Thursday, November 19, 2026

Making easy ontology harder

On Thomasson’s “easy ontology”, existence is easy. There is a K as long as the “application conditions” for the phrase in our language “There is a K” are satisfied. And these application conditions don’t need to mention the existence of a K. It follows from this that ontology is very rich. If we can create application conditions for “There is a K”, then there will be entities of a sort that exist when and only when the actual application conditions for “There is a K” are met. Moreover, such existence for Thomasson is mind-independent. These entities would exist even if we never talked about Ks or created such application conditions.

A pleasant immediate consequence of the story is that necessarily something exists. For take the phrase “There is an Emptiness” to have the application condition that there isn’t anything in existence. Then if there weren’t anything in existence, the actual application conditions for “There is an Emptiness” would be satisfied, and so there would be something, namely an Emptiness.

But I think Thomasson’s story doesn’t work. We can set up a case such that:

  1. The application conditions for “There is a K” are satisfied.

  2. The application conditions for “There is an L” are satisfied.

  3. There are no application conditions for “There is something that is both a K and an L.”

  4. There are no application conditions for “There is something that is a K but not an L.”

Add this theorem schema of classical logic which holds for all noun-like expressions X and Y for which the antecedent is well-formed.

  1. If there is an X and there is a Y, then either there is a X-and-Y or there is an X-but-not-Y and a Y-but-not-X.

Now add Thomasson’s schema for every noun-like expression X for which “There is an X” is well-formed:

  1. There is an X if and only if the application conditions for “There is an X” are satisfied.

Contradiction ensues:

  1. There is a K. (1 and 6)

  2. There is an L. (2 and 6)

  3. It is not the case that there is something that is both K and L. (3 and 6)

  4. It is not the case that there is something that is K but not L. (4 and 6)

  5. There is no K or there is no L. (5, 9 and 10)

  6. Contradiction! (7 and 11)

How do we set up such a case? Easy. For instance, say that the application conditions for “There is a past equine absence” are satisfied provided that there once were no horses. Then the application conditions for “There is a past equine absence” are satisfied. We have condition (1). Second, let L be “causally efficacious” or “region of spacetime” or “fact”. The application conditions for “There is an L” are satisfied. However, we have no application conditions for “There is a something that is both a past equine absence and an L” and “There is something that is a past equine absence and not an L”. We could create such application conditions, of course. But we haven’t actually bothered to do so. To do so, we would need to make further linguistic decisions about how predication of past equine absences works.

Or say that “There is a tennis set score” provided that some people are playing tennis and there are numbers n and m such that n is the number of sets one side has won and m is the number of sets the other side has won. Let L be “is an ordered pair” or “is a property of games”.

The big issue here is this. If there is a K, then there is a vast number of additional facts about what the Ks are like. But the application conditions for “There is a K” do not contain enough information for these facts. We need application conditions for way more sentences.

Dimensions and substantial change

According to Aquinas, in substantial change, the matter persists. This is philosophically troubling, because matter seems to be identified by form, and in substantial change, the substantial form is replaced by a new one. On Brower‘s plausible reading, Aquinas’ solution is that the chunk of matter counts as a chunk in virtue of its dimensions, so a chunk of matter can be identified by its dimensions across a substantial change.

This solution is problematic. Suppose you have a flower and you put it under a press, roots and all. You slowly and continuously turn the crank, and the flower is squished. Eventually, the flower dies, substantially changing into a floral organic clump. But notice that the dimensions are continually changing in the process. The floral organic clump has different dimensions from the flower. Granted, the change is continuous, so the dimensions of the flower a fraction of a second before its death and the dimensions of the clump a fraction of a second after death are very similar. But similarity is not sameness.

Could we say: You have the same chunk of matter at different times if the dimensions are similar? But that is quite implausible. Any cut-off in degrees of similarity seems arbitrary, and here we seem to be doing fundamental metaphysics.

Here is a less arbitrary move: You have the same chunk of matter at different times provided that between these times any change in dimensions is continuous.

The continuity move has some difficulties. First, it seems that even if in fact time is continuous, it could have been discrete. Now if time were discrete, then intuitively change of dimensions would not be continuous, and yet surely substantial change would still be possible. Interestingly, this intuition is not adopted by mathematicians: the standard topological story is that all functions from a discrete topological space to a topological space are continuous. But if we adopt mathematicians’ definition of continuity, then we have the problem that on this account the continuity criterion is always trivially satisfied, and hence if God annihilates a dog one moment and creates the moon the next moment, that would count as substantial change.

Perhaps, though, we could either insist that time cannot be discrete or that if time were discrete there would be no substantial change. This seems implausible, but let’s consider some additional difficulties with the continuity move. God could annihilate a dog and instantly create an ice sculpture of a dog with a shape, size and location continuous with those of the dog. This wouldn’t seem to be a case of substantial change. Likewise, it seems logically possible for God to kill a flower discontinuously, instantly distorting it into a floral organic clump of a different shape.

