Algorithmic priors and human nature

One promising way to define priors is with algorithmic probability, such as Solomonoff priors. The idea is that we have a language L (say, one based on Turing machines), and we imagine generating random descriptions in L in a canonical way. E.g., add an end-of-string symbol to L and randomly and independently generate symbols until you hit the end-of-string symbol, and then conditionalize on the string uniquely describing a situation, and take the probability of a specific situation s to be the probability of that a random description so generated describes s.

These kinds of priors are rather appealing for science, since they appear to be induction-friendly, as they assign high probabilities to compressible—more briefly expressible—situations. Thus, if our situations are distributions of color among ravens, monochromatic distributions get much higher probability as they can be much more briefly described, like ∀x(B(x)) or ∀x(W(x)).

Philosophically, I think the big problem is with the choice of the language. It would be nice if we could let L be a language that cuts nature exactly at the joints. But we don’t know that language. And absent that language, we need something arbitrary.

Here is a particular version of the problem. Take Kuhn’s division of science into ordinary and revolutionary science. One aspect of this division is that in ordinary science, we have a scientific language, and are discovering things within it. In that case, it is reasonable to take L to be that language. However, when we are doing revolutionary science and creating new paradigms, we cannot do that. The new paradigms either cannot be described in the old language or their description is unwieldy in a way that does not do justice to the plausibility of the new paradigm. Indeed, much of the point of revolutionary science is to create a language within which the description of the world is simpler, and then argue that this language is therefore more likely to cut nature at the joints.

Another version of this problem is what language L we choose when we are generating fundamental priors. Practically speaking, we cannot use a scientific language that cuts nature at the joints, because we have not yet discovered it. If this was merely a practical concern, we could try to say that this doesn’t matter: the fundamental priors are ones that we ought to have rather than any that we actually have or could have—perhaps ought does not imply can. But the concern is not merely practical. For one of the main points of our inductive reasoning is to discover what concepts cut nature at the joints, and this is largely an empirical enterprise. If the right fundamental priors were to reflect the joints in nature, then the enterprise wouldn’t make much sense, as we would be obligated to have already completed much of the enterprise before we started it.

So, I think, we have to say L does not always cut nature at the joints, and yet this generating appropriate priors for us. But we still need a constraint on L. After all, we could imagine a language that thwarts our empirical enterprise, such as one where only fairies can be described briefly and anything else requires very long descriptions, so we have very high priors for fairies and very low priors for everything else. What will be the constraint? Practically, we pretty much have to start with some ordinary human language. I think our ideal should not be far from what is practical. Thus, I propose, if we are going to go with algorithmic priors, we should choose L to be a language that fits well with our human nature as communicators. This is an anthropocentric choice, and I think human epistemology is rightly anthropocentric.

But why think that the anthropocentric choice is apt to lead to truth? There are two stories to be told here. First, it may be that human nature requires a measure of trust in human nature. Second, that trust is vindicated if we are created by a good God who loves the truth.

Two decreases in tension between faith and science

Over the past two hundred years or so, one new tension point arose for the relationship between Christianity and science due to scientific progress—namely, evolution. At the same time, several tension points disappeared due two other instances of scientific progress.

The first instance of this scientific progress was the general abandonment of the Aristotelian eternal world model of the universe with Big Bang cosmology. In the middle ages, Jewish, Islamic and Christian thinkers struggled with the tension between the science/philosophy of the day strongly tending towards a universe that always existed and the theological commitment to a creation a finite amount of time ago. That problem is gone.

The second instance is our scientific understanding of the continuity of organic development from zygote to embryo to infant to adult, which has made quite implausible the old view of discontinuous transition in utero from vegetable to animal to human. This old view was the dominant scientific view of human origins until fairly recently, and it had serious tensions with Christian theology.

The first of these embryological tensions was with Christian moral views about abortion. While traditionally Christians opposed both contraception and abortion, abortion was morally seen as a form of homicide. But on the discontinuous transition view, abortion prior to human ensoulment would only be contraception.

The second embryological tension was a technical problem in Christology. Suppose that in the Incarnation we have the vegetable, mere animal and rational animal sequence. Then Aquinas observes there are two possibilities, neither of which is theologically appealing.

First, it could be that God becomes incarnate as a vegetable or a mere animal. But this seems, as Aquinas says, “unbecoming”. And he seem to be right. The Incarnation reveals to us the person of the Logos, and it would be unbecoming that the Logos become a non-personal being.

Second, it could be that the Incarnation happens only at the beginning of the third stage of development, namely once everything is ready for a rational animal. But then Aquinas says “the whole conception could not be attributed to the Son of God”. Indeed, don’t we even have a tension with the Apostles’ Creed line that Christ “was conceived by the Holy Spirit”? For on this option, Christ was not conceived at all. What was conceived was a vegetable, not Christ. (Indeed, none of us were conceived on this view.) Moreover, one might worry that then there would be a sense in which the flesh of Christ would pre-exist the Incarnation. And that makes it difficult to say that the Word became flesh—for the flesh that Scripture says he “became” would already in a sense have been there, and one can’t become this flesh, since this flesh already has its own identity. (Granted, there may well be some Aristotelian metaphysics one can do to lessen this last worry.)

Aquinas solves the problem by supposing that Christ is conceived fully formed in Mary’s womb, and hence has the rational soul from the first moment of his existence. But this solution is itself problematic. Absent gradual development from a zygote, is this conception at all? If God were to create an adult human either ex nihilo or out of some pre-existing matter, we would not consider that a conception. But neither should we then consider it a conception if God creates a fully-formed fetus, even if he does that out of the pre-existing matter of Mary. So we still have a problem with the Apostles’ creed’s “was conceived by the Holy Spirit”. Moreover, it seems that this deprives Mary of a significant chunk of her motherhood.

