Monday, 2 October 2017

In Search of All Souls

I recently sat the All Souls Fellowship exam, called by some the "world's hardest exam". It requires you to write twelve essays for four papers over two days; the breadth and novelty of the questions make it a fascinating experience. Two of the papers were "general papers" and two were in a humanities subject of your choice (in my case, philosophy); most papers had around 25 questions. I've summarised my answers below, as well as noting some of the other particularly interesting questions.

General paper I:

1. How should you prepare for the end of the world?
I started off with a somewhat emotional argument about the badness of death on an individual level. I referred to Epicurus' argument that we have nothing to fear from death because when we are dead, we will not have any preferences about it at all - but argued, in response, that having preferences about future states of the world is a foundation of our lives.

In terms of the end of the world, I identified some plausible ways in which humanity might go extinct in the near future, most notably the development of synthetic diseases or general AI. Then I discussed some cognitive biases which hold us back from addressing them: they're unpleasant to think about; we often ignore small probabilities of important outcomes; and general scope insensitivity about what's at stake (consider, for instance, those who believe that a child's death is a tragedy, but the obliteration of humanity is not so bad - even though the latter would entail millions of the former!). But we shouldn't allow such biases to dictate our thinking. Indeed, under any plausible theory of moral uncertainty (such as MacAskill's), we should place significant weight on the far future even if we don't personally think it matters. The original question is not hypothetical; the best preparation for the end of the world is to take action to prevent it.

2. Debunk a modern myth.
I identified the myth that companies have obligations first and foremost to maximise shareholder value. I talked about the duties of individuals, and then how they play out within the structure of a modern firm in a way which dilutes moral responsibility. I also talked about the cases in which businesses are exempted from moral duties which we usually expect from individuals. I won't say too much about this here as I'm working on an extensive essay about this.

3. Is the only good answer one which destroys the question?
No, I answered, but the best ones are. Why? Well, language has evolved partly to reflect our cognitive biases, but it also reinforces them - in particular, the bias which says that terms and categories which seem "obvious" to us reflect a real distinction in the world. This is what led, for instance, to centuries of debate over the Ship of Theseus and whether or not it remained the "same" ship. But these intuitions hold us back from the obvious conclusion of modern science: that although we perceive some everyday distinctions as fundamental (did x cause y? Is w the same person as z?), instead - as Democritus said - "in truth there are only atoms and the void". Of course such a reductionist viewpoint doesn't mean that it is pointless to talk about which distinctions to draw: some levels of explanation are more useful than others in a human context. But we should also recognise that many such discussions are "empty" and reflect only linguistic differences. The best answers help us realise that fact so that we are able to better evaluate the concepts which we are using, and perhaps discard them altogether.

Philosophy I:

1. Should we be Bayesians?
I explained Bayesianism as conditionalisation from a prior, also noting a slight modification which needs to be made in response to Arntzenius' objections to conditionalisation and reflection (in short, we use an ur-prior, which we conditionalise anew from total evidence whenever we receive more). I then identified three possible interpretations of the question:
a) Is Bayesianism the ideal of rationality? I argued that it is because of Dutch books, which lead non-Bayesians to contradictory probability judgments, or else suspension of belief. However, since it can't account for any revisions to the laws of logic, we also need a meta-theory which takes that possibility into account.
b) Does Bayesianism describe human thinking well? No, because of the computational impossibility of implementing it. In the limiting case where not much computation is required, we might expect that evolution has designed our brains draw conclusions in an approximately Bayesian manner; but it's unlikely that this applies in many cases.
c) Should we use Bayesianism as a guide to improving our thinking? Sometimes. For example, explicitly Bayesian reasoning helps us to avoid the base rate fallacy. However, in general it's rational to reason in ways which are explicitly ruled out by ideal Bayesianism - for example, revising our beliefs when we think of a new theory or notice a new implication of old evidence.

2. Were ancient slave-holders bad people?
I claimed that there is no good way of integrating the idea of being a "bad person" with consequentialism. If we base our evaluation of people on outcomes of acts, then in the hypothetical case where knowledge of the horrors of World War 2 prevented a full-scale nuclear war, Hitler would have been a good person; also, in general unpredictable 'butterfly effects' into the far future will dominate. If we base it on subjective judgments of the utility of acts, then those who deludedly believe that they are certainly saving the world will dominate. If we base it on rationally permissible subjective judgments, then some people simply could not be very good people if they were born in a time or place that became insignificant in the long term.

