Conceptual Analysis and Moral Theory

Part of the se­quence: No-Non­sense Me­taethics. Also see: A Hu­man’s Guide to Words.

If a tree falls in the for­est, and no one hears it, does it make a sound?

Albert: “Of course it does. What kind of silly ques­tion is that? Every time I’ve listened to a tree fall, it made a sound, so I’ll guess that other trees fal­ling also make sounds. I don’t be­lieve the world changes around when I’m not look­ing.”

Barry: “Wait a minute. If no one hears it, how can it be a sound?”

Albert and Barry are not ar­gu­ing about facts, but about defi­ni­tions:

...the first per­son is speak­ing as if ‘sound’ means acous­tic vibra­tions in the air; the sec­ond per­son is speak­ing as if ‘sound’ means an au­di­tory ex­pe­rience in a brain. If you ask “Are there acous­tic vibra­tions?” or “Are there au­di­tory ex­pe­riences?”, the an­swer is at once ob­vi­ous. And so the ar­gu­ment is re­ally about the defi­ni­tion of the word ‘sound’.

Of course, Albert and Barry could ar­gue back and forth about which defi­ni­tion best fits their in­tu­itions about the mean­ing of the word. Albert could offer this ar­gu­ment in fa­vor of us­ing his defi­ni­tion of sound:

My com­puter’s micro­phone can record a sound with­out any­one be­ing around to hear it, store it as a file, and it’s called a ‘sound file’. And what’s stored in the file is the pat­tern of vibra­tions in air, not the pat­tern of neu­ral firings in any­one’s brain. ‘Sound’ means a pat­tern of vibra­tions.

Barry might re­tort:

Imag­ine some aliens on a dis­tant planet. They haven’t evolved any or­gan that trans­lates vibra­tions into neu­ral sig­nals, but they still hear sounds in­side their own head (as an evolu­tion­ary biproduct of some other evolved cog­ni­tive mechanism). If these crea­tures seem meta­phys­i­cally pos­si­ble to you, then this shows that our con­cept of ‘sound’ is not de­pen­dent on pat­terns of vibra­tions.

If their de­bate seems silly to you, I have sad news. A large chunk of moral philos­o­phy looks like this. What Albert and Barry are do­ing is what philoso­phers call con­cep­tual anal­y­sis.1

The trou­ble with con­cep­tual analysis

I won’t ar­gue that ev­ery­thing that has ever been called ‘con­cep­tual anal­y­sis’ is mis­guided.2 In­stead, I’ll give ex­am­ples of com­mon kinds of con­cep­tual anal­y­sis that cor­rupt dis­cus­sions of moral­ity and other sub­jects.

The fol­low­ing para­graph ex­plains suc­cinctly what is wrong with much con­cep­tual anal­y­sis:

Anal­y­sis [had] one of two rep­u­ta­tions. On the one hand, there was ster­ile cat­a­loging of pointless folk wis­dom—such as ar­ti­cles an­a­lyz­ing the con­cept VEHICLE, won­der­ing whether some­thing could be a ve­hi­cle with­out wheels. This seemed like triv­ial lex­i­cog­ra­phy. On the other hand, there was meta­phys­i­cally loaded anal­y­sis, in which on­tolog­i­cal con­clu­sions were es­tab­lished by hold­ing fixed pieces of folk wis­dom—such as at­tempts to re­fute gen­eral rel­a­tivity by hold­ing fixed allegedly con­cep­tual truths, such as the idea that mo­tion is in­trin­sic to mov­ing things, or that there is an ob­jec­tive pre­sent.3

Con­sider even the ‘nat­u­ral­is­tic’ kind of con­cep­tual anal­y­sis prac­ticed by Ti­mothy Schroeder in Three Faces of De­sire. In pri­vate cor­re­spon­dance, I tried to clar­ify Schroeder’s pro­ject:

As I see it, [your book] seeks the clean­est re­duc­tion of the folk psy­cholog­i­cal term ‘de­sire’ to a nat­u­ral kind, ala the re­duc­tion of the folk chem­i­cal term ‘wa­ter’ to H2O. To do this, you em­ploy a nat­u­ral­ism-fla­vored method of con­cep­tual anal­y­sis ac­cord­ing to which the best the­ory of de­sire is one that is log­i­cally con­sis­tent, fits the em­piri­cal facts, and cap­tures how we use the term and our in­tu­itions about its mean­ing.

