has changed at all ‒ something must stay the same. There must be a fixed reference point, a superclass abstraction, that ‘before’ and ‘after’ concepts are subtypes of.

Changeableness is meaninglessness

Imagine throwing a ball in the air and seeing where it lands. The other person says it landed on area C, but you look and see it clearly fell on K. No, they explain, because area C has been ‘redefined’. You might reply that you do not care what the name may be, it fell on that area, next to the rose bush. But they then say that it is not about what the category is that is important, but what it should be. How does this make any sense?

——

“The category of ‘courier’ is socially constructed. Maybe it should be redefined to be ‘health worker’!” What could that even mean? It does not and cannot mean anything except in respect of what ‘courier’ and ‘health-worker’ already mean. And yet the ‘should’ only means anything as far as they are different.

There seems a very basic problem here: you cannot redefine the thing you are talking about. It is like grabbing something and declaring: “This thing! This is something else!”. But is there some way it can work? Can we not, in some way, adjust the boundaries of concepts? …

——

Imagine you sell fences, and find that people will equally accept hedges. So you change the definition of fence to include hedges. What has changed?

It seems to produce a kind of apples-and-oranges problem. The ‘should’ of changing the category is only substantial to the degree that the two are different. But the ‘should’ of changing the category is also only evaluable to the degree that the two are the same. So if you change the category slightly, how does it matter? And if you change it a lot, how is it comparable?

There can only be a discussion here if we are refining the border, asking if a bush this tall should really be called a tree ‒ tweaking the boundary edge. The validity of the discussion is inversely proportional to the amount of difference at stake ‒ more difference, less valid; less difference, more valid. We could not be talking about trees and biscuits, and deciding whether the concept we are talking about should be defined by one or the other ‒ they are just different things.

——

The degree of choice is in the degree of meaninglessness of the change. It is only because there is no way to evaluate the matter that there is a choice; if you could tell one way or the other you would be constrained accordingly. Where there is a choice, it is between particular features: you inspect the differences and see what gives you what you want. There is not some other freedom of movement, some leeway, there, that can be ‘conceptually engineered’.

When you are choosing between concepts, or coats, or cooking utensils, the question is what do they do, what features they have, and how they fit the circumstance. If they have different capabilities, you check them against your needs, but then what is the choice here exactly? It is wholly decided by the particular circumstances and the particular offerings. Is that freedom, is that choice? But if all the items do the same, are indistinct, you do have a free choice, and yet only because it is a choice of no substance, a meaningless choice. ‘Conceptual (re-)engineering’ seems to be trying to have it both ways: that we could somehow choose to change the shape of something and still have it fit the same slot in the world. Any question of selection or change of concepts here is not some abstract generality of ‘conceptual engineering’, it is about science, knowledge, real facts in the world. You choose things because they matter, and they matter because of how they relate to the world.

——

If there is a choice, the question is what are we looking at to make the choice, what features etc? There is no general answer that will make sense of ‘conceptual engineering’.

What is this special thing that can change its meaning? Can it be changed to anything? For any particular example, we must be talking about a particular thing, but then how could it be changed? What is the best meaning for the word or concept ‘shoe’? One immediately feels a struggle to answer. Can ‘man’ have its meaning changed to ‘tambourine’ or ‘narwhal’?
“What should count as men (or X, or whatever)?”
‒ “Men should. (How could anything else?)”.
The trivial inevitable tautology of that answer tells us about the question.
“But we have not decided what the concept is ‒ that is the point after all!”
‒ “Then what are we talking about, when we say ‘men’? Biscuits, spaceships … what?”
You must already know what ‘men’ (or X, or whatever) are to understand the question, and if you know that, then that thing is not open to choice, the answer is already foreclosed, and the question misleading.

Self-obfuscation or subject-change

Could you have a concept of trousers but not have a concept of leg? It seems imaginable to have the first not the second, but peculiarly ignorant/stupid. But is there a problem with the structure of understanding that makes it impossible ‒ ‘having’ concepts like that just would not count as understanding?

It seems a matter with two aspects: 1, concepts and the capabilities they bring, and 2, (material) things and their sub-parts. Understanding is relating things to others: extrapolatingly, that seems unbounded so offers no proper threshold there for crossing to understanding from not. Except at the lower end ‒ there must be a relation to at least one other thing. The two things ‒ eg, trousers and leg ‒ could be arbitrarily close, but they would still be different concepts, and so it seems possible to lack one. We could imagine them as just datastructures in the mind, and see that either could very well exist without the other.