But the problem entirely disappears once we think that the human beings begin their existence at conception. Christ is conceived by the Holy Spirit, presumably in that Mary’s ovum is transformed into a zygote by the infinite power of the Holy Spirit, which zygote is the Christ who then grows in utero like we all do.

(Catholics also note that the new scientific understanding of human embryonic development also helps with the doctrine of Mary’s immaculate conception—for only a rational being can be immaculately conceived, since original sin or freedom from it can only apply to a rational being.)

Comparing experiments

When you’re investigating reality as a scientist (and often as an ordinary person) you perform experiments. Epistemologists and philosophers of science have spent a lot of time thinking about how to evaluate what you should do with the results of the experiments—how they should affect your beliefs or credences—but relatively little on the important question of which experiments you should perform epistemologically speaking. (Of course, ethicists have spent a good deal of time thinking about which experiments you should not perform morally speaking.) Here I understand “experiment” in a broad sense that includes such things as pulling out a telescope and looking in a particular direction.

One might think there is not much to say. After all, it all depends on messy questions of research priorities and costs of time and material. But we can at least abstract from the costs and quantify over epistemically reasonable research priorities, and define:

1. E₂ is epistemically at least as good an experiment as E₁ provided that for every epistemically reasonable research priority, E₂ would serve the priority at least as well as E₁ would.

That’s not quite right, however. For we don’t know how well an experiment would serve a research priority unless we know the result of the experiment. So a better version is:

2. E₂ is epistemically at least as good an experiment as E₁ provided that for every epistemically reasonable research priority, the expected degree to which E₂ would serve the priority is at least as high as the expected degree to which E₁ would.

Now we have a question we can address formally.

Let’s try.

3. A reasonable epistemic research priority is a strictly proper scoring rule or epistemic utility, and the expected degree to which an experiment would serve that priority is equal to the expected value of the score after Bayesian update on the result of the experiment.

(Since we’re only interested in expected values of scores, we can replace “strictly proper” with “strictly open-minded”.)

And we can identify an experiment with a partition of the probability space: the experiment tells us where we are in that partition. (E.g., if you are measuring some quantity to some number of significant digits, the cells of the partition are equivalence classes under equality of the quantity up to those many significant digits.) The following is then easy to prove:

Proposition 1: On definitions (2) and (3), an experiment E₂ is epistemically at least as good as experiment E₁ if and only if the partition associated with E₂ is essentially at least as fine as the partition associated with E₁.

A partition R₂ is essentially at least as fine as a partition R₁ provided that for every event A in R₁ there is an event B in R₂ such that with probability one B happens if and only if A happens. The definition is relative to the current credences which are assumed to be probabilistic. If the current credences are regular—all non-empty events have non-zero probability—then “essentially” can be dropped.

However, Proposition 1 suggests that our choice of definitions isn’t that helpful. Consider two experiments. On E₁, all the faculty members from your Geology Department have their weight measured to the nearest hundred kilograms. On E₂, a thousand randomly chosen individiduals around the world have their weight measured to the nearest kilogram. Intuitively, E₁ is better. But Proposition 1 shows that in the above sense neither experiment is better than the other, since they generate partitions neither of which is essentially finer than the other (the event of there being a member of the Geology Department with weight at least 150 kilograms is in the partition of E₂ but nothing coinciding with that event up to probability zero is in the partition of E₁). And this is to be expected. For suppose that our research priority is to know whether any members of your Geology Department are at least than 150 kilograms in weight, because we need to know if for a departmental cave exploring trip the current selection of harnesses all of which are rated for users under 150 kilograms are sufficient. Then E₁ is better. On the other hand, if our research priority is to know the average weight of a human being to the nearest ten kilograms, then E₂ is better.

The problem with our definitions is that the range of possible research priorities is just too broad. Here is one interesting way to narrow it down. When we are talking about an experiment’s epistemic value, we mean the value of the experiment towards a set of questions. If the set of questions is a scientifically typical set of questions about human population weight distribution, then E₁ seems better than E₂. But if it is an atypical set of questions about the Geology Department members’ weight distribution, then E₂ might be better. We can formalize this, too. We can identify a set Q of questions with a partition of probability space representing the possible answers. This partition then generates an algebra F_Q on the probability space, which we can call the “question algebra”. Now we can relativize our definitions to a set of questions.

4. E₂ is epistemically at least as good an experiment as E₁ for a set of questions Q provided that for every epistemically reasonable research priority on Q, the expected degree to which E₂ would serve the priority is at least as high as the expected degree to which E₁ would.

5. A reasonable epistemic research priority on a set of questions Q is a strictly proper scoring rule or epistemic utility on F_Q, and the expected degree to which an experiment would serve Q is equal to the expected value of the score after Bayesian update on the result of the experiment.

We recover the old definitions by being omnicurious, namely letting Q be all possible questions.

What about Proposition 1? Well, one direction remains: if E₂’s partition is essentially at least as fine as E₁’s, then E₂ is better with regard any set of questions, an in particular better with regard to Q. But what about the other direction? Now the answer is negative. Suppose the question is what the average weight of the six members of the Geology Department is up to the nearest 100 kg. Consider two experiments: on the first, the members are ordered alphabetically by first name, and a fair die is rolled to choose one (if you roll 1, you choose the first, etc.), and their height is measured. On the second, the same is done but with the ordering being by last name. Assuming the two orderings are different, neither experiment’s partition is essentially at least as fine as the other’s, but the expected contributions of both experiments towards our question is equal.

Is there a nice characterization in terms of partitions of when E₂ is at least as good as E₁ with regard to a set of questions Q? I don’t know. It wouldn’t surprise me if there was something in the literature. A nice start would be to see if we can answer the question in the special case where Q is a single binary question and where E₁ and E₂ are binary experiments. But I need to go for a dental appointment now.

Animal experimentation

We have an intuitive line as to where the suffering in animal’s life is so great compared to the goods in it that when we are able to do so, we euthanize animals when the suffering causes the value of the animal’s life to fall below that line. On the other hand, the life of an animal that falls above this line is a life that is a benefit to the animal.