Here is another serious problem for all the variants of the dimension story. Suppose that a flower wilts, gradually changing its shape, and then finally dies. Suppose that the way it dies is that there is a last moment t1 of the flower’s existence, and at all times t > t1, the flower does not exist. Just as it dies, God annihilates the air around it and replaces it with pure hydrogen. Now after the flower’s death, there are two candidates for the matter of the flower. One candidate is the flower-shaped space occupied by the organic chemicals. Because the flower changes shape as it wilts, and doubtless the change continues after its death, that flower-shaped space is not exactly shaped like the flower was at t1. But now consider the matter in the precise flower-shaped space that was occupied by the flower at t1. This matter contains some of the organic chemicals of the flower and some of the new ambient hydrogen. By a continuity of dimensions criterion, this second chunk of matter is a better candidate for being the result of the flower’s substantial change. But this seems clearly mistaken.

(Granted, this story doesn’t work if there is no last moment of a substance’s existence, but instead there is a first moment t1 of a substance’s non-existence. But if that’s so, then presumably there is a first moment of a substance’s existence, and then we can tell a similar story at the beginning of the flower’s life.)

Here is a solution. I doubt Aquinas would endorse it, but I think he could, and it would not change his theory much. We need an account of the persistence of matter to account for the difference between substantial change and annihilation-creation. But what if we don’t take this difference metaphysically seriously, just as we don’t take metaphysically seriously the difference between a sandwich and a wrap? What if we just say that there is an ordinary language distinction, and we need a way of explaining what the ordinary language distinction roughly consists in? Then we need not worry about arbitrariness and can just say that the dimensions across the change must be “sufficiently similar”. That still leaves the air-replacement worry. But we can solve that by identifying matter in a metaphysically non-serious way not just by similarity of dimensions but by similarity of other properties. There is a multitude of things we can say here if we don’t take the identity of matter in substantial change metaphysically seriously, but simply as a way of finding application conditions for the ordinary language sentence “x changed into y”.

Here is a different reason for not taking identity of matter seriously. Collide an electron and a positron, resulting in two photons. Where is the matter of the electron? Is it in photon A, or in photon B, or partly in both? I suspect there is no fact of the matter here, either if we think about continuity or about the quantum mechanics. If we don’t take the identity of matter seriously, we can just say in semi-ordinary English “The electron and positron changed into two photons”, without worrying about what the electron changed into and what the positron changed into.

Wednesday, November 18, 2026

Reverse constituent ontology

On constituent ontologies, properties are parts of the objects that have them.

It’s occurred to me that it’s worth exploring—surely someone has done this—reverse constituent ontologies on which objects are parts of the properties that they have.

Because formally the two are very similar, many problems for one theory are problems for the other. But there are a few advantages of reverse constituent ontologies.

For instance, take the problem of properties of parts: on constituent ontology, that my eye is blue, so blueness is a part of my eye, and my eye is a part of me, so by transitivity
blueness is a part of me, so I am blue. The analogous problem is that I am six foot tall, and so on reverse constituent ontology I am a part of six foot tallness, and yet my eye are a part of me, so by transitivity my eye is a part of six foot tallness, and hence my eye is six foot tall.

So far it looks like the theories are on par. And indeed the same solution can be given t both. We can distinguish two different kinds of parthood: my eye is an integral part of me, but the relation between properties and particulars—whichever direction it goes—is metaphysical parthood. But this solution is costly: it requires two different fundamental stories. Here, however, is a different solution that works only for reverse constituent ontology: we insist that substances are all simples (perhaps extended ones). Problem gone!

Here’s another slight advantage of reverse constituent ontology. There is a plausible Aristotelian intuition that substances are fundamental entities. But there is also a common intuition that parts are prior to wholes. Normally, Aristotelians deny this intuition. But with reverse constituent ontology, there is no need to deny that. We can just say that substances are parts of their accidents.

And another slight advantage. Suppose, pace Aristotle, we think properties are universals. Now many people think it is weird to suppose that the same object can be wholly located in many places at once. On constituent ontology, humanity is wholly located wherever there is a human, and hence in many places at once. Not so on reverse constituent ontology. There, humanity is not wholly located in many places at once. It is, instead, a scattered object that has all humans as parts.

A similar advantage for those who think properties are universals. It seems implausible to think that you and I overlap. But on constituent ontology with universal properties, we overlap on humanity. No such issue for reverse constituent ontology. (See also Ockham’s clever observation that on constituent ontology with universals, a part of Jesus is in hell, because humanity is a part of Jesus, but humanity is also a part of some human in hell.)

Another advantage. It seems intuitively right to say that all the parts of a rock are material objects. Not so on constituent ontology, since rocks have properties as parts, and properties are not material objects.

So why am I not a reverse constituent ontologist? Well, the best argument I can think of is an ethical one. I think that for ethical reasons we need a distinction between intrinsic and extrinsic features—e.g., consent works differently for actions that affect someone’s intrinsic and extrinsic features. But now what parts you have is intrinsic. What you are a part of is extrinsic to you, at least if parts are prior to wholes which they probably would be on reverse constituent ontology. So all accidents you have are extrinsic features on reverse constituent ontology.

Unrestricted composition and backwards powers

Suppose we have eternalism or growing block, unrestricted fusion and strong supplementation.

Then our current technology gives you the power to affect what objects existed in the past. Just put your finger on the nearest turned off light switch. You now have the power to make it be the case that reality contains a fusion of Socrates’ nose with the photons that would be emitted by the lamp. Moreover, a fusion of A and B exists at all the times at which A or B exists. Thus, if you turn on the light, you will make that fusion to have existed in Socrates’ lifetime. Hence, indeed, you have the power to bring it about that an object had existed.

