« Evaluability (And Cheap Holiday Shopping) | Main | Unbounded Scales, Huge Jury Awards, & Futurism »

November 28, 2007


If we map 'the philosophy approach' and 'the literature approach' onto machine learning algorithms, philosophy will map to something like a causal bayesian network, while literature will map to something like k-nearest neighbor. (K-nearest neighbor algorithms answer classification problems given some learning data by just storing the data away as is and classifying a new point based on k 'closest' points in existing data).

Apparently classification algorithms are in some sense equivalent (no free lunch theorems).

"What evidence suggests that readers of such stories have gained such insight, instead of merely gaining an intuition telling them of such insights? Has anyone tried to measure this supposed increased insight in literature readers, such as in their ability to choose better actions?"

I have very often earnestly asked myself the same question. What benefit does the reader of literature really obtain from reading literature? While I think that literature benefits its readers in certain ways, i.e. increased imagination, perception of human situations, ability to understand complexes of motives and conflicting feelings, I also think that there is an upper bound to the benefit of literature, and this upper bound is reached relatively early in life (18-25 years, in some cases), and that many people continue fruitlessly beyond that upper bound, with detrimental effects on their character and outlook.

I think this is primarily because literature schools that part of our mind that is related to our affects, the emotional-experiential part of the mind, and whatever people may say, decreasing neuroplasticity and the reinforcement of our preferences cause us to adopt one way of being and feeling, so that reading literature after the upper bound is just indulging the desire for affective feeling while in no way challenging the mind to view things differently(I often find). This is not meant to slight readers of literature, but the propensity of literature to warp character is known to me through a few personal examples. Or perhaps what I have experienced is just the tendency of already-warped personalities to surround themselves with literature for purposes of self-reinforcement.

Christopher Hitchens, the famous alcoholic pro-Iraq war pundit, is a good and well publicized example of what I mean. His mastery of reading and writing, as well as his dealings with "great old dead writers", gave him the feeling that he exists on a higher plane than the rest of us mere mortals, so the tone of his articles and speeches is invariably condescending to the point of hilarity (or cringing), even while he demonstrates no real penetrating understanding of anything as far as I can tell. Literature, in giving him a feeling of superiority while teaching him to be self-indulgent, vain, self-centered and non-analytical, effectively ruined him, IMO.

Essentially, people select what they read. They tend (tend!) to read in ways that reinforce what they already think, they filter out what they do not want to hear, and literature allows them to recreate the world in an affect-centered way that reinforces whatever patterns of observation and reaction they had previously acquired. A communist will reinforce his communism (I found) to the exclusion of other points of view, as will others.

Science, on the other hand, demands that you lay aside your ego, your problems, your emotional affects, if you are to understand it. Afterwards you can bring these things back in, but in the moment of comprehending something, you have to go beyond yourself, because the truth ultimately does not care about your "taste" or your "feelings". Mental models of fictional characters are usually self-referential, so playing with mental models from literature is effectively "playing with yourself" and has in fact this self-indulgent element, while quantitative and logical models are usually not self-referential, i.e. you do not understand a math problem by thinking about your feelings towards that problem. Science allows you to go beyond yourself, and so does literature, but few people seem to use literature to go beyond themselves after age 25. This is why some literati make the impression of being grown-up children, which is literally amazing in light of their other well developed traits. I knew one literature grad student, one professional writer who had written 13 fiction books (2 published), one non-professional writer who wrote 2 literary books, two or three devotees of literature, and some other people through blogs and websites.

Humans seem to be innately predisposed to communicate in stories. This means that we will absorb ideas and concepts presented to us in narrative without significant conscious effort. In constrast, transmitting information in non-narrative forms requires a great deal of effort on the party of everyone involved, but most especially the receiver.

If you're looking to eliminate bias, then, non-narrative is the way to go. Only by consciously analyzing and examining presented information can we screen out misinformation, shoddy reasoning, and argument by association. Narratives, by being intuitive, magnify and conceal the flaws inherent in our intuition.

Of course, when the two are combined, the results can be extraordinarily powerful. But very few can pull that off.

My children are learning history through the use of the "Story of the World" series of books and audio-books.

They are close to riveting, in terms of presenting factual information about history in an interesting and emotionally-meaningful way.