In general, since our evaluations of people and our evaluations of outcomes are driven by fundamentally different mental processes, it's unlikely we'll ever get a good definition of the former in terms of the latter. So I moved on to the question of which definition would be most useful. That depends on at least three properties: whether it is intuitively acceptable enough to change people's behaviour; whether such behavioural changes will lead to very good outcomes; and whether it's consistent with other principles we value. In the case of slavery, since there are still millions of slaves across the world, then making excuses for past slave-holders will probably lessen the strength of our impulses to help them. Also, we might lose faith in the idea of moral progress, which has motivated many important progressive movements. So we should be very clear that ancient slave-holders were bad people.

(This argument relies on the idea that we can and should try to change our moral intuitions in some ways; I didn't have time to discuss the meta-ethical arguments for this).

3. Do scientific explanations raise the probabilities of the evidence they explain?
I noted that the question was weird, because we usually think that we are certain of the data we already have. I then described what I thought we should mean by subjective probabilities and scientific explanations. I argued that the quality of the latter should be evaluated based on simplicity, coherence, and also subjective human interests - because the way we decide which levels of explanation we want is based on what we want a science to achieve. For example, we can explain the fact that almost all animals have equal birth rates of each sex either in terms of its evolutionary benefit, or the mathematical model of genetics which brings it about, or (very impracticably) the underlying atomic movements which implement animal life.

I then claimed that under the Duhem-Quine thesis - that every scientific observation must be taken to assume a large number of auxiliary hypotheses - we might doubt that one of the auxiliary hypotheses is true, and therefore that our observation is correct. We would be more likely to doubt this if there is no good explanation for that observation which is consistent with all the auxiliary hypotheses; therefore, a good explanation can help raise our credence that the event we observed actually happened. If "evidence" is taken to be such observed events, that answers the question. If "evidence" is taken to be merely the fact that we made an observation at all, then it becomes a little more difficult - can we really doubt, for instance, that we are seeing the colour red? But we can certainly doubt that we saw red at a point in the past, because our memories are fallible - so in this sense too a good explanation can raise the subjective probability that we actually had the experience we think we did.

General paper II:

1. 'All lives matter.' Do they?
I noted that I would take the question entirely literally at first, but eventually move on to its political context. So, what is a life? Noting the difficulties define it in biological terms, I moved on the idea of what lives might matter, and argued for consciousness as a key criterion. I claimed that we can distinguish 3 levels on which we value human lives: intrinsic (simply because a life is human), welfare (based on how good a life is for the person leading it), and outcome (how good the effects of a life are). For simplicity's sake, I assumed that all effects could be described in terms of changes to welfare, collapsing the third level into the second. Also, while most of us have clear intuitions that even humans with very little welfare - late-stage embryos, people in comas, etc - do matter intrinsically, such value is difficult to reason about consequentially due to the virtue ethical/deontological intuitions behind it.

Zooming in on welfare, I explained the problems with hedonic, preference and objective list accounts of welfare, since each can be taken to unacceptable extremes. (I also explained how our intuitions about individual lives cannot consistently be extended to whole populations because of Arrhenius' impossibility theorems). I then noted that under any reasonable fusion of these three accounts, there are some people - those who will have and cause zero or negative welfare over the rest of their lives - about whose continued survival we should be agnostic or even opposed. This is the first sense in which a life can "matter". Secondly, when deciding where to invest resources, there are some people who we can help very much for very little, and others whom we cannot help much, or at all. In that sense, the former lives matter more. However, even lives we cannot improve still matter in the sense that if we could increase their welfare without any trade-offs, we should do it. These three definitions help explain the clash between "Black lives matter" and "All lives matter": all lives matter in the third sense, since if we could hypothetically increase their welfare, we should. But black lives matter more in the second sense, since there are ways we as a society could help them very much very easily, such as cracking down on police violence.

2. Are there any questions that should be beyond the pale of philosophical discussion?
I noted that there are two contexts for this question. The first is the entirely theoretical context; the second is the current context of ideological conflict over offense and censorship. In theory, it is clear that there are some discussions which should not be had: truth is not the only good which philosophers do or should value. Human flourishing is another, and certain truths - such as a true report to a genocidal dictator listing all the crimes committed by minorities - lead to the exact opposite.
In the current context, however, the question is whether we can identify those cases. I noted that from an outside view, people with the power of censorship have been very bad at identifying what to censor over the last few centuries. However, it's difficult to dispute that we should be very confident in censoring claims that certain ethnic groups, religious groups, women, and so on, are normatively inferior. The difficulty lies in distinguishing normative claims from descriptive claims - and I argue that in practice, social justice movements have often failed to do so. For example, the claim that sexual orientation is a choice, although mostly used by conservatives to argue against gay marriage, is only harmful to LGBT+ individuals given the further claim that intrinsic identities matter more than chosen ones. I think that there are very good reasons to consider the latter claim false.