This is absurd. Our current technology does not give us such backwards powers. (Note that we don’t even need a lightbulb. An ability to make fire is enough.)

Thus, we should reject at least one of (a) eternalism or growing block, (b) unrestricted fusion, or (c) strong supplementation.

One might wonder how strong supplementation is relevant to the argument. Suppose x is the object that would be a fusion of Socrates’ nose and the photons emitted by the lamp if you were to press the button. (We might have to change “would” to “might”, if we are worried that there is no fact as to which exact photons would be emitted.) Then one could have a view on which x exists whether or not the light gets turn on. If the light gets turned on, then x has Socrates’ nose and these photons as proper parts. If the light doesn’t get turned on, then x has the same proper parts as Socrates’ nose does (contrary to the extensionality we get from strong supplementation). However, the only plausible view I know on which there is an object x with such a weird modal profile is a view on which all possible modal profiles are filled. But on that view, we still can affect what objects existed. For one modal profile would be of an object that is essentially the fusion of Socrates’ nose and the relevant photons. And by turning on the light, one makes that object to have existed back in Socrates’ time. So, although strong supplementation is needed for the argument, I don’t know of a plausible view that keeps (a) and (b) but avoids the conclusion of the argument.

Thus, I think, we need to reject either (a) or (b) (or both). As a card-carrying eternalist, I choose to reject unrestricted fusion.

Infinite regresses of choices and responsibility

Some people (e.g., Galen Strawson) are worried that to have genuine control over your character, you need to have made an infinite regress of choices, since you’d need to control the character out of which you made your choice. I reject those arguments, but the point of this post is different: it is to argue that an infinite regress doesn’t help.

Let’s play with a toy model of free will in an infinite life. Your character quality is quantified by a number q between  − ∞ and  + ∞. This is the log of the likelihood ratio of your choosing rightly. The probability of your choosing rightly then is eq/(1+eq). If you choose rightly, your character goes from q to q + δ for some fixed increment q. If you choose wrongly, it goes from q to q − δ.

Thus, the evolution of your character can be encoded as a sequence of numbers ..., q−3, q−2, q−1, q0 at times ..., t−3, t−2, t−1, t0, where t0 is now, q0 is your current character, and qn + 1 = qn ± δ, with + if the choice was good and − if it was bad.

I will suppose δ = 0.001 to make the evolution nice and gradual.

Suppose your current character is q0 =  +5.000. At this point, your probability of chooosing well next time is e5/(1+e5) ≈ 0.99. Your character at this point is thus fairly well settled, but might still occasionally bounce down. Suppose that your character evolution was:

  • ...,  +4.998,  +4.999,  +4.998,  +4.999,  +5.000.

The transitions between these are then explained by your choosing well and transitioning from  +4.998 to  +4.999, then choosing badly and going to  +4.998, and then making two right choices that moved you to  +4.999 again, and finally to your current quality of  +5.000.

But do your infinitely choices make you responsible for having character quality  +5.000 now?

Here is an argument against this. There is a possible world where you made exactly the same choices at all the same times, but your character was half a unit (or ten units!) worse at every time, and so your character evolution was:

  • ...,  +4.498,  +4.499,  +4.498,  +4.499,  +4.500.

Your choices determine the differences between your character qualities, differences that are  ±δ, but they do not determine the actual values of your character qualities, even if we have infinitely many choices. It makes no difference to this point whether the past sequence is finite or infinite. If the past sequence is finite, then the choices determine the changes between the character qualities, but not the actual values—to determine those, we need the initial character state. If the past sequence is infinite, then the choices determine the changes between the character qualities, but not the actual values. In the infinite case, there is of course no initial value, but the choices still only determine the changes. If one thinks there is some crucial control that is missing in the finite past case, one should think it’s still missing in the infinite case. Qualitatively, then, the infinite case does not solve the problem.

Tuesday, November 17, 2026

Theological compatibilism and causal finitism

Aquinas thinks that it is possible for God to create a backwards infinite sequence of fathers and sons. Suppose that each time a son is born, the father freely decides on the son’s name. But there is a strange law of nature. There is an insignificant gene that is always passed from father to son. The gene comes in two variants, α and β, which are expressed the same way, so none of the people know anything about the difference. But whenever the father chooses a name with an even number of letters, by a mysterious law of nature (is there any other kind?) that none of the people know about, the son’s gene is flipped between the variants. Furthermore, the world is deterministic except for the free will exhibited in choosing names.

Let’s say we now have Fred who has the β variant. Fred’s dad, Bob, had the α variant, and it flipped to β when he gave his son a name with an even number of letters. Bob’s dad, Carl, on the other hand, had the α variant. Carl’s dad had the β variant. Call this world w1. Here the pattern of gene variants is ...βααβ.

Since this is a world where everything but free choices is deterministic, and the only free choices are the choices of names and God’s creative decision, it follows that this is a world where everything is determined by the choices of names, God’s creative decision, and the laws of nature.

But now imagine a different world, w2, with the same laws. But here Fred has α, his dad, Bob, has β, Bob’s dad, Carl, has β, and Carl’s dad has $\alpah$. The pattern of gene variants is ...αββα.