It seems to me that this intuitive line could be a helpful discernment criterion for animal experimentation. Animals that are used for experiments are often bred for that purpose. Thus, they wouldn’t exist absent the practice of experimentation. It seems, hence, that experiments where the stresses on the animals make the animal’s life fall below this intuitive line are very easy to justify: the animals are benefitted by the practice, even if the experiments do impose some suffering on the animal. It seems plausible that such experiments could thus be justified by the intrinsic value of the knowledge gained for its own sake, or even by pedagogical benefits for students.

On the other hand, if the stresses in the animal’s life are such as to make their life fall below the line, then stronger justification is needed: the prospective benefits of the research need to be rather more significant.

I don’t know how good we are at discerning where that line goes. People with pets and farm animals do make hard decisions about this, though, so we seem to have some epistemic access to the line.

(I do think that it is permissible to be much more utilitarian about animal life than about human life. I certainly would not generalize what I say above to the case of humans.)

Beauty and simplicity in equations

Often, the kind of beauty that scientists, and especially physicists, look for in the equations that describe nature is taken to have simplicity as a primary component.

While simplicity is important, I wonder if we shouldn’t be careful not to overestimate its role. Consider two theories about some fundamental force F between particles with parameters α₁ and α₂ and distance r between them:

1. F = 0.8846583561447518148493143571151840833168115852975428057361124296α₁α₂/r²

2. F = 0.88465835614475181484931435711518α₁α₂/r^{2 + 2−64}.

In both theories, the constants up front are meant to be exact and (I suppose) have no significantly more economical expression. By standard measures of simplicity where simplicity is understood in terms of the brevity of expression, (2) is a much simpler theory. But my intuition is that unless there is some special story about the significance of the 2 + 2⁻⁶⁴ exponent, (1) is the preferable theory.

Why? I think it’s because of the beauty in the exponent 2 in (1) as opposed to the nasty 2 + 2⁻⁶⁴ exponent in (2). And while the constant in (2) is simpler by about 106 bits, that additional simplicity does not make for significantly greater beauty.

Literature and science

I think we learn at least as much about ourselves as persons from literature as from science. This is surprising if physicalism is true.

Scoring rules and publication thresholds

One of the most problematic aspects of some science practice is a cut-off, say at 95%, for the evidence-based confidence needed for publication.

I just realized, with the help of a mention of p-based biases and improper scoring rules somewhere on the web, that what is going on here is precisely a problem of a reward structure that does not result in a proper scoring rule, where a proper scoring rule is one where your current probability assignment is guaranteed to have an optimal expected score according to that very probability assignment. Given an improper scoring rule, one has a perverse incentive to change one’s probabilities without evidence.

To a first approximation, the problem is really, really bad. Insofar as publication is the relevant reward, it is a reward independent of the truth of the matter! In other words, the scoring rule has a reward for gaining probability 0.95 (say) in the hypothesis, regardless of whether the hypothesis is true or false.

Fortunately, it’s not quite so bad. Publication is the short-term reward. But there are long-term rewards and punishments. If one publishes, and later it turns out that one was right, one may get significant social recognition as the discoverer of the truth of the hypothesis. And if one publishes, and later it turns out one is wrong, one gets some negative reputation.

However, notice this. Fame for having been right is basically independent of the exact probability of the hypothesis one established in the original paper. As long as the probability was sufficient for publication, one is rewarded for fame. Thus if it turns out that one was right, one’s long-term reward is fame if and only if one’s probability met the threshold for publication and one was right. And one’s penalty is some negative reputation if and only if one’s probability met the threshold for publication and yet one was wrong. But note that scientists are actually extremely forgiving of people putting forward evidenced hypotheses that turn out to be false. Unlike in history, where some people live on in infamy, scientists who turn out to be wrong do not suffer infamy. At worst, some condescension. And it barely varies with your level of confidence.

The long-term reward structure is approximately this:

• If your probability is insufficient for publication, nothing.

• If your probability meets the threshold for publication and you’re right, big positive.

• If your probability meets the threshold for publication and you’re wrong, at worst small negative.

This is not a proper scoring rule. It’s not even close. To make it into a proper scoring rule, the penalty for being wrong at the threshold would need to be way higher than the reward for being right. Specifically, if the threshold is p (say 0.95), then the ratio of reward to penalty needs to be (1−p) : p. If p = 0.95, the reward to penalty ratio would need to be 1:19. If p = 0.99, it would need to be a staggering 1:99, and if p = 0.9, it would need to be a still large 1:9. We are very, very far from that. And when we add the truth-independent reward for publication, things become even worse.

We can see that something is problematic if we think about cases like this. Suppose your current level of confidence is just slightly above the threshold, and a graduate student in your lab proposes to do one last experiment in her spare time, using equipment and supplies that would otherwise go to waste. Given the reward structure, it will likely make sense for you to refuse this free offer of additional information. If the experiment favors your hypothesis, you get nothing out of it—you could have published without it, and you’d still have the same longer term rewards available. But if the experiment disfavors your hypothesis, it will likely make your paper unpublishable (since you were at the threshold), but since it’s just one experiment, it is unlikely to put you into the position of yet being able to publish a paper against the hypothesis. At best it loses you the risk of the small negative reputation for having been wrong, and since that’s a small penalty, and an unlikely one (since most likely your hypothesis is true by your data), so that’s not worth it. In other words, the the structure rewards you for ignoring free information.

How can we fix this? We simply cannot realistically fix it if we have a high probability threshold for publication. The only way to fix it while keeping a high probability threshold would be by having a ridiculously high penalty for being wrong. But we shouldn’t do stuff like sentencing scientists to jail for being wrong (which has happened). Increasing the probability threshold for publication would only require the penalty for being wrong to be increased. Decreasing probability thresholds for publication helps a little. But as long as there is a larger reputational benefit from getting things right than the reputational harm from getting things wrong, we are going to have perverse incentives from a probability threshold for publication bigger than 1/2, no matter where that threshold lies. (This follows from Fact 2 in my recent post, together with the observation that Schervish’s characterization of scoring rules shows implies that any reward function corresponds to a unique up to additive constant penalty function.)