Of course, many do think that it is true. The argument that sexuality is intrinsic helps to counter their arguments without needing to challenge their fundamental normative claims. But it also has the potential to massively backfire - for instance if scientific evidence to the contrary emerges, then gay marriage advocates will be discredited as unscientific. I claimed that another example of a potentially harmful lack of clarity has been the continued conflation of "women are, on average, less interested in mathematics than men, due to intrinsic factors" with "we shouldn't give equal treatment to the women who are interested in mathematics." Of course you cannot derive either statement from the other. But since there are a number of respected psychologists who agree with the former statement, making it taboo is a bad strategy for feminism. Instead, we should all work to clearly separate normative and descriptive claims, so that arguments for equal rights do not need to depend on scientific findings.

3. 'When everyone is somebody, then no one's anybody' (W.S. GILBERT). Discuss.
I pointed out that depression and mental health issues are incredibly prevalent among millennials. We are in an uncanny valley between developing technology that we use to replace many social interactions, and developing technology that is good enough to replace most of the valuable aspects of social interaction. This harms our friendships; meanwhile, time spent online is generally unproductive and unsatisfying. Over time we will, I think, become better at dealing with the distractions of the internet. But right now it's not only distracting, it's also individualistic, since using technology is currently a very solitary activity.

The growth of individualism is not new; in fact, there have been many contributing factors over the last century or more. We've seen declines in organised religion, massive migration to cities (and, indeed, between countries), no-fault divorce, the normalisation of frequent job-switching, smaller families, widespread acceptance of individualist economic assumptions, and loss of local social structures (see the classic book Bowling Alone). We now focus on finding the ideal career without realising that it's communities, and relationships within them, that make us happy. Today everyone is 'somebody' - an individual - but few of us are anybody in the communal sense which matters the most.

Philosophy II:

1. Is scepticism irrefutable?
I framed philosophical scepticism as a problem of underdetermination: all evidence will always beconsistent with multiple possible worlds. As long as we do not rule any of them out a priori, then we will always have multiple options to which we assign non-zero probability. But can we refute scepticism in the sense of rationally assigning it very low probability? I present some attempts to do so from Blackburn, Harrod, Stove, etc, but they all fail. Then I turn to Solomonoff Induction. While it doesn't solve the traditional problem of induction, since it relies on a number of assumptions, I argue that it represents important progress. Further, when we think of hypotheses as descriptions in a certain language, then on average shorter descriptions must always be more probable; this helps to independently motivate Occam's Razor. (I hope to write a post soon to explore this last point further).

2. Should we be effective altruists?
I felt like I wrote fairly standard things here - but then again, they are standard things which many philosophers don't accept. I said that EA was based on two normative claims and a factual claim. The factual claim: doing a great deal of good costs us very little. The normative claims: 1. if we can do significant amounts of good without significant sacrifices, then we have a duty to do so; 2. fulfilling this duty requires us to focus on the most effective charities. The basic argument intuition behind 1 is evoked most strongly in Singer's Drowning Child thought experiment; however, it also relies on the factual claim above, and the further claim that the difference between saving a drowning child near us and a starving child in a faraway country is not morally significant. I think it's difficult to make strong arguments for this latter claim, as people either find it intuitively compelling, or not; hopefully, though, as the world gets "smaller" future generations will find it more convincing. In support of 2 I argued that the duty was never phrased in terms of "donating money" but rather "doing good"; replacing a donation to an effective charity with one to an ineffective charity is thus a choice to do less good, which means that you're not fulfilling that duty as well.

I then addressed some objections. Firstly, that we might not believe the factual claim, because the net effect of donations might include, for instance, propping up corrupt governments. But it would be strange if, in a world of such inequality, there are no ways we can significantly benefit others for relatively little money - and if there are, we can find out about them in the same way that we gain other scientific and economic knowledge. Secondly, that EA is too demanding. But you don't need to accept an infinite duty, only that you have a moral obligation to help others until the point that it causes you significant sacrifices. Different people might have different ideas of what that means, but it's difficult to believe that giving up a slightly fancier car or a new TV is really such a significant sacrifice - especially given that psychological research indicates that donating that same money will probably actually make you happier! Thirdly, that far future concerns will end up dominating. But they probably should dominate, because of the huge amounts at stake; and anyway, EAs who believe that the far future doesn't matter can simply avoid contributing to those causes. Being an EA isn't defined by participating in the movement, but rather by acting on the beliefs above.