Schroeder con­firmed this, and it’s not hard to see the mo­ti­va­tion for his pro­ject. We have this con­cept ‘de­sire’, and we might like to know: “Is there any­thing in the world similar to what we mean by ‘de­sire’?” Science can an­swer the “is there any­thing” part, and in­tu­ition (sup­pos­edly) can an­swer the “what we mean by” part.

The trou­ble is that philoso­phers of­ten take this “what we mean by” ques­tion so se­ri­ously that thou­sands of pages of de­bate con­cern which defi­ni­tion to use rather than which facts are true and what to an­ti­ci­pate.

In one chap­ter, Schroeder offers 8 ob­jec­tions4 to a pop­u­lar con­cep­tual anal­y­sis of ‘de­sire’ called the ‘ac­tion-based the­ory of de­sire’. Seven of these ob­jec­tions con­cern our in­tu­itions about the mean­ing of the word ‘de­sire’, in­clud­ing one which asks us to imag­ine the ex­is­tence of alien life forms that have de­sires about the weather but have no dis­po­si­tions to act to af­fect the weather. If our in­tu­itions tell us that such crea­tures are meta­phys­i­cally pos­si­ble, goes the ar­gu­ment, then our con­cept of ‘de­sire’ need not be linked to dis­po­si­tions to act.

Con­trast this with a con­ver­sa­tion you might have with some­one from the Sin­gu­lar­ity In­sti­tute. Within 20 sec­onds of ar­gu­ing about the defi­ni­tion of ‘de­sire’, some­one will say, “Screw it. Ta­boo ‘de­sire’ so we can ar­gue about facts and an­ti­ci­pa­tions, not defi­ni­tions.”5

Disput­ing definitions

Ar­gu­ing about defi­ni­tions is not always mis­guided. Words can be wrong:

When the philoso­phers of Plato’s Academy claimed that the best defi­ni­tion of a hu­man was a “feather­less biped”, Dio­genes the Cynic is said to have ex­hibited a plucked chicken and de­clared “Here is Plato’s Man.” The Pla­ton­ists promptly changed their defi­ni­tion to “a feather­less biped with broad nails.”

Like­wise, if I give a lec­ture on cor­re­la­tions be­tween in­come and sub­jec­tive well-be­ing and I con­clude by say­ing, “And that, ladies and gen­tle­men, is my the­ory of the atom,” then you have some rea­son to ob­ject. No­body else uses the term ‘atom’ to mean any­thing re­motely like what I’ve just dis­cussed. If I ever do that, I hope you will ar­gue that my defi­ni­tion of ‘moral­ity’ is ‘wrong’ (or un­helpful, or con­fus­ing, or some­thing).

Some un­for­tu­nate words are used in a wide va­ri­ety of vague and am­bigu­ous ways.6 Mo­ral terms are among these. As one ex­am­ple, con­sider some com­monly used defi­ni­tions for ‘morally good’:

  • that which pro­duces the most plea­sure for the most people

  • that which is in ac­cord with the di­v­ine will

  • that which ad­heres to a cer­tain list of rules

  • that which the speaker’s in­tu­itions ap­prove of in a state of re­flec­tive equilibrium

  • that which the speaker gen­er­ally ap­proves of

  • that which our cul­ture gen­er­ally ap­proves of

  • that which our species gen­er­ally ap­proves of

  • that which we would ap­prove of if we were fully in­formed and perfectly rational

  • that which ad­heres to the poli­cies we would vote to en­act from be­hind a veil of ignorance

  • that which does not vi­o­late the con­cept of our personhood

  • that which re­sists en­tropy for as long as possible

Often, peo­ple can’t tell you what they mean by moral terms when you ques­tion them. There is lit­tle hope of tak­ing a sur­vey to de­cide what moral terms ‘typ­i­cally mean’ or ‘re­ally mean’. The prob­lem may be worse for moral terms than for (say) art terms. Mo­ral terms have more pow­er­ful con­no­ta­tions than art terms, and are thus a greater at­trac­tor for sneak­ing in con­no­ta­tions. Mo­ral terms are used to per­suade. “It’s just wrong!” the moral­ist cries, “I don’t care what defi­ni­tion you’re us­ing right now. It’s just wrong: don’t do it.”

Mo­ral dis­course is rife with mo­ti­vated cog­ni­tion. This is part of why, I sus­pect, peo­ple re­sist dis­solv­ing moral de­bates even while they have no trou­ble dis­solv­ing the ‘tree fal­ling in a for­est’ de­bate.

