Might machine learning hurt the machine?

Machine learning has the computer generate parameters for a neural network on the basis of a lot of data. Suppose that we think that computers can be conscious. I wonder if we are in a position, then, to know that any particular training session won’t be unpleasant for the computer. For we don’t really know what biological neural configurations, or transitions between them, constitute pain and other forms of unpleasantness. Maybe in the course of learning, among the vast number of changing network parameters or the updates between them there will be some that will hurt the computer. Perhaps it hurts, for instance, when the value of the loss function is high.

This means that if we think computers can be conscious, we may have ethical reasons to be cautious about artificial intelligence research, not because of the impact on people and other organisms in our ecosystem, but because of the possible impact on the computers themselves. We may need to first solve the problem of what neural states in animals constitute pain, so that we don’t accidentally produce functional isomorphs of these states in computers.

If this line of thought seems absurd, it may be that the intuition of absurdity here is some evidence against the thesis that computers can be conscious (and hence against functionalism).

Two open-ended cosmological arguments

First argument:

1. There are no infinite causal regresses or causal loops.

2. Every ordinary entity has a cause.

3. So, there is an extraordinary entity.

Second argument:

4. There is a causal explanation why there are any ordinary entities.

5. Causal explanations are not circular.

6. So, there is an extraordinary entity.

Yet another counterexample to Nicod's Principle

Nicod’s Principle says that the claim that all Fs are Gs is confirmed by each instance.

Here’s yet another counterexample. Consider the claim:

1. All unicorns are male.

We take this claim to be true, albeit vacuously so, since there are no unicorns.

But suppose an instance of (1), namely a male unicorn, were found. We would immediately conclude that (1) is probably false. For if there is a male unicorn, likely there is a female one as well.

The problem here is that when we learn of Sam that it is a male unicorn, we also learn that there are unicorns. And as soon as we learned that there are unicorns, that undercut the reason we had for believing (1), namely that we thought (1) was vacuously true.

Slavery, forced marriage and unjust laws

Slavery is the ownership of one person by another. Since a person no more owns another than a thief owns the purloined goods, there has never been any slavery. But of course there have been institutions thought to be slavery: institutions in which a person was thought to be the property of another. These were not institutions of slavery in the above strict sense but forms of unjust imprisonment, kidnapping, etc.

This seems to be merely a point about words, and a mistaken one at that. “Of course, Alexander II ended serfdom in Russia while Lincoln ended slavery in the US. Words mean what they are used to mean, and to dispute historical claims like these is to be like the fusty grammarian who claims that ‘It’s me’ is bad English.”

I agree that the question of words is unimportant. But here is what is important. Institutions are defined in large part by their norms. It is a defining feature of the norms of slavery (and, with some differences, of serfdom) that one person has property-style rights over another who has onerous obligations corresponding to these rights. But in fact, nobody has such rights over another, and the supposed obligations do not obtain. The institution that the “masters” saw themselves as a part of did not in fact exist, because the rights and obligations that they took to be integral to the institution did not, and could not, in fact exist.

We can use the term “slavery” (and cognates in other languages) for that non-existent institution, just as we use the term “phlogiston” for the substance that chemists mistakenly believed in before oxygen was discovered.

But we could also use distinguish and use two terms. Maybe slavery_h is the historical form of social organization that actually (and deplorably) existed and slavery_n is the normative institution that the masters_h (and maybe some of the slaves_h, as well) incorrectly thought to exist and thought to be coextensive with slavery_h.

Again, the words don’t matter, but it matters that there was a morally condemnable attempt to create a certain social institution which attempt failed because the norms that were attempted to be instituted were incapable of institution.

This is a pattern we find in many other cases. There is no such thing as a forced marriage, since the norms of love and sexuality that define marriage do not come into existence apart from the free consent of the parties. But of course over the course of history there have been morally condemnable attempts to force people—especially women—into the institution of marriage. These attempts always failed, and what the victims were forced into was a different institution, one subjecting them to such injustices as kidnapping, unjust imprisonment, rape, etc.