What’s the solution? Maybe it’s this: reward people for publishing lots of data, rather than for the data showing anything interestingly, and do so sufficiently that it’s always worth publishing more data?

Physicalism and the progress of science

People sometimes use the progress of science to argue for physicalism about the mind. But it seems to me that Dostoevskii made more progress in understanding the human mind by existential reflection than anybody has by studying the brain directly. More generally, if we want to understand human minds, we should turn to literature and the spiritual masters rather than to neuroscience.

Thus, any argument for physicalism about the mind from the progress of science is seriously flawed. And perhaps we even have some evidence against physicalism. For it is a surprising fact that we learn more about the mind by the methods of the humanities than by study of the brain if the mind is the brain.

Samuel Clarke on our ignorance of the essence of God

At times I am made uncomfortable by this objection to arguments for the existence of God: there feels like there is something fishy about inferring the existence of a being about which we know so very little. It may be that theism is the only reasonable explanation of the universe’s existence, but if we know so very little about that explanation, can the inference to the truth of that explanation be a genuine version of inference to best explanation?

Newton's disciple Samuel Clarke has a nice answer to this objection:

There is not so mean and contemptible a plant or animal, that does not confound the most enlarged understanding upon earth; nay, even the simplest and plainest of all inanimate beings have their essence or substance hidden from us in the deepest and most impenetrable obscurity.

In other words, all our ordinary day-to-day inferences are to things whose essence is hidden.

It may be thought that now that we know about DNA, we do know the essences of plants and animals. But even if that is true, which I am sceptical of, it doesn’t matter: for belief in plants and animals was quite reasonable even before our superior science. And even this day, our knowledge of the essences of the fundamental entities of physics (e.g., particles, fields, wavefunctions) is basically nil. All we know is some facts about the effects of these entities.

A pedagogical universe

Our science developed over milennia, progressing from false theory to less false theory. Why did we not give up long ago? I take it this is because the false theories, nonetheless, had rewards associated with them: although false, they allowed for prediction and technological control in ways that were useful (in a broad sense) to us.

Thus, the success of our science depends not just on a “uniformity of nature” on which the correct fundamental scientific theories are elegant and uniform. Most of our historical progress in physics has not involved correct scientific theories—and quite possibly, we do not have any correct fundamental theories in physics yet. The success of our science required low-hanging fruit for us to pick along the way, fruit that would guide us in the direction of truth.

We can imagine worlds where the ultimate physics requires an enormous degree of sophistication (much as we expect to be the case in our world) and there is little in the way of low-hanging fruit (except maybe for the lowest level of low-hanging fruit, involving the regularities needed to enable evolution of intelligence in the first place) in the form of approximately true theories that rewards us with prediction and control so that beings like us would just give up on science. Our world is better than that.

Indeed, our world seems to be pedagogically arranged for us, arranged to gradually teach us science (and other things), much as we teach our children, with intellectual and practical rewards. There is a design argument for the existence of God from this (closely related to this one).

Pain and water

One way for physicalists to handle the apparent differences between mental and physical properties is to liken the difference to that between water and H₂O. It is a surprising a posteriori fact that water is H₂O. Similarly, it is a surprising a posteriori fact that pain is physical state ϕ₁₃₅ (say).

Now, a posteriori facts are facts that are knowable by observation. But it is not clear that the proposition that pain is physical state ϕ₁₃₅ is knowable by observation.

Here is why. There are two main candidates for what kind of a state ϕ₁₃₅ could be: a brain state or a functional state. The choice between these two candidates depends on how strongly one feels about multiple realizability of mental states. If one is willing to say that only beings with brains like ours—say, complex vertebrates—feel pain, one might identify ϕ₁₃₅ with a brain state. If one has a strong intuition that beings with other computational systems anatomically different from those of complex vertebrates—cephalopods, aliens, and robots—could have consciousness, one will opt for identifying ϕ₁₃₅ as a functional state.

But in fact, assuming pain is a physical state, there is a broad spectrum of physical state candidates for identifying pain with, depending on how far we abstract from the actual physical realizers of our pains while keeping fixed the broad outlines of functionality (signaling damage and leading to aversive behavior). If we abstract very little, only brain states found in humans—and perhaps not all humans—will be pain. If we abstract a bit more, but still insist on anatomical correspondence, then brain states found in other complex vertebrates will be pain. If we drop the insistence on anatomical correspondence but do not depart too far, we may include amongst the subjects of pain other DNA-based organisms such as cephalopods. Further abstraction will let in living organisms with other chemical bases, and yet further abstraction will let in robots. And even when talking of the fairly pure functionalism applicable to robots, we will have serious questions about how far to abstract concepts such as “damage” and “aversive behavior”.

The question of where in this spectrum of more and more general physical states we find the state that is identical with pain does not appear to be a question to be settled by observation. By internal observation, we only see our own pain. By external observation, however, we cannot tell where in the spectrum of more and more general (perhaps along multiple dimensions) physical states pain is present, without begging the question (e.g., by assuming from the outset that certain behaviors show
the presence of pain, which basically forces our hand to a functionalism centered on those behavior).

Objection 1: An experimenter could replace the brain structures responsible for pain in her own brain by structures that are further from human ones, and observe whether she can still feel pain. Where the feeling of pain stops, there we have abstracted too far.

Response: There are serious problems with this experimental approach. First, mere replacement of brain pain centers will not allow one to test hypotheses on which what constitutes pain depends on the larger neural context. And replacement of the brain as a whole is unlikely to result in the experimenter surviving. Second, and perhaps more seriously, if replacements of the brain pain centers commit the same data to memory storage as brain pain centers do, after the experiment the agent will think that there was pain, even if there wasn’t any pain there, and if they have the same functional influence on vocal production as brain centers do, the agent will report pain, again even if there wasn’t any pain there.