So it absolutely could be done with other types of reports, but it takes tremendous skill to do it right. Also, it's simplified, because it's aimed at children, and more complex reports would be harder to present in a literary style.

I suppose if you insist on looking to literature as a source of evidence about something, it might work like this:

Literary writing is typically more sensuous and evocative than academic writing. Done right, it can communicate more about the texture of certain kinds of experience. In a manner of speaking it can cheat: it has a free pass to indulge in rhetorical tricks, take risks with outlandish metaphors, and generally ride roughshod over the strictures of impartial inquiry.

Another important fact is that it often contains dialogue, which most academic writing doesn't. This means literature can have a stab at representing one of the defining features of what it's like to be in a given milieu - namely, how people talk to one another.

Sometimes a novel is written about life under certain conditions, and it causes people who have actually lived under those conditions to say things like: "Yes - it's just like that." These rare occurrences might have some value as evidence - weak, ambiguous evidence - about what it's like to do various things.

I must stress, though, this is far from being the point of literature. You might as well object to Kahneman and Tversky on the grounds that their papers lack lyricism. Science is meant to be true. Literature is meant to give pleasure. On occasion you can kill two birds with one stone, but you don't have to.

I don't know about reading literature, but in my experience writing literature (even poorly-written literature that you never intend to show anyone) is quite useful. You can use the characters as a sort of protective glove, to explore concepts you would otherwise shy away from as threatening to your worldview. It is a sort of a workaround for your tendency to avoid asking hard questions of your beliefs.

Following up to Gray Area:

I would say a better analogy to algorithms is as follows. Philosophy is like GOFAI: attempt to reason your way to a solution, through mathematical calculation and logical deduction, without any learning (=experience). Literature is like machine learning, except with a really noisy data set. Reading nonfiction may be better, in the sense that there's less noise, but it could also be systematically biased by the emphasis on interesting times, places, and events.

A book I would really like to read is 1000 one-page biographical summaries of totally normal people. How did they live their lives, what were the sources of their happiness and sadness, etc.

Normal for what trait?

Couple thoughts--not terribly confident of them, though I think there's something there:

1. Fiction as a fast and frugal heuristic. You generate a scenario and evaluate it based on how much it seems like reality. This could be useful when you can't test a hypothesis, or testing is expensive beyond reasonableness, or there's very little in the past to study regarding the problem, so you generate cases based on what seem to be the laws or quasi-laws of the world. The story is better for some problems than academic approaches.

2. Fiction as integrator of facts, theories, and values (or what you believe to be your values) to create a unified normative-positive theory of what- is/what-should-be-done. "These are the important facts, this is what they mean, and this is what should be done given this." Individual facts are instances of general realities, so many can relate. The story is an inside view represenation of the outside view. Plausibility is judged by how much the story 'feels' true, and 'errors' in the story can be discussed among the readers of a story.

Would popular stories represent some kind of wisdom of crowds?

I think that the value of fiction is related to El's planning fallacy - fiction draws from experience. Research has a tendency to focus on how we think people *should* act (given certain situations/assumptions, of course). Fiction tends to focus on how we think people *would* act in given situations. If you focus excessively on the "should" you can lose sight of what you know about the "would."

Litterature offers the same insights as anecdotes: it suggests situations, often true, that may not have occured to the reader. Statistical academic reports focus instead on the measurable prevalence of a specific situation.

Litterature, and anecdotes, are useful for two reasons: because our rationality is bounded, and because they can pack a lot more ideas into a small space than an academic report can.

Consider the three situations:
-"Statistics says that health system X cures people much better than system Y, for all diseases" vs "Maybe, but my friend went to system X and died of cancer, but another one went to system Y, and survived!"
Verdict on the anecdote: completely trumped by the statistical evidence, hence can be discarded.

-"Statistics says that health system X cures people much better than system Y, for all diseases" vs "Yes, but the doctors in system X are so uncaring!"
Verdict on the anecdote: useful for phrasing a research question, irrelevant to the first statement.