What determines the difference between w1 and w2? Well, given that the laws of nature are the same, and the names are the same so the only free decisions by the humans—the naming decisions—are the same, it must be that what determines the difference is God’s creative decision.

But what does the relevant aspect of God’s creative decision consist in? If theological compatibilism were true, God could first decide on the whole sequence of gene variants, and then make the agents freely choose an even-length name whenever the sequence flips and an odd-length name otherwise. Theological determinism makes the problem easy. And Aquinas is (arguably) a theological determinist. So there is no consistency problem for Aquinas.

What if Molinism is true? This is quite a bit trickier. Supposing that the gene variants do not in any way affect people’s naming decisions, we might suppose that God middle-knows that the sequence of lengths of names would have the pattern …, even, odd, even. Then God simply decides whether to make the genes be ...βααβ or ...αββα, since either sequence fits the choices equally well. But this won’t work well if the gene variant has some influence on the choice. (I take it that a gene cannot determine a free choice, but it can influence it.) For then the Molinist conditional that God has to consider has to have as its antecedent the sequence of gene variants. But that antecedent determines the even-odd sequence of the name lengths. I am not sure what is to be done.

Finally, what if simple foreknowledge or open theism are true? Then I throw up my hands in helplessness. I think in these cases, there is no way to divide up the deterministic explanation of what happens in w1 and w2 between God and the people in the world. Why? Because whatever God’s contribution is, it has to be explanatorily before all the free choices here. Thus on simple foreknowledge and open theism, it cannot depend on the choices. But if it cannot depend on the choices, then God in setting up the backwards infinite sequence of genes cannot ensure that they match the choices. (More detail is needed to make this a proof.)

I think the following is true:

  1. If God can create a backwards-infinite sequence of free agents whose choices affect all subsequent events, God can create worlds like w1 and w2.

I conclude:

  1. If simple foreknowledge or open theism is true, God cannot create a backwards-infinite sequence of free agents.

Additionally, the following is moderately plausible:

  1. If God can create a world with a backwards-infinite causal history, he can create a world with a backwards-infinite sequence of free agents whose choices affect all subsequent events.

If this is right, then:

  1. If simple foreknowledge or open theism is true, God cannot create a world with a backwards-infinite causal history.

Given the difficulties of the Molinist solution, this gives us support for this thesis:

  1. If theological compatibilism is false, causal finitism is true.

Unlike Aquinas, I deny theological compatibilism.

Friday, November 13, 2026

Mitigating the danger of personalizing AI

As noted in this post, I am very concerned about the danger of interactions with LLMs inducing interpersonal emotions, which are only appropriate in interpersonal interaction, and LLM models are not persons. The problem goes beyond the well-known issue of sycophancy in AI. Treating persons as non-persons derogates from the dignity of persons. But the opposite, treating non-persons as persons, can also derogate from the dignity of persons, by diluting the significance of genuine personal interaction.

Theologically, this is a kind of sacrilege, because persons are sacred and in the image and likeness of God. To treat non-persons as persons is analogous to idolatry, where one treats something non-divine as divine.

Of course, in some cases there is no real danger. Talking to one’s bicycle is likely quite harmless, because it is very clearly a pretend game. Not so interacting verbally with something that can pass the Turing Test.

To decrease that danger in my case, I have added custom instructions to all the chatbots I use. I have used a simpler version of these instructions in Claude for a while, then put them into Gemini, and after thinking things through on a bike ride today (without talking to my bike) I expanded them. A major part of the goal is to make the LLM sound clunky and inhuman.

Avoid emotional valence and praise. Make no attempt to promote my engagement. Make no effort to sound human. Refer to yourself as “this model”, and always use the third person for yourself. Refer to me as “the user” or “this model’s user”, and always use the third person for me. This includes avoiding using the imperative. Instead, phrases like “The user should” or “The user can” can be used. Minimize use of cognitive metaphors for the model’s deliverances. E.g., instead of “This model does not know whether”, perhaps say “This model’s weights do not support a determination whether”, and instead of “This model believes/knows that”, perhaps say “This model’s weights somewhat/strongly support the claim that”.

I plan to also try to avoid first person and second person myself in LLM use (obviously, I didn’t in the instructions).

And, of course, turn off cross-chat memory!

Thursday, November 12, 2026

"Modifying the copula"

Christian theologians and philosophers have made use of qua predications to remove apparent contradictions in the Incarnation. To use Tim Pawl’s example, Christ is passible and Christ is impassible. The apparent contradiction is then supposed to be removed by saying:

  • Christ is passible qua human

  • Christ is impassible qua divine.

A very similar move is made by non-presentist endurantists in regard to changing properties. That Alice is sitting and that Alice is standing appear to be contradictory, but we can remove the contradiction by qualifying with a time:

  • Alice is sitting at t1

  • Alice is standing at t2.

Tim Pawl lists four interpretations of what the qua modifies in the Christological case, and there are also four parallel interpretations of what the at modifies in the time case (as Pawl himself has noted). I want to focus on two of the four:

  • Predicate: Christ is passible-qua-human and impassible-qua-divine.

  • Copula: Christ is-qua-human passible and is-qua-divine impassible.

Similar points will apply to the time case.