You would think that to understand is to have a particular, successful, kind of relation between one's concepts or capabilities. And an object and its sub-parts seems like it might be an important or crucial such relation. To understand is to disassemble, recombine, re-use, and so on. Understanding is mapping and navigating the structure of the world. The original question, or rather the answer to that type of question, is ad hoc. And there are some things that if you had the one and not its close relation, then you could not be said to understand that dependent one. So it is with trousers. What is it in trousers that makes them that? Nothing at all but that they are a garment for legs.

——

We can find in this a rule regarding ‘conceptual engineering’. The more closely related two things are, the less you can claim to understand one without understanding the other, and therefore you cannot redefine one to the other ‒ since that means overwriting hence deleting one. And the less related, the more you can redefine one by the other, but since they are unrelated you are merely changing the subject.

And so this is more clarification of the dilemma inherent in redefinition: if you stay talking about one thing, then you pull the rug out from under yourself; but if your redefinition was valid, you slid into talking about two different things.

Functionless or deceptive

If someone says you could ‘redefine’ ‘person’, you might reply “redefine what? What thing are you talking about?”. We must already know what the thing is ‒ but then what do you mean (to do) by ‘redefining’ it?

Well, it seems there are two possibilities for ‘redefining’. First, you are proposing to actually change some fact somewhere. In which case, that is the important thing that we want to know about. And furthermore, you will have to explain using words that you have not changed, or no-one will understand you.

The second possibility is that you are not changing any facts. But then what are you doing? You must be merely changing words: rewriting labels or moving them around ‒ you take a bottle of water and you do not change it, you merely stick a wine label on. Now, assuming you also explain what you did, what do you achieve? You changed labels, but then by explaining, changed their meaning too ‒ you attached a wine label, but then stated that ‘wine’ now means ‘water’. So this cancels itself out and does nothing. But if you do not explain your label changes, it amounts to giving people different things to what they expected from the labels. This falls somewhere in the spectrum between miscommunication and lying.

There are two elements here: the word/name and the thing it points to, and those are what can be changed. Either you grip hold of the name (‘water’), and change its target (refill the bottle with wine); or you grip hold of the target (bottle of water), and change its name (to ‘wine’). Both of these operations produce confusion.

Take people (the customers/users) before you change the concept of X, and people after. Are they talking about the same thing? If they are not talking about the same thing, you merely changed the subject, and have done nothing significant to the original point of interest. If they are talking about the same thing, then that is the stable agreed fact of X, and again you have done nothing significant to the original point of interest.

So ‘redefining’ either means doing something real and explaining it in old unmodified words, or it is a superficial name change that if not cancelled out is a deliberate miscommunication.

——

You could tell people they should not buy cars, or you could increase the tax on cars, or enact a law to stop the manufacture of cars. Those would be ways to bring about your environmental policy aims. But you cannot simply ‘redefine the concept of car’. But how does conceptual engineering work otherwise? If it is not about effecting actual material change, but is confined to changing concepts, then how can it work except by withholding information? Any substantial meaningful change to any concept, if plain, would cause people to ignore it as merely a new label, and stay with the material goods they originally wanted.

And when you set out to ‘conceptually engineer’ and ‘ameliorate’ a concept, how do you know that you have succeeded, that you have improved matters? There must be something you look at to check it has improved. But then that is the reference point, that is the real subject here, and that has not changed ‒ that must stay the same for you to find it and check if it is better. So what was being changed by the ameliorative conceptual hocus-pocus?

——

The proposal that some category should be such-and-such transpires to this: a coercive misrepresentation. 1: you have to already know what the category is, else the proposal is meaningless; 2: the new definition of the ‘ought’ must be different, else the proposal is functionless. So for the proposal to be substantial, everyone must know the truth, and the proposal must deny it, and everyone is told they must accept it. That is, a lie by coercion.

The hinge of change

Perhaps this redefinition endeavour only makes sense if there is a tacit understanding of an intermediate abstraction in between word and use.

So one is saying not ‘this word should be used in a different way’ (which is vacuous ‒ there is nothing in that word's letter sequence itself), but really (eg) ‘this word means this kind of room, and such rooms should be used in a different way’. There are two things: the room, and its usage.