Thomas Aquinas, similarly, holds that there are no unjust laws. Of course, legislators may attempt to enact laws that would be unjust (or they may simply be exercising power and not even trying to legislate). But when they do so, they fail to enact laws. What they enact are mere demands masquarading as laws (philosophical anarchists think all “laws” are like that). Again, the question of words is unimportant, but what is important is the pattern: the legislator is deplorably attempting to create a social institution—a law—and failing to do so, but instead creating another institution.

The particular cases of this pattern are interesting, and so is the pattern itself. A central part of the pattern is an attempt to create an institution (or an instance of an institution) that misfires, and instead another institution is created that is widely but mistakenly thought to be the one that was the target of the attempt. But the cases of slavery, forced marriage and unjust laws also share another feature that not all the cases of misfiring do. For instance, suppose due to an honest mistake in the counting of ballots, there is a mistake as to who the mayor of a town is. The false mayor then attempts to legislate something quite just. The attempt fails, because the false mayor lacks the standing to legislate. But there need be nothing morally deplorable here, as there is in the slavery, forced marriage and unjust law cases.

Moreover, the three cases I started with are not just morally deplorable, but there seems to be an important connection between moral evil and performative misfire. Slavery_h is morally horrific, but slavery_n would be even worse, as the slaves_n would be under genuine obligations to do the enforced labor required of them and not to escape. This would, as it were, make morality itself complicit with the master, and the properly formed conscience of the slave into a whip in the master’s hand. And the same holds in the other two cases: morality itself would be a tool of oppression.

There are, alas, times when morality is a tool of oppression. The duties that exist between relatives are frequently exploited by repressive regimes as a means of social control: If you are an Uyghur or Tibetan defecting to a free country and speaking out against the Chinese regime, your relatives back home will suffer, and this restricts your activity because of the duties you have to your relatives. But the kinds of cases where the wicked use morality as a lever against the righteous seem different and less direct from what would be the case if slavery, forced marriage and unjust laws had the normative force that they pretend to. It is a mere coincidental effect of duties to family when these duties make it morally impossible or difficult to stand up to a wicked regime. But it would be of the very nature of the norms induced by slavery, forced marriage and unjust laws—if these norms really came into existence—that they would oppress.

This still leaves an interesting puzzle, which different moral theories will answer differently: Why is it the case that morality does not innately oppress?

Objection: Maybe slavery_h does create norms, but not moral norms.

Response: I myself don’t think there are any non-moral norms. But in any case slavery_h does not create any kind of obligation on the slave to obey the master, whether moral or not, except in some, but not all, cases a prudential one.

The search for new truths

I know that I have two hands. With a bit of thought, I now know a number of truths that it seems no ordinary person has ever known before:

• I have two hands or there is a palomino painted green and pink with someone in a Darth Vader costume on its back.

• I have two hands or the number of pigs born in 1745 is odd.

• I have two hands or Sir Patrick Stewart will consume a prime number of calories tomorrow.

• I have two hands or Donald Trump issued a series of anti-Klingon tweets yesterday.

And so on, ad infinitum. The search for new truths is thus really easy. I just need to search for silly propositions that no one has thought about, and disjoin them with something I know.

External time as such doesn't matter to us

Suppose a deity threatened to move us all to a universe where everything is pretty much as in our world, except that electric charges are reversed and the laws of nature are tweaked to ensure that this reverse doesn’t affect our lives. Thus, in that world, we are based not on carbon atoms, but on anti-carbon anti-atoms (they will have six anti-protons and six positrons, etc.), but the laws are tweaked so that the anti-atoms would behave just like atoms.

Assuming we can survive the shift, it seems that except for sentimental considerations (maybe when Grandma’s old wedding ring is replaced by a ring of anti-gold, it’s no longer the same ring) it would make no difference to us.

Similarly, if the deity threatened to spatially rotate the world by 180 degrees around some axis, that would make no difference to us.

What if the deity offered to rotate our world in time by 180 degrees, with causation now running temporally backwards, with us being born in the future and dying in the past, but everything being kept intact. It seems to me that this would make no difference to us.

Similarly, it seems to me that if the deity offered to rotate our four-dimensional world so that the temporal dimension and a spatial dimension were swapped, so that we would be born and die at the same time, but in different places along a spatial axis, and causation would run unidirectionally along the spatial axis, again that would make no difference to us.