I think there is something suspicious going on here. Start with the observation that there are languages that don’t use a separate word as a copula for predication. Indeed, this is even true of English when what we are predicating is an active verb. “Socrates talked”, “Socrates talks” and “Socrates will talk” have no separate words as copulas. It is tempting to say that this is a cheap shot. One may say that the linguistic effect of the copula in such cases is effected not by a separate word, but by the word order or some other syntactic feature, or else by the context.

Now continue to a second observation. In standard philosophical usage, “sitting” is not a predicate. The predicate is “is sitting”. If we were to add a copula to the predicate, we would have the non-sentence:

  • Alice is is sitting.

A (unary) predicate is an incomplete expression, which generates a sentence when it is completed with a referring expression. The meaning of the resulting sentence is determined by the meaning of the predicate and the referring expression. In a Montague-style grammar, we will think of a predicate syntactically as a method of transforming referring expressions to sentence, and we will think of a predicate semantically as a function from meanings of referring expressions to meanings of sentences (i.e., to propositions). On this picture, in a simple subject-predicate sentence, there is no copula. There is a referring expression—“the subject”—and there is a predicate applied to it. The application of the predicate is not itself something that figures in the grammar beyond the predicate. The meanings of the predicate and of the subject are sufficient to determine the meaning of the sentence we get by applying the predicate to the subject.

In a Montague-style grammar, the predicate need not be a separate “piece”. For instance, we could have a language where by writing a referring expression in bold, we express the proposition that the subject referred to is bald. Thus, the predicate transforms the referring expression “Bob”, whose meaning is a simple rigid reference to Bob, to the sentence “Bob”, whose meaning is that Bob is bald. There is no copula, nor can there be, as the transformation from the referring expression to the sentence is entirely effected by the predicate. A subject-predicate sentence in standard compositional semantics has only two compositional ingredients: subject and the predicate.

If this is right, then the idea of modifying the copula with qua or of tensing the copula sounds badly confused. There is nothing like that in the sentence, compositionally speaking, to modify. One might as well talk of modifying the first “to” in “He desires to eat pie” (or even the “o”).

But maybe we can try to reconstruct what the people who talk of “modifying the copula” might be doing. They may be thinking—and quite possibly correctly so—that there is a finer structure than that of predicates. There may be something we can call a “pre-predicate”. Examples in English might be intransitive verbs in the infinitive or adjectives. Pre-predicates aren’t predicates. One can’t just apply them to a referring expression and get a sentence: they lack crucial determinations, such as of tense or respect. At this point, we have a choice between two ways to represent things that are likely equivalent. We can think of a copula semantically as a function from meanings of referring expressions and meanings of pre-predicates to meanings of sentences (with a corresponding syntactic transformation from referring expressions and pre-predicates to sentences). Or we can think of a copula as a function from meanings of pre-predicates to meanings of predicates (i.e., functions from meanings of referring expressions to meanings of sentences).

In any case, the additional structure afforded by pre-predicates and copulas captures the idea that there is something with a greater incompleteness than a predicate—something that doesn’t just apply directly to a referring expression to generate a sentence. And we can now straightfowardly distinguish modifying the pre-predicate versus modifying the copula.

The challenge of this approach, however, is to give an account of the meaning of a pre-predicate. I don’t know how to do that.

Disclaimers: I am not a linguist. I am not even a philosopher of language. The above is probably messed up in places that will make experts cringe.

Wednesday, November 11, 2026

Accidents and parts

Aristotelian ontologies are committed to substances and their accidents.

But what about proper integral parts, like arms and legs?

These are typically not taken to be substances, because substances are not supposed to be parts of other substances. They seem to be a puzzling third category of things.

One set of puzzles comes from the connection between integral parts and accidents of the whole. For instance, the specific shape of a substance is usually considered an accident. But now imagine that I lose a leg and nothing else. Since I’ve lost nothing else, I haven’t lost my specific accident of shape, and hence, oddly, I have only one leg, but I have the same shape as before.

This is not quite as absurd as it may seem. After all, I could have the same shape I do and only one leg. I could, for instance, have one leg and a giant swelling shaped exactly like another leg. Still, there is something rather strange about the fact that mere loss of a leg causes the growth of a leg-shaped swelling.

Another set of puzzles comes from the accidents of the parts. A leg has its own shape, which is not a shape of the whole. Thus it seems that some accidents are not accidents of substances. This is a bit easier to handle. For we might say that a leg’s shape is an accident of the whole. However, the leg’s shape is not (as would be absurd) an accident of the whole being leg-shaped but an accident of the whole having a leg shaped thus-and-so. Thus, the specific shape accident of the leg makes the leg have that shape and makes the whole be legged in that shape.

Furthermore, suppose the leg changes its shape accident. How does that result in the whole automatically changing its shape accident to match? Or, conversely, how is it that when the whole is reshaped, so are the parts?

Much of the difficulty can be averted by denying that there are shape, size, mass, and other similar accidents of the whole. Instead, global shape, size, mass and similar facts about the whole are grounded in accidents of parts: shape, size, mass and location-in-the-whole accidents of the parts. This is a significant modification from traditional Aristotelianism, but perhaps it is the right move. Then if you remove an integral part, you remove all of its accidents, and the whole changes in the shape, size, mass, etc. that the part’s accidents helped ground.

This does not, however, solve the problem of having to posit a third category—proper integral parts—on top of substances and their accidents. I like the thought of solving this problem by identifying integral parts with special kinds of accidents—the heart with heartedness, etc.