-"On average, this NGO is very well run" vs "Yes, but there's a problem with motivating staff, and the managers don't tend to listen well, and they drive around in huge jeeps, alianating locals. They're good with accounts and getting projects starting and ending on time, I'll grant you that. They also know the culture of the area pretty well"
Verdict on the anectdote: provides nuance (if true) in a very short space of time. If the time/cost of transforming the anecdotal insights into statistical measures (and then checking them) is high (or if it can't be done), then we may find the anecdote more usefully informative than the statistics.

Since we don't have detailled statistical reports on every moment of every person's life, we'll often find that litterature and anecdotes are more useful to us in the everyday. Maybe more meaningful too, but that's a very different issue.

Of course, academic writing lies between litterature and statistics, but tends to be narrow and often closer to the statistical end of things (apart from survey articles).

I have a microeconomics textbook coming out in Feb 2008 that supplements the traditional analysis with many original fictional stories. I believe that the stories will help students better understand microeconomics in part through making the material more interesting. It is my hope that for many students my stories will provide "important insights that [they] cannot (or at least do not) express and understand abstractly" If the book is a success it will provide some evidence of the power of fiction to help give insights into abstract ideas. Since there are objectively right answers in basic microeconomics (as opposed to Philosophy) it will be possible to test if the stories really do provide students with insights that they don't get from the traditional approach.

The book's webpage is here:

This study purports to measure some of the effects of reading literature


It may be of some interest to consider authors who wrote both philosophy and literature, and see which turned out better. Maybe the innate advantages of one type of writing help the same person to say better things.

The first two examples that come to my mind are:

William Godwin (better philosophy)
Ayn Rand (better fiction)

I stopped reading fiction long before I read this, but the same idea was there. The problem here is that literature is being compared to philosophy, which makes more of a pretense of imparting real knowledge, but I don't of any reason to believe that is actually the case (as Robin noted with respect to moral philosophy here).

I think that Plato's critique of the poets has some relevance, since it can certainly influence us in ways that strictly academic writing can't. It's anecdotal, and it has a way of playing on the emotions. We learn more from our experiences than we do from simply reading about abstract ideas, which literature is more akin to. Other types of writing rely on logic and empirical evidence.

I think that it would be perfectly justified to question whether or not works, even great ones, portray things truly. Just because we find something to be beautiful, it still may be the case that its gives a faulty understanding of society. We can only say that a work is accurate if its views of society and human interactions are in some way similar to how they are in real life. Plato was right on this point (that these things can be dangerous, but I wouldn't advocate censorship as he did), but wrong in not realizing the benefits of literature. Homer, for instance, was regarded, and still is, as portraying life truly--Plato criticized him mostly on theological and moral grounds, and failed to understand Homer's real appeal.

Plato's point was not that philosophy was a superior truth to literary truth since, quite correctly, he regarded that way of talking about truth as bogus. His problem with poets was that the poets misled people into believing *falsehoods*.

Who said literature was about the same order of insight as philosophy or statistics ? literature excels in broadening, deepening, and heightening human emotional experience, while philosophy and statistics do the same for the rational experience. Complementarity. No reader of this blog will need reminding of the power of the emotional to 'fix' the experience. Think of Pirsig and Gaarder, what do you remember of their books, the philosophical content or the narrative envelope ?
Analogue afterthought : could a photographer have created the 'Madonna of the Rocks' ?

Sean, are you sure Plato failed to understand Homer's true appeal?

Remember, the famous "ancient quarrel" quote is from Republic, and has a political context. Homer is subversive because he speaks the truth about human affairs, in a way that calls into question the Noble Lie. Also, many of Socrates's criticisms of Poetry can also be applied to Plato's Republic quite easily, so it would be strange to ascribe those views to Plato himself.

I wonder whether the importance and deeply ambiguous status of Poetry might be because any sufficiently novel conceptual communication is in a sense poetic. (Cf. Nietzsche's claim that all speech is lying.) That is, to describe things for which you did not yet have a language, you must use your existing language in incorrect or metaphorical ways.