Disput­ing the defi­ni­tions of moral terms

So much moral philos­o­phy is con­sumed by de­bates over defi­ni­tions that I will skip to an ex­am­ple from some­one you might hope would know bet­ter: re­duc­tion­ist Frank Jack­son7:

...if Tom tells us that what he means by a right ac­tion is one in ac­cord with God’s will, right­ness ac­cord­ing to Tom is be­ing in ac­cord with God’s will. If Jack tells us that what he means by a right ac­tion is max­i­miz­ing ex­pected value as mea­sured in he­dons, then, for Jack, right­ness is max­i­miz­ing ex­pected value...

But if we wish to ad­dress the con­cerns of our fel­lows when we dis­cuss the mat­ter—and if we don’t, we will not have much of an au­di­ence—we had bet­ter mean what they mean. We had bet­ter, that is, iden­tify our sub­ject via the folk the­ory of right­ness, wrong­ness, good­ness, bad­ness, and so on. We need to iden­tify right­ness as the prop­erty that satis­fies, or near enough satis­fies, the folk the­ory of right­ness—and like­wise for the other moral prop­er­ties. It is, thus, folk the­ory that will be our guide in iden­ti­fy­ing right­ness, good­ness, and so on.8

The mean­ings of moral terms, says Jack­son, are given by their place in a net­work of plat­i­tudes (‘clauses’) from folk moral dis­course:

The in­put clauses of folk moral­ity tell us what kinds of situ­a­tions de­scribed in de­scrip­tive, non-moral terms war­rant what kinds of de­scrip­tion in eth­i­cal terms: if an act is an in­ten­tional kil­ling, then nor­mally it is wrong; pain is bad; ‘I cut, you choose’ is a fair pro­ce­dure; and so on.
The in­ter­nal role clauses of folk moral­ity ar­tic­u­late the in­ter­con­nec­tions be­tween mat­ters de­scribed in eth­i­cal, nor­ma­tive lan­guage: coura­geous peo­ple are more likely to do what is right than cow­ardly peo­ple; the best op­tion is the right op­tion; rights im­pose du­ties of re­spect; and so on.
The out­put clauses of folk moral­ity take us from eth­i­cal judge­ments to facts about mo­ti­va­tion and thus be­havi­our: the judge­ment that an act is right is nor­mally ac­com­panied by at least some de­sire to perform the act in ques­tion; the re­al­iza­tion that an act would be dishon­est typ­i­cally dis­suades an agent from perform­ing it; prop­er­ties that make some­thing good are the prop­er­ties we typ­i­cally have some kind of pro-at­ti­tude to­wards, and so on.
Mo­ral func­tion­al­ism, then, is the view that the mean­ings of the moral terms are given by their place in this net­work of in­put, out­put, and in­ter­nal clauses that makes up folk moral­ity.9

And thus, Jack­son tosses his lot into the defi­ni­tions de­bate. Jack­son sup­poses that we can pick out which plat­i­tudes of moral dis­course mat­ter, and how much they mat­ter, for de­ter­min­ing the mean­ing of moral terms—de­spite the fact that in­di­vi­d­ual hu­mans, and es­pe­cially groups of hu­mans, are them­selves con­fused about the mean­ings of moral terms, and which plat­i­tudes of moral dis­course should ‘mat­ter’ in fix­ing their mean­ing.

This is a de­bate about defi­ni­tions that will never end.

Aus­tere Me­taethics vs. Em­pathic Metaethics

In the next post, we’ll dis­solve stan­dard moral de­bates the same way Albert and Barry should have dis­solved their de­bate about sound.

But that is only the first step. It is im­por­tant to not stop af­ter sweep­ing away the con­fu­sions of main­stream moral philos­o­phy to ar­rive at mere cor­rect an­swers. We must stare di­rectly into the heart of the prob­lem and do the im­pos­si­ble.

Con­sider Alex, who wants to do the ‘right’ thing. But she doesn’t know what ‘right’ means. Her ques­tion is: “How do I do what is right if I don’t know ex­actly what ‘right’ means?”

The Aus­tere Me­taethi­cist might cross his arms and say:

Tell me what you mean by ‘right’, and I will tell you what is the right thing to do. If by ‘right’ you mean X, then Y is the right thing to do. If by ‘right’ you mean P, then Z is the right thing to do. But if you can’t tell me what you mean by ‘right’, then you have failed to ask a co­her­ent ques­tion, and no one can an­swer an in­co­her­ent ques­tion.