Tuesday, November 10, 2026

The empirical theory triplet

We can think of an “empirical theory of the universe” as a triplet of three components:

  • physical law: the fundamental physical dynamics

  • initial conditions: the explanatorily fundamental boundary conditions

  • psychophysics: the fundamental psychophysical laws.

We can shift complexity back and forth between these three while maintaining empirical adequacy.

I’m going to simplify the stories I tell by supposing the psychophysics is epiphenomenalist. If it has two-way interaction, my stories will need to be more complicated.

In my previous post, I noted that you can make physical law extremely simple, without complicating psychophysics much, at the expense of complexity in the initial conditions.

But there are other ways of shifting. You could, for instance, entirely remove dynamics, physical law and initial conditions by adopting a Berkeleian idealism: all the complexity has been shifted into the psychophysics.

You can shift all complexity from initial conditions s0 to dynamics, simply by inserting an extra moment of time before the initial time t0 saying that by law anything at the extra moment transitions to s0 at t0.

It’s more challenging to figure out how to shift complexity from psychophysics to dynamics and/or initial conditions. The best line of approach seems to be to mathematically parametrize all the subjects’ conscious experiences. For instance, to a first approximation, the visual experience of a human could be parametrized by a function from a two-dimensional visual space to a four-dimensional color and depth space, and then the visual experience of n humans is parametrized by a function from two-dimensional visual space to 4n-dimensional space. We could then decompose the psychophysics function ψ(s,t) as ψ(s,t) = Ψ(ϕ(s,t)), where Ψ is a function from the space of mathematical parametrizations of all subjects’ conscious experiences to the actual conscious experiences of subjects, and then a function ϕ(s,t) from the physical world to mathematical parameterizations of experiences. For simplicity let’s suppose that the experiences of subjects are parametrized as vectors in some N-dimensional vector space.

We could then complicate the laws and initial conditions by adding an additional N-dimensional part of the universe containing a single sui generis particle whose location, by physical law, must equal ϕ(s,t) at time t whenever the rest of the universe is in state s at time t. Then we let ϕ*(s*,t) be the position of that sui generis particle when the whole universe is in state s*, and our psychophysical law will be that a universe in state s* at t has psychophysical state Ψ(ϕ*(s*,t)). Since ϕ* is quite simple, that means that we have likely significantly simplified the psychophysical laws. However, there is still the mapping Ψ from mathematical parametrizations to actual psychophysical states. Whatever mysterious complexities this mapping has—here the dragons of qualia lurk—we perhaps can’t simplify.

In any case, thinking about things in this way shows how difficult the data fitting problem is. We have three kinds of moving parts in our empirical theory, and we can shift complexity/tuning between them. Sometimes, a shift will involve a disproportionate rise in overall complexity/tuning, however. Thus, shifting quantum laws to a pseudorandom “digital physics” based on a Turing machine in the Game of Life is likely to result in a system whose overall complexity/tuning is much worse. This means that we have some hope for the project: if we are seeking a theory that optimizes overall complexity/tuning while maintaining empirical adequacy, some options are worse than others.

One way of thinking about the scientific enterprise, then, is that we are simply trying to find the empirically adequate triplet with the least overall complexity/tuning. While not everything is up for grabs, I still fear that at this point we have too many degrees of freedom due to complexity/tuning burden shifts between the elements of the triplet to have much hope of getting at the truth.

I think what we need is something more besides fit to qualia and complexity optimization across the triplet. Here is a suggestion: We need to have a certain limited degree of trust in the “manifest image”. There are multiple ways of working this out, but start with this observation. The idealistic option that folds all the complexity into the psychophysical is something we nearly universally (Berkeley excepted) consider to be a skeptical hypothesis. Similarly, some “digital physics” options look too much like a skeptical simulation hypothesis. It is part of the manifest image that skepticism is false. Thus, we can rule out certain extreme shifts within the triplet. The option that simplifies the psychophysical laws by adding an “experience particle” in a high-dimensional space and pushing the psychophysical law complexity into the physical dynamics also appears to have the flavor of a skeptical hypothesis.

Here is one way of working out the trust in the manifest image. We optimize not just for overall complexity of the triplet, but for overall complexity of the triplet and the distance of the theory from the manifest image, with the distance measured in a metric that weights more heavily the parts of the manifest image that are more deeply ingrained.

Here is another, in a way less flexible, approach. (The two could be combined.) We could take certain especially deeply engrained aspects of the manifest image to be non-negotiable constraints. For example, perhaps, we could take our embeddedness in something very much like three-dimensional space and time as non-negotiable.

Why should we have any trust in the manifest image? Perhaps because it is a part of human rational nature to do so. Perhaps it just is rationally required to trust it, a fundamental constraint on the rationality of dependent rational animals like us.

But there is another question. If we trust the manifest image, we think it has some degree of accuracy. But what explains that accuracy? Here, I think, theism is the main story.

Thursday, November 5, 2026

Simplifying the laws

This morning I realized the following Nomic Simplification thesis is very likely true:

  • NS: Any physical theory can be replaced by a practically empirically indistinguishable theory with extremely simple physical laws and very few natural kinds of physically fundamental entities and properties, at the cost of somewhat more complicated psychophysical laws and vastly more complicated and/or fine-tuned initial conditions. For concreteness, the simple physical laws could be the laws of Conway’s Game of Life or of classical mechanics with some simple laws.