3. Does it make sense to say that one false theory is closer to the truth than another?
I claim that the idea of being "closer to the truth" is an inherently vague one which doesn't generalise well to scientific theories. An intuitive definition would include at least two criteria: closeness to observations, and closeness to fundamental composition (i.e. ontology). It's far easier to measure the former (although Kuhn disagrees, claiming that observations made in different paradigms are incomparable). When it comes to everyday phenomena, we only really care about the former - for instance, most of us agree that cats are made out of atoms, but if it turns out that the science was wrong and atoms just don't exist, we wouldn't say that all our statements about cats had been wrong this whole time. (I claim therefore that Putnam drew incorrect conclusions from his 'Twin Earth' thought experiment). On the other hand, we care about fundamental composition more in a scientific context: if atoms didn't exist then it's probably reasonable to say that we've been wrong about molecules this whole time. Van Fraassen thinks that there's a clear distinction between these two cases, based on what's "observable" and "unobservable", but the line he draws just doesn't make sense. So the best we can do is to accept some combination of the two criteria, with the understanding that it will always be vague and impossible to formalise. In fact, we can interpret 'structural realism' as such a compromise.

Other interesting questions from the general papers:
1. Why are there still Communists?
Obviously because real Communism hasn't been tried yet. /s
2. If you were a dictator, what would you ban?
It was surprisingly difficult to think of an answer to this one - perhaps because bans, as very blunt tools, are seldom the best option. Maybe smoking, but that would've been a pretty boring essay.
3. Account for the rise of internet memes.
4. What is the use of magic?
5. Can a deal be art?
6. Invent a new idiom - then spill the beans.
7. For what should we ask forgiveness?
8. Why is the price of housing so high?
Supply-side stupidity!
9. Is incompetence the most underrated force in human history?
10. 'The first thing we do, let's kill all the lawyers' (WILLIAM SHAKESPEARE). Discuss.
11. What if there were no hypothetical questions?

From the philosophy papers:
1. "I have tried to show that what matters in the continued existence of a person are, for the most part, relations of degree" (PARFIT). Discuss.
Awesome question, wish I knew enough to properly answer it. It seems difficult to deny the force of Parfit's argument that our future selves are different people in the same way that everyone else is, especially because of its thoroughly reductionist grounding. But in practice it may well be psychologically impossible to fully accept it.

2. Is the golden rule a good guide to morality?
3. What is distinctive about moral disagreement?
Two fairly standard ethics questions, although they touch on important issues.

4. Is it possible to refute those who deny the Law of Non-Contradiction?
5. Is there any more reason to doubt the existence of root(-1) than to doubt the existence of root(2)?
6. Does the Generalised Continuum Hypothesis have a truth-value?
7. Are there any infinitesimal quantities?
Some fascinating questions in philosophy of mathematics. Again, I wanted to answer them but simply haven't read up enough on different theories of what it means for numbers or logical proofs to 'exist' or 'be true'.

8. Does music have meaning?
9. What, if anything, unifies surprise, anger, sorrow, disgust, guilt, contempt, amusement and wonder as emotions?
Two thought-provoking questions touching on everyday life.

10. Is quantum non-locality consistent with the Special Theory of Relativity?
11. What is the best interpretation of probabilities in quantum mechanics?

And from the Economics papers, which I had a chance to look at, but didn't sit:
1. Does the experience of China undermine the proposition that democracy is good for economic growth?
Not at all; rather, it supports the idea that democracy is low-variance for economic growth, while autocracies can be very good for it (China in the last few decades) or very, very bad (China right before that). Since potential downside far outweighs potential upside, democracy is still superior.

2. Is there a tension between global and national equality?
Without better redistribution, yes there is: one of the best things for global equality is the outsourcing of jobs to developing countries, which obviously hits blue-collar workers hard.

3. Can a price be put on human life?
Empirically, yes: a few thousand dollars in some African countries; a few million dollars in most Western ones. But don't confuse price with value! Water, for example, is the most valuable thing to most people, but its price is very cheap.


  1. What is the purpose of a price on life if it doesn't confer value?

    1. It tells you how much it costs to "obtain" (in this case, save) a life. I think that's useful so we know where we should be directing our money.