I think these thought experiments suggest that external time as such is not important. What matters is how the distribution of things interacts with the causal order.

To be honest, though, I am not completely confident that any of these thought experiments make sense. It could be that any dimension along which causation runs much as causation runs along the temporal direction in our world is therefore a time dimension. But if so, then I think it's still true that it is causation, not external time as such, that matters.

I am less confident of this in the case of internal time.

Disembodied trees

Here’s an interesting thesis:

1. If x has the ys among its parts, and for each z among the ys, x can survive losing z without gaining anything, then x can survive simultaneously losing all the ys without gaining anything.

There are obvious apparent counterexamples. A boat that has sufficient redundancy can survive the loss of any plank, but cannot survive losing them all. An oak tree can lose any cell but cannot lose all cells.

But counterexamples aside, wouldn’t (1) be a nice metaphysical thesis to have? Then essential parts wouldn’t be made of inessential ones. You can see all the nasty ship-of-Theseus questions that would disappear if we had (1).

I think an Aristotelian can embrace (1), and can get around the counterexamples by biting some big bullets. First, like some contemporary Aristotelians, she can deny that artifacts like boats (or bullets) exist. Second, she can say that oak trees can survive the loss of all their matter, becoming constituted by form alone, much as some philosophers say happens to human beings after death (before the resurrection). The second part seems a bigger bullet to bite, as one would need a story as to why in fact oak trees perish when they lose all their cells, even if they don’t have to. But perhaps that’s just contingently how it happens, though an all powerful being could make an oak tree survive the destruction of all its cells.

The big question here is exactly what philosophical advantages embracing (1) has.

Distinguishing between properties

Some philosophers worry about “principles of individuation” that make two things of one kind be different from another. Suppose we share that worry. Then we should be worried about Platonism. For it is very hard to say what make two fundamental Platonic entities of the same sort different, say being positively charged from being negatively charged, or saltiness from sweetness.

However, the light-weight Platonist, who denies that predication is to be grounded in possession of universals, has a nice story to tell about the above kinds of cases. For here is a qualitative difference between saltiness and sweetness:

• saltiness is necessarily had by all and only salty things, but

• sweetness is not necessarily had by all and only salty things.

But for the heavy-weight Platonist to tell this story would involve circularity, for what it is for a thing to be salty will be to exemplify saltiness.

Of course, this story only works for properties that aren’t necessarily coextensive. But it’s some progress.

Acting without existing (any more)

Thesis: It is possible for an object to be acting while it does not exist.

Argument:

Imagine a rattlesnake that is ten light-years long, all stretched out. For all one hundred years of life it has been deliberately rattling its rattle. And then at the end of its hundred years, its head is destroyed, and I assume that the destruction of the head of a snake is sufficient for its death.

Rattling continues for at least about ten years even after the snake is dead, since the nerve signals the brain had sent while the snake was alive are continuing to rattle.

If this post-mortem rattle counts as the snake’s activity, the Thesis is established. But it is not clear that this ten years of post-mortem rattle is the snake’s activity.

But now consider the last year of pre-mortem rattling, call it R₉₉ (since it starts in year 99 of the snake’s life). Whatever one says of the post-mortem rattling, clearly R₉₉ is the snake’s activity. However, there is a reference frame—the way I set the length of the snake and the times in the story guarantees this—in which R₉₉ occurs after the snake’s head has been destroyed, and hence occurs after the death of the snake. But R₉₉ is the snake’s activity. Hence, there is a reference frame where an activity of the snake occurs after the snake is dead.

Scholium:

Obviously, only existent things can act. But while existence simpliciter is important for activity, existence-at-a-time does not have the same kind of significance. Obviously, often an actor’s action has a relationship R to some thing x that the actor itself does not have. For instance, an agent’s action may be known by me without the agent being known by me (here, R is being known and x is me).

Now, when we say that Elizabeth II exists as Queen of Canada, that is just an awkward way of saying that she has a monarchic relationship to Canada, rather than being a claim about that mysterious thing deep ontology studies: existence. I think we should think of existing-at-a-time as not really existence but simply as a particular kind of relationship—an occupation or presence relationship. It is not surprising in general that activities can stand in relationships that the agents do not. So, why can’t an activity stand in an occupation relationship to a time that the agent does not?