Objection 2: We could know which physical state pain is identified with if God told us, and being told by God is a form of a posteriori knowledge.

Response: It seems likely that God’s knowledge of which physical states are pains, or of the fact that water is H₂O, would be a priori knowledge. God doesn’t have to do scientific research to know necessary truths.

Objection 3: We can weaken the analogy and say that just as the identity between water and H₂O is not a priori, so too the identity between pain and ϕ₁₃₅ is not a priori, without saying that both are a posteriori.

Response: This is probably the move I’d go for if I were a physicalist. But by weakening this analogy, one weakens the position that it defends. For it is now admitted that there is a disanalogy between water-H₂O and pain-ϕ₁₃₅. There is something rather different about the mental case.

Mystery and religion

Given what we have learned from science and philosophy, fundamental aspects of the world are mysterious and verge on contradiction: photons are waves and particles; light from the headlamp on a fast train goes at the same speed relative to the train and relative to the ground; objects persist while changing; we should not murder but we should redirect trolleys; etc. Basically, when we think deeper, things start looking strange, and that’s not a sign of us going right. There are two explanations of this, both of which are likely a part of the truth: reality is strange and our minds are weak.

It seems not unreasonable to expect that if there were a definitive revelation of God, that revelation would also be mysterious and verge on contradiction. Of the three great monotheistic religions, Christianity with the mystery of the Trinity is the one that fits best with this expectation. At the same time, I doubt that this provides much of an argument for Christianity. For while it is not unreasonable to expect that God’s revelation would be paradoxical, it is a priori a serious possibility that God’s revelation might be so limited that what was revealed would not be paradoxical. And it would also be a priori a serious possibility that while creation is paradoxical, God is not, though this last option is a posteriori unlikely given what we learn from the mystical experience traditions found in all the three monotheistic religions.

So, I am not convinced that there is a strong argument for Christianity and against the other two great monotheistic religions on the grounds that Christianity is more mysterious. But at least there is no argument against Christianity on the basis of its embodying mysteries.

Constructive empiricism and pairs

Van Fraassen thinks that when we accept a scientific theory, we should be bracket the theory’s claims about unobservable entities, but believe everything else.

An oddity has occurred to me. Suppose a theory talks about certain microorganisms that are just under the minimum size for human visual observation. But when you have two things that are just under the minimum observable size side-by-side, the pair is observable. So, oddly, we will believe in pairs without believing in individuals.

For further oddity, now imagine that Alpha and Beta are such a side-by-side pair. Then we believe in Alphabeta, the pair of Alpha and Beta. Suppose Alpha swims a little away from Beta. Now, Alphabeta disappears from view. But Alphabeta is still observable. To observe Alphabeta, all we need to do is to coax Alpha and Beta to swim to each other. So on van Fraassen grounds, we should still continue to believe in Alphabeta even when temporarily we cannot see it due to the separation of Alpha from Beta. (Compare: a very thin sheet is still observable when it is edge-on, even though it can only be seen when tilted to the line of sight.)

But it is absurd to believe in a pair of organisms, at different ends of a test-tube, without believing in either organism.

One way out for van Fraassen is to adopt a sparse ontology on which there are no pairs. But while I like such an ontology, I don’t think van Fraassen will want to do that, as he wants to believe in observable objects that science talks about, such as planets.

Approximatable laws

Some people, most notably Robin Collins, have run teleological arguments from the discoverability of the laws of nature.

But I doubt that we know that the laws of nature are discoverable. After all, it seems we haven’t discovered the laws of physics yet.

But the laws of nature are, surely, approximatable: it is within our power to come up with approximations that work pretty well in limited, but often useful, domains. This feature of the laws of nature is hard to deny. At the same time, it seems to be a very anthropocentric feature, since the both the ability to approximate and the usefulness are anthropocentric features. The approximatability of the laws of nature thus suggests a universe whose laws are designed by someone who cares about us.

Objection: Only given approximatable laws is intelligence an advantage, so intelligent beings will only evolve in universes with approximatable laws. Hence, the approximatable laws can be explained in a multiverse by an anthropic principle.

Response: Approximatability is not a zero-one feature. It comes in degrees. I grant that approximatable laws are needed for intelligence to be an advantage. But they only need to be approximatable to the degree that was discovered by our prehistoric ancestors. There is no need for the further approximatability that was central to the scientific revolution. Thus an anthropic principle explanation only explains a part of the extent of approximatability.

A tension in some theistic Aristotelian thinkers

Here is a tension in the views of some theistic Aristotelian philosophers. On the one hand, we argue:

1. The mathematical elegance and discoverability of the laws of physics is evidence for the existence of God

but we also think:

2. There are higher-level (e.g., biological and psychological) laws that do not reduce to the laws of physics.

These higher-level laws, among other things, govern the emergence of higher-level structures from lower-level ones and the control that the higher-level structures exert over the lower-level ones.

The higher-level laws are largely unknown except in the broadest outline. They are thus not discoverable in the way the laws of physics are claimed to be, and since no serious proposals are yet available as to their exact formulation, we have no evidence as to their elegance. But as evidence for the existence of God, the elegance and discoverability of a proper subset of the laws is much less impressive. In other words, (1) is really impressive if all the laws reduce to the laws of physics. But otherwise, (1) is rather less impressive. I’ve never never seen this criticism.

I think, however, there is a way for the Aristotelian to still run a design argument.

Either all the laws reduce to the laws of physics or not.

If they all reduce to the laws of physics, pace Aristotelianism, we have a great elegance and discoverability design argument.