Those Plato passages about the conflict between poetry and philosophy must be some of the most misread passages in intellectual history (and I'm aware there are some pretty strong contenders). He put them in the mouth of a character in a dramatic dialogue - one which had gone to great pains to portray characters, particularly that of Thrasymachus in the early sections. Plato was using irony, which is one of the things that literature does very well, and exploring what he felt to be real, irresolvable conflict, which is another thing that literature does very well.
The irony of literature enables writers, at their best (I'm thinking of writers like Dostoyevsky here) to construct the strongest version of opposing arguments, to really explore them, take them for a ride, try them on for size in a way that a philosopher really rarely will; it also enables them to explore the many ways in which someone's perspective can shape and colour his/her views. Sure, philosophy or psychological/sociological research can do this too, but there's strong grounds for thinking that literature's gestalt approach might be better.
It's also a good way to identify genuine conflicts or sources of fissure within thought. I'm recalling Eliezer's comments about the Would Have Banned shop and the real possibility of its being true both that such shops ought, on the whole, to exist and that such shops would inevitably lead to undeserving and painful deaths. Real ethics is so often about making hard choices. I don't see why it isn't then sensible to have a forum for examining and acknowledging the difficulty of those choices.
If you are reading this and thinking 'all very well, but literature does not in general do this', then you are quite right for the simple reason that literature is not in general very good. But then, neither, in general, is philosophy.

I would venture two explanations for people thinking they have a lot of insight from literature:

1- The Narrative Fallacy. We love stories, and like to think of everything in stories and understand things better when given to us in stories. It is easier to lie to people if you couch your lie within a sequence of events than to just say the lie on its own.

2- Literature allows people to glean the "insight" they want to get without having to indulge with data, science, real critical thinking, nuance and the world being very complex and not very malleable to generalizations. Development is a perfect example: a proper academic study will fail to establish much, will be very nuanced, will have strong detractors and critics, and will be very conservative in its outcomes. It's hard to read anything in development and walk out with a feeling of revelation, and generally, the more you learn, the less sure you are of any preconceived notions you had. A novel on development will just put forward views on development within a nice story and convince us that they are ultimate truths. If you do not like the truths, you won't like the book and that's the end of it. If you liked these truths, and they match with your previous biases which you seek to confirm, then you go around telling everyone in the world that these truths are self-evident, that literature provides you with unmatched insight into these issues, that anyone who hasn't read this novel is an idiot, and that anyone who relies on science, rational thinking and hard-work to come to their views is wrong, because it can all be had on a red-eye flight from an airport novel.

Why go through hundreds of studies, data-sets, books, surveys and lectures to understand the determinants of economic growth, when an author could write a nice novel that attributes it all to the factors you want to believe are the determinants, and come out feeling justified. The beauty of literature, unlike shoddy academic writing, is that it doesn't even have to grapple with the real world to find one confirming example; it'll just make it up.

But, then again, all of this might just be the narrative fallacy of a development grad student who hasn't read fiction in years, and is jealous of his Creative Writing roommate's workload of drinking and reading fiction, drinking and watching movies and drinking and having chats with friends.

Fiction is useful for suggesting questions that might not otherwise have occurred to you, but it's completely useless for supplying answers.

They tend (tend!) to read in ways that reinforce what they already think, they filter out what they do not want to hear

Good literature - and good readers - do precisely the opposite; they open a world that otherwise be foreclosed from the reader. That's the strongest epistemological claim of lit, and it's a good one, IMHO.

I'm struck by how passionate, detailed, diverse, and numerous are the opinions expressed on this subject. So why isn't there a book on this subject? Seems easy to write and seems a large audience is interested.

Ah, but what *sort* of book?

Discussing the comparative merits of literature and the more philosophical/scientific approaches has always been tricky for me because of what I recognize as my own innate biases -- I've spent most of the last several years studying literature and literary theory and writing fiction. It seems to me that every person is going to have tendencies one way or the other, and may even have mindframes that directly oppose one of the means of abstraction, if not both together. Because of this, it seems that one of the most redeeming values of literature is a discussion of it memetically. Don't ask me for figures or statistics on the matter, but it seems intuitive that literature, and hybrids of entertainment/idea-positing in general disseminate much more rapidly throughout the populace. While one could easily make the argument that this doesn't mean that these ideas are sound, the more scientifically oriented could use them as a guiding point for futher exploration of a certain generation's tendencies. For example, does a resurgence of Hemingway sales represent a growing interest and sympathy in machismo? &tc.

"Litterature offers the same insights as anecdotes: it suggests situations, often true, that may not have occured to the reader."