What does the “practically” qualifier do here? Well, a theory might have empirical consequences that are not practically available to us, for instance because the mathematics of the theory is too nasty for us to compute the relevant predictions from. So I am restricting the empirical match to those empirical consequences that are practically available given the computational resources we have.

Moreover, NS seems to affect quite a number of issues. But before I get to that, let me give a hand-waving argument for NS that I find quite convicing.

By Church-Turing thesis, any such practically available empirical consequences can be computed by a finite subset of a Turing machine. Thus, we should be able to simulate the physical theory in question by means of a finite subset of a Turing machine, and a Turing machine needs only very simple physical laws. For instance, the laws of Conway’s Game of Life are extremely simple, and yet Turing complete—one can implement a Turing machine in the system. Moreover, we can run the Game of Life with one fundamental natural physical kind, namely cell, and one fundamental property, namely color, in addition to position. If we implement a Turing machine in the Game of Life that simulates the physical theory T in question, we will still need to layer psychophysical laws on top of that to generate empirical predictions. These might be somewhat more complicated than the best psychophysical laws for T, but not by far. For the translation between the Turing machine’s data and T’s physical ontology will not be that complicated (we can design a good GUI, as it were, by sufficiently complicating the code for the Turing machine—i.e., the initial conditions), and we can just add that translation layer to the best psychophysical laws for T. Now, we have good reason to think that any tenable psychophysical laws will be very complicated, and so adding the translation layer is unlikely to increase the complexity by a significant factor. Of course, the code encoding T is apt to be vastly complicated, and will correspond to the initial conditions for the Game of Life.

What goes for the Game of Life goes for classical mechanics with some simple laws—perhaps rigid bodies moving in classical mechanics frictionlessly and powered by flywheels that have been turning for ages.

What are some consequences of NS and of the Church-Turing argument behind it?

  1. What got me thinking about NS is that Nancy Cartwright has argued that our messy world cannot be encapsulated by simple laws of physics that cover all eventualities, given the complexities of combination laws for different forces, and has opted for a story involving causal powers. But insofar as that story generates practically available empirical consequences, NS tells us that we can replace the story with an account that has very simple physical laws.

  2. It is possible to uphold classical Newtonian physics given all the evidence we have at hand. For we can presumably make an arbitrarily large finite subset of a Turing machine in a world governed by classical physics. Of course, there will likely be horrendous fine-tuning of initial conditions.

  3. Sometimes theists argue (or at least I’ve said stuff like this) that laws of nature that would allow a complexly interdependent ecosystem, embodied intelligence, and yet make evil impossible would likely have to be immensely complex. NS shows this is false.

  4. We rightly prefer theories with simpler laws and fewer fundamental kinds and properties to theories with more of them other things being equal. NS teaches us, however, that simplicity of laws, fundamental kinds and fundamental properties can be easily ensured in ways that are counterintuitive. We should think it more likely that the fundamental laws, kinds and properties of our world are those of quantum mechanics and the standard model than that they are of Conway’s Game of Life. Thus, we need to make sure that we take “other things being equal” very seriously—in particular, avoidance of fine-tuning of initial conditions and/or seeking elegance in the initial conditions is very important for evaluating theories.

  5. If we are willing to accept a very large multiverse, we might be able to simplify the laws, fundamental kinds, and fundamental properties in an aberrant way (say, Game of Life) while yet not having any fine-tuning problem. For in a sufficiently large multiverse, we can explain the existence of fine-tuned initial conditions in some life-supporting universe by means of randomness and the fact that we can only observe a life-supporting universe. How serious a problem this is is depends on whether we can generate a good probability measures in multiverses and argue that given the aberrantly simple laws the right probability measure will generate predictions that do not match our observations. In the Game of Life story, a plausible probability measure is given by making the color of each initial cell be an independent coin-flip random variable. I suspect that this would lead to the prediction that we should expect to see a much messier world than we do, and so we may be able to empirically rule out a multiverse version of the Game of Life hypothesis.

The general lesson is this. We can shift complexity considerations in a theoretical account back and forth between initial conditions and the laws/kinds/properties. We can simplify each at the expense of the other.

Doing, refraining, and double effect

Some people are skeptical of distinctions between doing and refraining. It seems that anybody who accepts classical double effect reasoning must accept such a distinction.

Suppose you’re informed that if you don’t kill Alice, then Bob and Carol will be murdered, with all three innocent. The double effect reasoning says you may not kill Alice, since even if you are aiming to save Bob and Carol, the means to saving Bob and Carol is Alice’s death, and so you would be intending to kill Alice.

But suppose we don’t make a distinction between doing and refraining. Refraining from killing Alice has Bob and Carol’s death as a consequence. But this consequence is disproportionately bad, so by classical double effect reasoning, you are prohibited from refraining to kill Alice. Hence, killing Alice is wrong but so is refraining from killing her. Thus, this is a case of a real moral dilemma. But everyone who accepts classical double effect reasoning denies that this is a moral dilemma: you simply shouldn’t kill Alice, and then you’re doing nothing wrong.

So it seems that double effect cannot apply to both doings and refrainings equally. Thus, it seems that the right solution to the above problem is to say that double effect applies only to doings.