I think much confusion in philosophy comes from thinking of existence-at-a-time and existence-in-a-place as something special, somehow deeply ontologically different from other relations.

Reattachment of fingers and a new Aristotelian argument for presentism

Suppose Alice loses her finger at t₁ and at t₂ it is reattached. Intuitively, after t₂ she has the same finger as she had before t₁. But now suppose that at t₃ she loses her finger again, and it is reattached again at t₄. Then at t₁ and t₃ she is shedding the numerically same finger. Do the two sheddings result in the numerically same severed finger (which is a finger in name only)?

It seems that the answer is affirmative—if we were right in thinking the reattached finger to be the same finger as the original one. For there seems to be a symmetry between re-shedding and re-attachment as we could replace shedding by a transplant operation when a finger is moved back and forth between two people (Terry Pratchett’s Igors probably do that sort of thing for fun sometimes), after all.

But here’s something metaphysically odd. Call the finger F and the severed finger S. Then there is a major metaphysical difference between the first and the second severing. Let’s think about the difference assuming eternalism. Then the first severing causes a severed finger to exist simpliciter. But the second severing does not cause a severed finger to exist simpliciter, but only to come to exist at t₃. This is puzzling. In both cases, it seems that we have the same kind of cause, namely the severing of a finger, but the first time this has an ontic effect, a new being exists, and the second time it has no ontic effect. This seems wrong: the same kind of cause should have the same kind of effect, barring something indeterministic.

Maybe we could say that the finger’s coming to exist simpliciter is overdetermined by the severings. But this is counterintuitive. It shouldn’t be possible to add overdetermination to an effect already achieved, in the way that the second severing does. (Moreover, the overdetermination view conflicts with strong origins essentialism, which I accept, and the plausible counterfactual thesis that if the second severing didn’t happen, the very same severed finger S would have come into existence at t₁ as actually did. For by strong origins essentialism, if an object was overdetermined in its origination, it could not exist without being thus overdetermined. But then if the second severing didn’t happen, S wouldn’t have been overdetermined, so it couldn’t exist.)

So we have a puzzle for eternalism (and growing block, too). One could even take the above line of thought as a direct argument for presentism. Informally:

1. After reattachment, one has the same finger F as originally.
2. If after reattachment one has the same finger as originally, then each severing results in the same severed finger S.
3. The first severing causes S to exist simpliciter and presently.
4. The second severing only causes S to exist presently.
5. Both sheddings have the same kind of effect.
6. So, existing simpliciter must be nothing but existing presently.
7. So, presentism is true.

What should the eternalist (or growing blocker) say? It seems to me that the best move is to deny that both sheddings result in the same severed finger. The first results in S₁ and the second results in S₂ and S₁ ≠ S₂. By symmetry between re-shedding and re-attachment, I think we have to say that the reattached finger is numerically different from the original one, and deny (1). That is counterintuitive, but it seems the least costly response.

Objection: God could ensure that the reattached finger is the same as the original.

Respose: I think so. But that would be a miraculous intervention. And the symmetry would then require a similar miraculous intervention to ensure that the severed finger after the second shedding is the same as the severed finger after the first shedding. And this makes the second shedding causally different from the first, since no such miraculous intervention was needed to modify the first shedding. And with the two sheddings being different in kind, (5) will no longer be plausible.

The sacredness of the individual

There is a deontic prohibition on killing innocent people. But in general I think there is no similar deontic prohibition on destroying communities. For instance, there seems to be no deontic prohibition on a government dissolving a village or a city. Indeed, the reasons the state would need to have for such dissolution would have to be grave, but not outlandishly so. The state could permissibly intentionally dissolve a village or a city to end a war, but could not permissibly intentionally kill an innocent for the same end.

One might think this means that individuals are more valuable than the communities they compose. But we shouldn’t think that in general to be true, either. For instance, if a foreign invader were to threaten to dissolve a city without however killing anyone there, and the citizens could repel the invader at an expected cost of, say, six defenders’ and six attackers’ lives, it would be reasonable for the city to conscript its citizens to repel the invader. Thus the value of the shared life of the citizens is worth sacrificing some individual lives to uphold. But it is still not permissible to intentionally kill these innocent civilians.