Even more than this, for the more scientifically minded, literature can act as a model, taking an idea into world scenarios where natural (or, granted, unnatural) extensions of the idea can be examined. The emotive aspects of literature provide the objective-minded individual to wonder, "What does my response towards this model mean?", while a person oriented towards the subjective will be forced to confront the rationale of the literature, even if only a little. Philosophy is not without its anecdotes and metaphors as a means to highlight an idea -- literature, through interpretation, is simply the reverse of this process.

It's also been argued that literature is a good facilitator for discussion for people who normally wouldn't bother with that sort of thing. Every person, with their a priori ideas and experiences, is going to bring a different view (of varying degree) to a particular work, which usually turns into a forum for the ideas of the novel in and of themselves.

We're dealing with a toolbox here, with each tool useful for different things. After all, when all you have is a hammer...

There are several fascinating themes here: the emotional effects of literature, the tendency for narratives to be more compelling than abstract accounts (even aside from their emotional content), exploration of moral issues, etc. I agree with Robin that a book is warranted (maybe several, along with much research).

I want to focus on another theme that came up only a few times, however. I agree with Dan Burfoot that abstract theories are like good old fashioned AI, and literature is more like statistical learning. However the implications are much more serious than he draws out.

Models and typical social statistics are great if they are on topic and incorporate the important aspects of the situation. Like the attempts at predicate calculus "common sense" in GOFAI, any strong models define a framework for reasoning that makes some sorts of understanding much easier and more reliable, at the cost of excluding a lot of the complexity of the world. They work great on their target domain, and fail badly if the model doesn't fit the particular situation.

Once we have models and data, and people whose professional role is to use them, often we let these models and data determine the topic and "what's important". In that case they can become fatal traps and cause enormous human misery, which may be completely invisible from within the model.

Furthermore models are often hard to build and change, and data is hard to collect. So we may know of important aspects of the situation that we can't model or quantify. If we talk in terms of models and data we are then unable to even talk about these important aspects and they tend to become invisible.

Literature can always be extended to express whatever aspects of a situation we're aware of, and through the writing process we may considerably develop our awareness. As Acheron remarks, most literature is bad, like most of everything else, so most literature won't capture important aspects of a situation that would otherwise go missing -- but if that's what we want to do, literature is a much more reliable tool than models and data.

This pattern shows up vividly in international development, as mentioned in several comments. Theories have tended to sharply limit the vision of planners, who then blithely prescribe policies that lead to disaster. Often the potential for disaster could be detected by listening to people's stories and trying to fit them together, but the results could not be translated into the theoretical framework, and didn't count as "data", and so were ignored until after the catastrophe was undeniable.

This pattern is dissected though case studies in Seeing like a State by James C. Scott, which is pretty much a book length study of biases driven by this sort of abstract model dependence in certain domains.

(One could argue that statistical learning can get us out of this problem, and I might agree. But that is a whole different discussion, and I believe the points above are an accurate reflection of how we use models and data today.)

Has Robin Hanson ever noticed how much self-imposed stupidity there is in social science? "Men Prefer Attractive Women" No kidding.


"Sean, are you sure Plato failed to understand Homer's true appeal?

Remember, the famous "ancient quarrel" quote is from Republic, and has a political context. Homer is subversive because he speaks the truth about human affairs, in a way that calls into question the Noble Lie. Also, many of Socrates's criticisms of Poetry can also be applied to Plato's Republic quite easily, so it would be strange to ascribe those views to Plato himself."

With regard to Homer's "appeal," I probably worded that wrongly. I should have said something like "value" instead. What I was getting at was that Plato had a point that literature can lie, and still be convincing. Plato's reasons for dismissing Homer were wrong (at least for me, because I don't share the philosophical commitments in that text).

I agree that the view shouldn't necessarily be ascribed to Plato--but rather to Socrates in that specific text. In fact, the discussions he has of art throughout his writings are often conflicting.

How many of Plato's students were successful
advisers to the rulers they advised? How many
of Hegel's disciples improved the life-span,
living conditions, and increased the freedom
of the people whose countries they governed?

Is there not error, bias, and folly in every
endeavor to which the human mind has turned?