We all must already know, and stay knowing, that intermediate concept, that kind of room. So the word is not really being changed, and so the suggestion of re-definition remains somewhat deceptive. (And notice: aren't rooms often defined by their uses …)

What about cases like redefining the word/concept ‘woman’. Unlike a room, ‘woman’ is not a container for interchangeable objects and uses. The reply will then be that the equivalent is the ‘social space’ into which women conventionally fit. But here the core of the problem is approached. How do you define ‘women's social space’? You must already know what ‘woman’ is! But was that not the (apparent) subject of re-definition? There is some kind of circular question-begging going on.

So it seems an intermediate abstraction does not help the fundamental problem in conceptual engineering, but it has allowed us to home in on what seems its root defect.

——

If someone says they will change that cup into an apple, maybe they would perform some magic and misdirect your attention away while replacing one object with the other. Or they could even simply take the cup out of the house and bring an apple in. But they cannot merely point to the cup then point to an apple, and say that is a change. In the first, you hold the pointing fixed (though distracting momentarily), and change some of the facts; in the second the facts are fixed, and you change the pointing. Changing something is not the same as talking about different things. To change something, it must have something static.

With a physical object, say a car, with the right tools you could actually change it into a bicycle. It seems that you will have succeeded in substantially changing the thing, rather than only ‘changing the subject’. But this is because with physical objects there is a kind of ‘place’ that holds the thing, and that is really the subject: that piece(s) of material in that position in the world. (You have changed a car, not the definition of ‘car’.)

——

When someone says they want to redefine the concept of ‘car’, there is a syntactic/parsing illusion: it seems to make sense only because of a confused idea that there is a car and the definition of the car, and so one can be changed separate from the other. But there are not two separate things there; the concept, or thing, is the definition.

There is the word, and the thing it refers to. And those are the only items that can be changed. If you change the word ‒ ie make everyone use another sign to think and do things with that object, what have you achieved? And if you change the thing ‒ ie make everyone think and do something else when using the word, the original target object is abandoned and untouched.

Superclass abstraction

For a thing to be described as changed, something must stay the same. There must be a fixed reference point, otherwise you cannot say what it is that changed. This is just the essential syntax and structure of conceiving of change: change is a particular relation of those two elements. A statement of a change is really a statement of a kind of similarity of the things, and (the rule of) that similarity is the fixed part standing outside the items under consideration.

Look at how ‘social construction’ arguments are made from this misunderstanding. People say ‘male/female is socially constructed’, and their point is that they can be changed, or rather we could have a different interpretation. So we have an implied instance of the two-part structure, but it is also presented as being about only one thing. If you ask them ‘How might female/male change? Give an example’, you have them stumped. (Ask them if they can think of another way that humans might reproduce besides by male/female small/large gametes ‒ but do not expect an answer.)

They would have to say that it is a matter of what we, society, does with male/female, the way it is treated ‒ that is what can be changed. But in that they are forced into a clarity that it is not female/male that is ‘socially constructed’ after all, but instead activities etc that are built on top of male/female, and so require female/male to be the reference point that is the static element of the structure.

So their argument evaporates. They tacitly chose female/male as what they are talking about, but say it is the thing changing. But it cannot be the thing that changes, and the point from which the change is measured.

——

In software engineering terms, this is a confusion of ‘entity objects’ and ‘value objects’. ("Domain-Driven Design"; Evans; 2003 / ISBN-0321125215 / book.) Consider the product of some calculation, a variable called ‘result’: it might have a value ‘42’, or ‘0’ or something else. Here, we have a kind of box that contains an integer: we hold that variable/box ‘result’, and the integer value inside it can change. This is an entity. But compare that with an integer itself. If you apply, eg, an increment to the value ‘1’, you do not change that object into ‘2’; instead you create a new value ‘2’ ‒ the integer ‘1’ (or any other) can never be anything else. And this is a value. Where an entity object ‒ like ‘result’ ‒ can change, a value object ‒ like ‘42’ ‒ is immutable; with an entity you hold a ‘box’, and with a value you hold the thing.

This is a basic distinction, not in the way things are, but in how we can communicate and model. To forestall confusion when we talk about concepts, we must be clear whether we are talking about entities or values. Concepts in themselves are value objects ‒ immutable, and cannot be redefined. Redefinability can only be possible where there is also a second concept, forming an entity object by playing the role of the ‘box’ subsuming the modifiable value.

——

To propose a change one must implicitly be proposing a non-changing reference point: to change from A to B there must also be an X above both, that joins the two together. Without the third part, the superclass abstraction, redefining becomes circular and impossible. And this looks like the diagnosis of conceptual engineering (of the second kind), that it is a confusion of what is changed and what is static.