I think it’s not that persons are more valuable than the villages and cities they compose, but rather they are sacred.

It is worth noting that where a community is sacred (two potential examples: sacramental marriages; God’s chosen people), there could very plausibly be a deontic prohibition on dissolution.

More and more I think the sacred is an ethical category, not just a theological one.

Vector display for Arduino-type boards

It's a nuisance to buy an LCD for each new Arduino project that requires a display, so I wrote an Android app that lets you use a tablet or phone as a display for an Arduino-type project. As a result, one can use a $2 board as a rudimentary oscilloscope with a phone or tablet providing the power, display and UI.

Instructions and links here.

Substantial change

The following seems to me to be a central tenet of classical Aristotelianism:

1. The identity of a parcel of matter is grounded in form.

But it seems to me that matter is introduced by Aristotelianism primarily to solve the problem of distinguishing (a) one substance changing into another (or into several others) from (b) one substance perishing and a new substance coming to be. The solution seems to be that in case (a) the matter persists, but not so in case (b).

But if the identity of a parcel of matter comes from form, then it is very puzzling indeed how a parcel of matter can remain selfsame while a change of form occurs. In other words, there is a tension between (1) and the motive for the introduction of matter into the ontology.

I am inclined to hold on to (1) in some sense, but reject the idea that matter solves the problem of substantial change.

Here is my currently best deflationary solution to the problem of substantial change. Certain kinds of causal interactions are described as “transfers of qualities”. For instance, when billiard ball A strikes billiard ball B in such a way that A stops moving and B begins moving, the momentum of A is “transferred” to B. However, we certainly do not want a metaphysics of momentum transfer on which there exists an entity m that previously was in A and later the numerically same m is present in B. That would be taking the talk of “transfer” too literally. Similarly, we talk of heat transfer.

I do not have an account of quality transfer, but a rough necessary condition for it is that there is a causal interaction where A causes B to gain a property that it itself is losing. There is an obvious difference between the momentum transfer story and the case where A is miraculously stopped by God from moving while B is simultaneously and miraculously set by God in motion.

Now, a special case of quality transfer is when a causal interaction not only transfers a quality but also creates one or more new substances. For instance, suppose a gecko chased by a predator drops its tail, whose writhing confuses the predator. In doing so, the gecko transfers some of its mass, extension, color, motion and other qualities to a new substance (or aggregate of substances), a substance that comes to exist as a result of the same causal interaction.

The technical neo-Aristotelian term for the gecko’s loss of its tail is excretion. Excretion comes in two sorts. The kind of excretion in the case of the gecko’s autotomy is productive excretion, where qualities, notably including mass and extension (understood broadly to include temporal extension for aspatial temporal things), are transfered to one or more substances that are produced in the same causal interaction. Another kind of excretion is accretive excretion, where qualities are transferred to one or more substances that exist independently of this causal interaction. For instance, if an animal were to swallow a living plant, perhaps the plant in the animal’s digestive system could be accretively excreting: its qualities, notably including mass and extension, would come to be gradually lost to the plant while gained by the animal. (This is a bit more complicated in real life, I expect: I doubt the digested bits immediately become parts of the animal.)

Substantial corruption of a material substance, then, is total excretion, a causal interaction where all of a substance’s extension and mass is excreted to one or more substances. This comes in two basic varieties: substantial change where the the beneficiary substances result from the same causal interaction and accretive substantial corruption where the beneficiary substances exist independently of this causal interaction (and typically are preexistent). And one can have a combination case where some of the beneficiaries result from the interaction and some are not dependent on it.

But there is nothing metaphysically deep about substantial corruption.

Possibly giving a finite description of a nonmeasurable set

It is often assumed that one couldn’t finitely specify a nonmeasurable set. In this post I will argue for two theses:

1. It is possible that someone finitely specifies a nonmeasurable set.

2. It is possible that someone finitely specifies a nonmeasurable set and reasonably believes—and maybe even knows—that she is doing so.