Suppose now that they don’t. Then there is, presumably, a great deal of complex connection between structural levels that is logically contingent. It would be logically possible for minds to arise out of the kinds of arrangements of physical materials we have in stones, but then the minds wouldn’t be able to operate very effectively in the world, at least without massively overriding the physics. Instead, minds arise in brains. The higher-level laws rarely if ever override the lower-level ones. Having higher-level laws that fit so harmoniously with the lower-level laws is very surprising a priori. Indeed, this harmony is so great as to be epistemically suspicious, suspicious enough that the need for such a harmony makes one worry that the higher-level laws are a mere fiction. But if they are a mere fiction, then we go back to the first option, namely reduction. Here we are assuming the higher level stuff is irreducible. And now we have a great design argument from their harmony with the lower-level laws.

Properties, relations and functions

Many philosophical discussions presuppose a picture of reality on which, fundamentally, there are objects which have properties and stand in relations. But if we look to how science describes the world, it might be more natural to bring (partial) functions in at the ground level.

Objects have attributes like mass, momentum, charge, DNA sequence, size and shape. These attributes associate values, like 3.4kg, 15~kg m/s north-east, 5C, TTCGAAAAG, 5m and sphericity, to the objects. The usual philosophical way of modeling such attributes is through the mechanism of determinables and determinates. Thus, an object may have the determinable property of having mass and its determinate having mass 3.4kg. We then have a metaphysical law that prohibits objects from having multiple same-level determinates of the same determinable.

A special challenge arises from the numerical or vector structure of many of the values of the attributes. I suppose what we would say is that the set of lowest-level determinates of a determinable “naturally” has the mathematical structure of a subset of a complete ordered field (i.e., of something isomorphic to the set of real numbers) or of a vector space over such a field, so that momenta can be added, masses can be multiplied, etc. There is a lot of duplication here, however: there is one addition operator on the space of lowest-level momentum determinates and another addition operator on the space of lowest-level position determinates in the Newtonian picture. Moreover, for science to work, we need to be able to combine the values of various attributes: we need to be able to divide products of masses by squares of distances to make sense of Newton’s laws of gravitation. But it doesn’t seem to make sense to divide mass properties, or their products, by distance properties, or their squares. The operations themselves would have to be modeled as higher level relations, so that momentum addition would be modeled as a ternary relation between momenta, and there would be parallel algebraic laws for momentum addition and position addition. All this can be done, one operation at a time, but it’s not very elegant.

Wouldn’t it be more elegant if instead we thought of the attributes as partial functions? Thus, mass would be a partial function from objects to the positive real numbers (using a natural unit system) and both Newtonian position and momentum will be partial functions from objects to Euclidean three-dimensional space. One doesn’t need separate operations for the addition of positions and of momenta any more. Moreover, one doesn’t need to model addition as a ternary relation but as a function of two arguments.

There is a second reason to admit functions as first-class citizens into our metaphysics, and this reason comes from intuition. Properties make intuitive sense. But I think there is something intuitively metaphysically puzzling about relations that are not merely to be analyzed into a property of a plurality (such as being arranged in a ball, or having a total mass of 5kg), but where the order of the relata matters. I think we can make sense of binary non-symmetric relations in terms of the analogy of agents and patients: x does something to y (e.g. causes it). But ternary relations that don’t reduce to a property of a plurality, but where order matters, seem puzzling. There are two main technical ways to solve this. One is to reduce such relations to properties of tuples, where tuples are special abstract objects formed from concrete objects. The other is Josh Rasmussen’s introduction of structured mereological wholes. Both are clever, but they do complicate the ontology.

But unary partial functions—i.e., unary attributes—are all we need to reduce both properties and relations of arbitrary finate arity. And unary attributes like mass and velocity make perfect intuitive sense.

First, properties can simply be reduced to partial functions to some set with only one object (say, the number “1” or the truth-value “true” or the empty partial function): the property is had by an object provided that the object is in the domain of the partial function.

Second, n-ary relations can be reduced to n-ary partial functions in exactly the same way: x₁, ..., x_n stand in the relation if and only if the n-tuple (x₁, ..., x_n) lies in the domain of the partial function.

Third, n-ary partial functions for finite n > 1 can be reduced to unary partial functions by currying. For instance, a binary partial function f can be modeled as a unary function g that assigns to each object x (or, better, each object x such that f(x, y) is defined for some y) a unary function g(x) such that (g(x))(y)=f(x, y) precisely whenever the latter is defined. Generalizing this lets one reduce n-ary partial functions to (n − 1)-ary ones, and so on down to unary ones.

There is, however, an important possible hitch. It could turn out that a property/relation ontology is more easily amenable to nominalist reduction than a function ontology. If so, then for those of us like me who are suspicious of Platonism, this could be a decisive consideration in favor of the more traditional approach.

Moreover, some people might be suspicious of the idea that purely mathematical objects, like numbers, are so intimately involved in the real world. After all, such involvement does bring up the Benacerraf problem. But maybe we should say: It solves it! What are the genuine real numbers? It's the values that charge and mass can take. And the genuine natural numbers are then the naturals amongst the genuine reals.

Two kinds of functionalism

There are two kinds of functionalism about the mind.

One kind upholds the thesis that if two systems exhibit the same overall function, i.e., the same overall functional mapping between sequences of system inputs and sequences of system outputs, then they have the same mental states if any. Call this systemic functionalism.

The other kind says that mental properties depend not just on overall system function, but also on the functional properties of the internal states and/or subsystems of the system. Call this subsystemic functionalism. The subsystemic functionalist allows that two systems may have the same overall function, but because the internal architecture (whether software or hardware) that achieve this overall function are different, the mental states of the systems could be different.

Systemic functionalism allows for a greater degree of multiple realizability. If we have subsystemic functionalism, we might meet up with aliens who behave just like we do, but who nonetheless have no mental states or mental states very different from ours, because the algorithms that are used to implement the input-to-output mappings in them are sufficiently different.

If subsystemic functionalism is true, then it seems impossible for us to figure out what functional properties constitute mental states, except via self-experimentation.