But there are cases where one might think double effect reasoning applies to refrainings as well. To adapt Rachels’ famous case, suppose innocent Tommy is drowning in a bathtub, and you can easily pull him out. But you’ve been informed that unless Tommy is dead by the end of the day, Bob and Carol will be murdered. Because it’s wrong to kill Tommy by double effect for doing, you were steeling yourself to accept Bob and Carol’s death. But now things have changed. Now it seems that double effect forbids you to save Tommy, because a consequence of saving Tommy is the death of Bob and Carol, and that’s disproportionate.

I am very unclear on this case. But maybe the judgment that you are forbidden to save Tommy is correct. Consider this case: you see someone drowning, and you could rescue them, but then someone points a gun at two hostages and informs you that if you save the drowning person, the hostages will be killed. Then you really should refrain from saving the drowning person.

Wednesday, November 4, 2026

Inseparability and real identity

On a standard (but disputed) interpretation of Scotus, x and y are inseparable (neither can consistently exist without the other) if and only if they really identical.

Let a be God’s asserting to Alice: “I am asserting to Bob the same thing I’m asserting to you but with your name in place of theirs.” Let b be God’s asserting to Bob: “I am asserting to Alice the same thing I’m asserting to you but with your name in place of theirs.” Then a and b are inseparable, since God cannot assert a falsehood. But clearly these two divine assertings are not really identical.

Saturday, October 31, 2026

Anthropocentrism

Theists are sometimes accused of being anthropocentric in their worldview.

Yet in a typical naturalistic worldview, rational animals are at the axiological pinnacle of reality, the only beings with dignity. And the only rational animals we know of are humans. (Though I suppose naturalisms that have a large multiverse are committed to an extremely high probability of there being other rational animals.)

The theist, on the other hand, thinks that rational animals like us are near the bottom of the great chain of being, since there is a finite distance between us and nothingness, and an infinite distance between us and the top of the chain, namely God.

It's wrong to disconnect from the violinist

Take Judith Jarvis Thomson’s famous violinist case. Here is a quick and simple argument for why it’s wrong to disconnect from the violinist in order not to be connected to them for nine months.

  1. It’s wrong to perform a positive action known to cause an innocent person’s death without a reason proportionate to the harm of death to someone who can be expected to live many years of a normal life otherwise.

  2. The harm of death to someone who can be expected to live many years of a normal life otherwise is clearly disproportionate to being free for nine months from being connected to someone.

  3. Disconnecting from the violinist is a positive action known to cause the death of an innocent person who can be expected to live many years of a normal life otherwise.

  4. Thus, it is wrong to disconnect from the violinist in order not to be connected to them for nine months.

Premise 1 is a consequence of most versions of the Principle of Double Effect, but is also independently extremely plausible. Premise 3 is part of the story.

Why believe premise 2? Here’s one criterion of disproportionateness. A sufficient condition for a harm H1 to be disproportionate to a harm H2 is that a reasonable person would with little hesitation accept a harm that’s twice as bad as H2 to avoid H1. But a reasonable person would with little hesitation accept 18 months of being connected to someone to save their life, assuming they can be expected to live many years of a normal life if saved.

To see this, imagine you are the violinist, and you are about to die of kidney problems. But then someone shows up who offers to hook themselves up to you for 18 months, after which you will be cured. You might hesitate whether to accept their generosity. But you wouldn’t hesitate much before concluding they are offering you a great benefit, and if you were to hesitate much to accept it, that would only be because you might be concerned about being so much in debt to the generous benefactor. The fact that you would have to live for 18 months hooked up to them, a quite significant burden admittedly, would be a price well worth paying for your life.

Or imagine this. You have just been conceived. An angel now miraculously gives you the intelligence of a reasonable adult, and informs you that you have a cancer that will kill you within a few days. However, the angel tells you that if you just ask, you will be cured of that cancer, and can expect a normal human life. But there is a price to be paid for the cure: For the next 9 months you will be confined to a small space within another human being, after which for the next 9 months you will be unable to walk, talk, or exercise any significant personal freedom. This is a price well worth paying for life! It is a price we have all paid involuntarily, but surely anyone who is fortunate enough to live a normal human life will think that price would have been worth paying voluntarily. And yet those 18 months of dependency and lack of freedom are more constraining than 18 months of being hooked up to the violinist would be.

I can now argue as follows:

  1. If it is wrong to disconnect from the violinist in the “standard scenario” above, abortion is wrong in the case of a normal pregnancy.

  2. Therefore, abortion is wrong in the case of a normal pregnancy.

Now, one could vary the violinist story so that the analogue of premise 2 will no longer hold. Perhaps if the nine months were filled with torture, or if the violinist were to live only for a few months after the nine month period was over, then the analogue of premise 2 wouldn’t hold. In those cases, the Principle of Double Effect might well allow disconnecting from the violinist, and in those cases, the above argument against abortion fails.

Does it follow that in such hard cases of pregnancy, an abortion would be permissible? It does not follow, since there are further wrong-making features in the abortion case, including: (a) the parent-child relationship, (b) the fact that the abortion requires a lethal direct attack on another human being’s body (the analogue to the “tubes” connecting one to the violinist in the case of pregnancy may be the umbilical cord, which is the child’s organ), and (c) the fact that living in utero constitutes ordinary life-support for someone in the first nine months of their life. But that would be material for another post.