Was Richard III as evil as Shakespeare wrote
him? Did witches prophesy Macbeth's rise and
fall? How many reports of development
economists are any more anchored in verifiable
observation than Macbeth's of Banquo's ghost?
More than one, perhaps? More than one on each
of the seven sides of an issue? If there are
seven grant making agencies with competing
agendas, who here would bet against it?

Literature never of itself brought us as much
science as the boiling point of water.
But while cummings proclaimed:

"when serpents bargain for the right to squirm. . .

"and march denounces april as a saboteur

"then we'll believe in that incredible
unanimal mankind (and not until)"

how much philosophy, economics, psychology
and the rest of science is suited only to
a unending variety of "unanimal" mankinds?

Some small points:

First. Sartre published both works of literature and works of philosophy. Does anyone know why he chose both forms? Surely not just to reach more people?

Second. Many of the philosophers of the past had the notion that the written word "told" the brain what to think and that important ideas should be understood (and not just read) and therefore the best way was to explain between the lines. This goes well with the idea that literature surely can explain some extra quality.

Is there insight in, say, George Orwell's 1984?

Thanks Robin for motivating this fascinating discussion.
In case other readers didn't have a chance to follow the link posted above by anonymous, it linked to a study "Bookworms versus nerds: Exposure to fiction versus non-fiction, divergent associations with social ability, and the simulation of fictional social worlds" by Raymond Mar et al.
The study finds that people who read fiction (vs non-fiction) have better skills in comprehending their peers. (yes, yes, I know we need to understand the selection bias) - but if you take the result as is for a moment, it does suggest that one can gain insight - if you define that to mean understanding of the people around you - by reading fiction. Of course, they didn't measure whether this is the most efficient way to gain such understanding. It's possible that one could gain it as easily by reading psychology journals. But that would be much less fun.

Of course, they didn't measure whether this is the most efficient way to gain such understanding. It's possible that one could gain it as easily by reading psychology journals.

Neither the fiction nor the journals are likely to contain genuine insights - although there will certainly be exceptions, the vast majority of both will consist of trash and nonsense.

But the trash and nonsense embedded in the fiction is far more likely to take root in your mind than the trash and nonsense in the journals.

The trash and nonsense in Marx's journals
seems to have taken root in more minds
than much fiction. But, then, Marx was
fiction, wasn't it.


The advice at the link below may be of benefit for certain other writers at this blog:


The heart might know things the mind knows nothing of, but the heart cant' be challenged. Literature and the arts can provide insights, but reason and philosophical thought allow ideas to be developed, analyzed challenged and determined to be correct or incorrect based on the standard of reason. Ideas challenged create the structure by which civilization's ideas of Truth advance. Good literature (and the arts in general) doesn't have to be reasonable to convey a message of Truth (it has to convey a message of Truth to be good however) and therefore can't be reasonably determined to be correct or incorrect.

Just my 2 cents...

(i say this without having read the other comments - their combined length seems about four times that of the article itself).

Fiction's insights are implicit, philosophy's are explicit. In fiction aiming at a reasonable-level of realisim, the author is aiming at working out a realistic scenario.

They are usually trying to highlight certain features of that scenario, and because of this, it is designed for the reader to draw certain conclusions about it (e.g. about the nature of NGOs).

But what the author explicitly specifies is the scenario - the conclusions and, in particular, the model of why those conclusions are the case, is left somewhat implicit.

We are pretty good at evaluating concrete descriptions of scenarios, which helps the author to construct a reasonably realistic scenario.

Philosophers, on the other hand, tend to work on the level of general models: general descriptions that are abstracted away from specific scenarios. Descriptions that explicitly talk about why certain things are the case, /in general/.

These are harder for us to evaluate. And it is harder for us to incorporate nauances within models than within stories.

I'm not trying to say that stories "are better". I think that's why there is value to be had in both, and personally, I'm more interested in trying to draw the general, explicit lessons - the kind of thing philsophy is more targed at.

The comments to this entry are closed.

Less Wrong (sister site)

May 2009

Sun Mon Tue Wed Thu Fri Sat
          1 2
3 4 5 6 7 8 9
10 11 12 13 14 15 16
17 18 19 20 21 22 23
24 25 26 27 28 29 30