For instance, we would want to know whether the functional properties that constitute mental states are neuronal-or-above or subneuronal. If they are neuronal-or-above, then replacing neurons with prostheses that have the same input-to-output mappings will preserve mental states. If they are subneuronal, such replacement will only preserve mental states if the prostheses not only have the same input-to-output mappings, but also are functionally isomorphic at the relevant (and unknown to us) subneuronal level.

But how could we figure out which is the case? Here is the obvious thing to try: Replace neurons with prostheses whose internal architecture does not have much functional resemblance to neurons but which have the same input-to-output mappings. But assuming standard physicalist claims about there not being “swervy” top-down causation (top-down causation that is unpredictable from the microphysical laws), we know ahead of the experiment that the subject will behave exactly as before. Yet if we have rejected systemic functionalism, sameness of behavior does not guarantee sameness of mental states, or any mental states at all. So doing the experiment seems pointless: we already know what we will find (assuming we know there is no swervy top-down causation), and it doesn’t answer our question.

Well, not quite. If I have the experiment done on me, then if I continue to have conscious states after complete neuronal prosthetic replacement, I will know (in a Cartesian way) that I have mental states, and get significant evidence that the relevant system level is neuronal-or-above. But I won’t be able to inform anybody of this. If I tell people: “I am still conscious”, if they have rejected systemic functionalism, they will just say: “Yeah, he/it would say that even if he/it weren’t, because we have preserved the systemic input-to-output mappings.” And there will be significant limits to what even I can know. While I could surely know that I am conscious, I doubt that I would be able to trust my memory to know that my conscious states haven’t changed their qualia.

So with self-experimentation, I could know tht the relevant system level is neuronal-or-above. Could I know even with self-experimentation that the relevant system level is subneuronal. That’s a tough one. At first sight, one might consider this: Replace neurons with prostheses gradually and have me observe whether my conscious experiences start to change. Maybe at some point I stop having smell qualia, because the neurons involved in smell have been replaced with subsystemically functionally non-isomorphic systems. Oddly, though, given the lack of swervy top-down causation, I would still report having smell qualia, and act as if I had them, and maybe even think, albeit mistakenly, that I have them. I am not sure what to make of this possibility. It’s weird indeed.

Moreover, a version of the above argument shows that there is no experiment that we could do that would persons other than at most the subject know whether systemic or subsystemic functionalism is true, assuming there is no swervy top-down causation.

Things become simpler in a way if we adopt systemic functionalism. It becomes easier to know when we have strong AI, when aliens are conscious, whether neural prostheses work or destroy thought, etc. The downside is that systemic functionalism is just behaviorism.

On the other hand, if there is swervy top-down causation, and this causation meshes in the right way with mental functioning, then we are once again in the experimental philosophy of mind business. For then neurons might function differently when in a living brain than what the microphysical laws predict. And we could put in prostheses that function outside the body just like neurons, and see if those also function in vivo just like neurons. If so, then the relevant functional level is probably neuronal-or-above; if not, it's probably subneuronal.

Bayesianism and anomaly

One part of the problem of anomaly is this. If a well-established scientific theory seems to predict something contrary to what we observe, we tend to stick to the theory, with barely a change in credence, while being dubious of the auxiliary hypotheses. What, if anything, justifies this procedure?

Here’s my setup. We have a well-established scientific theory T and (conjoined) auxiliary hypotheses A, and T together with A uncontroversially entails the denial of some piece of observational evidence E which we uncontroversially have (“the anomaly”). The auxiliary hypotheses will typically include claims about the experimental setup, the calibration of equipment, the lack of further causal influences, mathematical claims about the derivation of not-E from T and the above, and maybe some final catch-all thesis like the material conditional that if T and all the other auxiliary hypotheses obtain, then E does not obtain.

For simplicity I will suppose that A and T are independent, though of course that simplifying assumption is rarely true.

I suspect that often this happens: T is much better confirmed than A. For T tends to be a unified theoretical body that has been confirmed as a whole by a multitude of different kinds of observations, while A is a conjunction of a large number of claims that have been individually confirmed. Suppose, say, that P(T)=0.999 while P(A)=0.9, where all my probabilities are implicitly conditional on some background K. Given the observation E, and the fact that T and A entail its negation, we now know that the conjunction of T and A is false. But we don’t know where the falsehood lies. Here’s a quick and intuitive thought. There is a region of probability space where the conjunction of T and A is false. That area is divided into three sub-regions:

1. T is true and A is false

2. T is false and A is true

3. both are false.

The initial probabilities of the three regions are, respectively, 0.0999, 0.0009999 and 0.0001. We know we are in one of these three regions, and that’s all we now know. Most likely we are in the first one, and the probability that we are in that one given that we are in one of the three is around 0.99. So our credence in T has gone down from three nines (0.999) to two nines (0.99), but it’s still high, so we get to hold on to T.

Still, this answer isn’t optimistic. A move from 0.999 to 0.99 is actually an enormous decrease in confidence.

But there is a much more optimistic thought. Note that the above wasn’t a real Bayesian calculation, just a rough informal intuition. The tip-off is that I said nothing about the conditional probabilities of E on the relevant hypotheses, i.e., the “likelihoods”.

Now setup ensures:

4. P(E|A ∧ T)=0.

What can we say about the other relevant likelihoods? Well, if some auxiliary hypothesis is false, then E is up for grabs. So, conservatively:

5. P(E|∼A ∧ T)=0.5
6. P(E|∼A ∧ ∼T)=0.5

But here is something that I think is really, really interesting. I think that in typical cases where T is a well-established scientific theory and A ∧ T entails the negation of E, the probability P(E|A ∧ ∼T) is still low.