The Em­pathic Me­taethi­cist takes up a greater bur­den. The Em­pathic Me­taethi­cist says to Alex:

You may not know what you mean by ‘right.’ You haven’t asked a co­her­ent ques­tion. But let’s not stop there. Here, let me come alongside you and help de­code the cog­ni­tive al­gorithms that gen­er­ated your ques­tion in the first place, and then we’ll be able to an­swer your ques­tion. Then not only can we tell you what the right thing to do is, but also we can help bring your emo­tions into al­ign­ment with that truth… as you go on to (say) help save the world rather than be­ing filled with pointless ex­is­ten­tial angst about the uni­verse be­ing made of math.

Aus­tere metaethics is easy. Em­pathic metaethics is hard. But em­pathic metaethics is what needs to be done to an­swer Alex’s ques­tion, and it’s what needs to be done to build a Friendly AI. We’ll get there in the next few posts.

Next post: Plu­ral­is­tic Mo­ral Reductionism

Pre­vi­ous post: What is Me­taethics?


1 Eliezer ad­vises against read­ing main­stream philos­o­phy be­cause he thinks it will “teach very bad habits of thought that will lead peo­ple to be un­able to do real work.” Con­cep­tual anal­y­sis is, I think, ex­actly that: a very bad habit of thought that ren­ders many peo­ple un­able to do real work. Also: My thanks to Eliezer for his helpful com­ments on an early draft of this post.

2 For ex­am­ple: Jack­son (1998), p. 28, has a differ­ent view of con­cep­tual anal­y­sis: “con­cep­tual anal­y­sis is the very busi­ness of ad­dress­ing when and whether a story told in one vo­cab­u­lary is made true by one told in some allegedly more fun­da­men­tal vo­cab­u­lary.” For an overview of Jack­son’s kind of con­cep­tual anal­y­sis, see here. Also, Alonzo Fyfe re­minded me that those who in­ter­pret the law must do a kind of con­cep­tual anal­y­sis. If a law has been passed declar­ing that ve­hi­cles are not al­lowed on play­grounds, a judge must figure out whether ‘ve­hi­cle’ in­cludes or ex­cludes rol­ler­skates. More re­cent pa­pers on con­cep­tual anal­y­sis are available at Philpa­pers. Fi­nally, read Chalmers on ver­bal dis­putes.

3 Brad­don-Mitchell (2008). A fa­mous ex­am­ple of the first kind lies at the heart of 20th cen­tury episte­mol­ogy: the defi­ni­tion of ‘knowl­edge.’ Knowl­edge had long been defined as ‘jus­tified true be­lief’, but then Get­tier (1963) pre­sented some hy­po­thet­i­cal ex­am­ples of jus­tified true be­lief that many of us would in­tu­itively not la­bel as ‘knowl­edge.’ Philoso­phers launched a cot­tage in­dus­try around new defi­ni­tions of ‘knowl­edge’ and new coun­terex­am­ples to those defi­ni­tions. Brian Weather­son called this the “anal­y­sis of knowl­edge merry-go-round.” Tyrrell McAllister called it the ‘Get­tier rab­bit-hole.’

4 Schroeder (2004), pp. 15-27. Schroeder lists them as 7 ob­jec­tions, but I count his ‘try­ing with­out de­siring’ and ‘in­tend­ing with­out de­siring’ ob­jec­tions sep­a­rately.

5 Ta­boo­ing one’s words is similar to what Chalmers (2009) calls the ‘method of elimi­na­tion’. In an ear­lier post, Yud­kowsky used what Chalmers (2009) calls the ‘sub­script gam­bit’, ex­cept Yud­kowsky used un­der­scores in­stead of sub­scripts.

6 See also Gal­lie (1956).

7 Eliezer said that the clos­est thing to his metaethics from main­stream philos­o­phy is Jack­son’s ‘moral func­tion­al­ism’, but of course moral func­tion­al­ism is not quite right.

8 Jack­son (1998), p. 118.

9 Jack­son (1998), pp. 130-131.


Brad­don-Mitchell (2008). Nat­u­ral­is­tic anal­y­sis and the a pri­ori. In Brad­don-Mitchell & Nola (eds.), Con­cep­tual Anal­y­sis and Philo­soph­i­cal Nat­u­ral­ism (pp. 23-43). MIT Press.

Chalmers (2009). Ver­bal dis­putes. Un­pub­lished.

Gal­lie (1956). Essen­tially con­tested con­cepts. Pro­ceed­ings of the Aris­tote­lean So­ciety, 56: 167-198.

Get­tier (1963). Is jus­tified true be­lief knowl­edge? Anal­y­sis, 23: 121-123.

Jack­son (1998). From Me­ta­physics to Ethics: A Defense of Con­cep­tual Anal­y­sis. Oxford Univer­sity Press.

Schroeder (2004). Three Faces of De­sire. Oxford Univer­sity Press.