Here’s the argument for (1).

Imagine we live an uncountable multiverse where the universes differ with respect to some parameter V such that every possible value of V corresponds to exactly one universe in the multiverse. (Perhaps there is some branching process which generates a universe for every possible value of V.)

Suppose that there is a non-trivial interval L of possible values of V such that all and only the universes with V in L have intelligent life. Suppose that within each universe with V in L there runs a random evolutionary process, and that the evolutionary processes in different universes are causally isolated of each other.

Finally, suppose that for each universe with V in L, the chance that the first instance of intelligent life will be warm-blooded is 1/2.

Now, I claim that for every subset W of L, the following statement is possible:

3. The set W is in fact the set of all the values of V corresponding to universes in which the first instance of intelligent life is warm-blooded.

The reason is that if some subset W of L were not a possible option for the set of all V-values corresponding to the first instance of intelligent life being warm-blooded, then that would require some sort of an interaction or dependency between the evolutionary processes in the different universes that rules out W. But the evolutionary procesess in the different universes are causally isolated.

Now, let W be any nonmeasurable subset of L (I am assuming that there are nonmeasurable sets, say because of the Axiom of Choice). Then since (3) is possible, it follows that it is possible that the finite description “The set of values of V corresponding to universes in which the first instance of intelligent life is warm blooded” describes W, and hence describes a nonmeasurable set. It is also plainly compossible with everything above that somebody in this multiverse in fact makes use of this finite description, and hence (1) is true.

The argument for (2) is more contentious. Enrich the above assumptions with the added possibility that the people in one of the universes have figured out that they live in a multiverse such as above: one parametrized by values of V, with an interval L of intelligent-life-permitting values of V, with random and isolated evolutionary processes, and with the chance of intelligent life being warm-blooded being 1/2 conditionally on V being in L. For instance, the above claims might follow from particularly elegant and well-confirmed laws of nature.

Given that they have figured this out, they can then let “Q” be an abbreviation for “The set of all values of V corresponding to universes wehre the first instance of intelligent life is warm-blooded.” And they can ask themselves: Is Q likely to be measurable or not?

The set Q is a randomly chosen subset of L. On the standard (product measure) understanding of how to probabilistically make sense of this “random choice” of subset, the event of Q being nonmeasurable is itself nonmeasurable (see the Sawin answer here). However, intuitively we would expect Q to be nonmeasurable. Terence Tao shares this intuition (see the paragraph starting “Intuitively”). His reason for the intuition is that if Q were measurable, then by something like the Law of Large Numbers, we would expect the intersection of Q with a subinterval I of L to have a measure equal to half of the measure of I, which would be in tension with the Lebesgue Density Theorem. This reasoning may not be precisifiable mathematically, but it is intuitively compelling. One might also just have a reasonable and direct intuition that the nonmeasurability is the default among subsets, and so a “random subset” is going to be nonmeasurable.

So, the denizens of our multiverse can use these intuitions to reasonably conclude that Q is nonmeasurable. Hence, (2) is true. Can they leverage these intuitions into knowledge? That’s less clear to me, but I can’t rule it out.

Simultaneous and diachronic causation

The main problem with the idea that all causation is simultaneous is to make sense of the obvious fact of diachronic causation, as when setting an alarm in the evening causes it to go off in the morning. Here is a theory that has both simultaneity and diachronicity that bears further examination:

• All causation between substances is simultaneous

• There is diachronic causation within a substance.

We now have a model of how setting the alarm works, on the simplifying assumption that the alarm clock is a substance. In the evening, by simultaneous causation, I cause the clock to have a certain state. A sequence of diachronic causal interactions within the clock—accidents of the clock causing other accidents of the clock, say—then causes the alarm to go off. The alarm’s going off then, by means of simultaneous causation between substances, causes particles in the air to move, etc. In other words, the diachronicity of the causation is all internal to the substances.

An even more interesting theory would hold that:

• All causation between substances is simultaneous

• All causation within a substance is diachronic.

If we were willing to swallow this, then we would have a very elegant account of the internal time of a substance as constituted by the causal relations within the substance (presumably, the causal relations between the accidents of the substance).