The reason is that all the evidence that we have gathered for T even better confirms the hypothesis that T holds to a high degree of approximation in most cases. Thus, even if T is false, the typical predictions of T, assuming they have conservative error bounds, are likely to still be true. Newtonian physics is false, but even conditionally on its being false we take individual predictions of Newtonian physics to have a high probability. Thus, conservatively:

7. P(E|A ∧ ∼T)=0.1

Very well, let’s put all our assumptions together, including the ones about A and T being independent and the values of P(A) and P(T). Here’s what we get:

8. P(E|T)=P(E|A ∧ T)P(A|T)+P(E|∼A ∧ T)P(∼A|T)=0.05
9. P(E|∼T)=P(E|A ∧ ∼T)P(A|∼T)+P(E|∼A ∧ ∼T)P(∼A|∼T) = 0.14.

Plugging this into Bayes’ theorem, we get P(T|E)=0.997. So our credence has crept down, but only a little: from 0.999 to 0.997. This is much more optimistic (and conservative) than the big move from 0.999 to 0.99 that the intuitive calculation predicted.

So, if I am right, at least one of the reasons why anomalies don’t do much damage to scientific theories is that when the scientific theory T is well-confirmed, the anomaly is not only surprising on the theory, but it is surprising on the denial of the theory—because the background includes the data that makes T “well-confirmed” and would make E surprising even if we knew that T was false.

Note that this argument works less well if the anomalous case is significantly different from the cases that went into the confirmation of T. In such a case, there might be much less reason to think E won’t occur if T is false. And that means that anomalies are more powerful as evidence against a theory the more distant they are from the situations we explored before when we were confirming T. This, I think, matches our intuitions: We would put almost no weight in someone finding an anomaly in the course of an undergraduate physics lab—not just because an undergraduate student is likely doing it (it could be the professor testing the equipment, though), but because this is ground well-gone over, where we expect the theory’s predictions to hold even if the theory is false. But if new observations of the center of our galaxy don’t fit our theory, that is much more compelling—in a regime so different from many of our previous observations, we might well expect that things would be different if our theory were false.

And this helps with the second half of the problem of anomaly: How do we keep from holding on to T too long in the light of contrary evidence, how do we allow anomalies to have a rightful place in undermining theories? The answer is: To undermine a theory effectively, we need anomalies that occur in situations significantly different from those that have already been explored.

Note that this post weakens, but does not destroy, the central arguments of this paper.

Degrees of freedom

The number of degrees of freedom in a system is the number of numerical parameters that need to be set to fully determine the system. Scientists have an epistemic preference for theories that posit systems with fewer degrees of freedom.

But any system with n real-valued degrees of freedom can be redescribed as a system with only one real-valued degree of freedom, where n is finite or countable. For instance, consider a three-dimensional system which is fully described at any given time by a position (x, y, z) in three-dimensional space. We can redescribe x, y and z by real-valued variable X, Y and Z in the interval from 0 and 1, for instance by letting X = 1/2 + π⁻¹arctan x and so on. Now write out these new variables in decimal:

• X = 0.X₁X₂X₃...
• Y = 0.Y₁Y₂Y₃...
• Z = 0.Z₁Z₂Z₃...

Finally, let:

• W = 0.X₁Y₁Z₁X₂Y₂Z₂X₃Y₃Z₃....

Then W encodes all the information about X, Y and Z, which in turn encode all the information about (x, y, z) and hence about our system at a given time. (This obviously generalizes to any finite number of degrees of freedom. For a countably infinite one, things are slightly more complicated, but can still be done.)

There is a lesson here, even if not a particularly deep one. The epistemic preference for theories that have fewer degrees of freedom cannot be separated from the the epistemic preference for simpler theories. For of course rewriting a theory that made use of (x, y, z) in terms of W is in practice going to make for a significantly messier theory. So we cannot replace a simplicity preference by a preference for a low number of degrees of freedom.

Objection: Instead of a simplicity preference, we may a priori specify that laws of nature be given by differential equations in terms of the variables involved. But when, say, x, y and z vary smoothly over time, it is very unlikely that W will do so as well.

Response: But one can find a replacement for W that is smoothly related to x, y and z up to any desired degree of precision, and hence we can give a differential-equation based theory that fits the experimental data pretty much equally well but has only one degree of freedom.

Vertical uniformity of nature

One often talks of the “uniformity of nature” in the context of the problem of induction: the striking and prima facie puzzling fact that the laws of nature that hold in our local contexts also hold in non-local contexts.

That’s a “horizontal” uniformity of nature. But there is also a very interesting “vertical” uniformity of nature. This is a uniformity between the types of arrangements that occur at different levels like the microphysical, the chemical, the biological, the social, the geophysical and the astronomical. The uniformity is different from the horizontal one in that, as far as we know, there are no precisely formulable laws of nature that hold uniformly between levels. But there is still a less well defined uniformity whose sign is that same human methods of empirical investigation (“the scientific method”) work in all of them. Of course, these methods are modified: elegance plays a greater role in fundamental physics than in sociology, say. But they have something in common, if only that they are mere refinements of ordinary human common sense.

How much commonality is there? Maybe it’s like the commonality between novels. Novels come in different languages, cultural contexts and genres. They differ widely. But nonetheless to varying degrees we all have a capacity to get something out of all of them. And we can explain this vague commonality quite simply: all novels (that we know of) are produced by animals of the same species, participating to a significant degree in an interconnected culture.

Monotheism can provide an even more tightly-knit unity of cause that explains the vertical uniformity of nature—one entity caused all the levels. Polytheism can provide a looser unity of cause, much more like in the case of novels—perhaps different gods had different levels in nature delegated to them. Monotheism can do something similar, if need be, by positing angels to whom tasks are delegated, but I don’t know if there is a need. We know that one artist or author can produce a vast range of types of productions (think of a Michelangelo or an Asimov).

Any case, the kind of vague uniformity we get in the vertical dimension seems to fit well with agential explanations. It seems to me that a design argument for a metaphysical hypothesis like monotheism, polytheism or optimalism based on the vertical uniformity might not have some advantages over the more standard argument from the uniformity of the laws of nature. Or perhaps the two combined will provide the best argument.