Steve Reads

Conversations with Josh recently remind me that a lot of people think science is cold, scary, and inhuman — that it tends to focus on subjects that are remote from what we humans actually care about. Rather than focusing on love, goes the argument, science focuses on things like neurons or evolutionary theory — explaining to us that our feelings for another human being are merely the emergent behavior of many electrical spikes or a sophisticated way for our genes to propagate. Science should focus on the “primal” subject, which is the love itself — not any of the abstractions that it has introduced to understand the world.

I think I understand the fear that underlies this objection to science: that it threatens to strip all the beauty from our very real feelings for other human beings, or that it turns artwork and poetry into something cold and dead. I think that’s what Merleau-Ponty is getting at when he says (quote via Josh):

To say that the world is, by nominal definition, the object x of our operations is to treat the scientist’s knowledge as if it were absolute, as if everything that is and has been was meant only to enter the laboratory . . . If this kind of thinking were to extend its dominion over humanity and history; and if, ignoring what we know of them through contact and our own situations, it were to set out to construct them on the basis of a few abstract indices — then, since the human being truly becomes the manipulandum he thinks he is, we enter into a cultural regimen in which there is neither truth nor falsehood concerning humanity and history, into a sleep, or nightmare from which there is no awakening.

Additionally, Merleau-Ponty seems to be saying that once we view ourselves as scientific abstractions, we cease to treat one another as human beings. I no longer treat you with love because it feels like the right thing to do — I do so because I am, for instance, a rational utility-maximizer and I can shoehorn my love into that framework, or because natural selection dictates that I behave in that way to maximize the chances of propagating.

There’s certainly a danger of this happening. Stephen Jay Gould (via Marx, apparently) noted our habit of inventing an abstraction — often a mathematical one — then assuming that it corresponds to a real thing; Gould (and Marx) call this “reification.” IQ is Gould’s big example of reification: it’s a single number that presumes to rank all humans linearly, and is at best a model of a much more complex underlying reality. It’s useful for certain things — it was, according to Gould, originally invented to identify children who needed more help in school (if I recall correctly) — but our perhaps-innate habit leads us to believe that we actually can rate humans in this way and using this measure. The abstraction becomes the thing.

It’s worth noting that this is by no means peculiar to science. We all invent abstractions about one another all the time, then use them to guide the way we view one another. I have abstracted George Bush as a cold, calculating, selfish, bloodthirsty, amoral, power-hungry bastard, even though it’s almost certainly true that there’s a much more complicated set of facts sitting inside his head. When I deal with my friends, I use a much more nuanced set of abstractions: this guy is nice, this girl is a flirt, this one is egotistical. Knowing my friends as I do — I certainly know them better than I know Bush — my model is likely to have much more predictive power. If one of my friends suddenly blew up a schoolbuilding, I would be very surprised, because I’ve built an intuitive model in my head of how other people behave. Violations from that model are, by definition, surprising.

Surely I’m missing something in my understanding of Merleau-Ponty’s view of science. Because as it stands, I view science as just a more-refined species of the intuitive models that we all build. Assume that people are utility-maximizers and see what happens. Maybe the model will fail, but at least you have something formal that can break — all your cards are on the table; at least when you view me as a self-obsessed mercenary, you’re doing so on the basis of assumptions that I can see and question and modify.

The difficulty comes, of course, when we prematurely sanctify the model as the reality. At some point, it seems to me, we are justified in behaving as though models were reality; we have to, if we’re going to get anything done. And as philosophers are fond of pointing out, there doesn’t exist a bare fact — every fact is model-laden. So if we’re going to bother to look at the world at all, we probably have no choice but to use some model to do so.

If that’s granted, then the question is which models we’ll use. May we sometimes use formal models? Of course; we shouldn’t constrain ourselves to only use human intuition to understand the world. We should use whatever is useful to understand the world around us.

So is it useful to use science to understand the beauty that we feel when we listen to a Coltrane song? I don’t understand why we’d object to such a thing. The objection seems to come from the level at which we study Coltrane: when we hear Coltrane and feel a stirring in our souls, we’re not thinking about the movement of air over our timpanic membrane. Perhaps a science that truly deserved Coltrane would study the higher level: why is it that I get such a stirring in my soul when I hear him?

My fundamental block in understanding is that I don’t see why understanding the physiological level — which is, I think, the part that a lot of people find cold about science — interferes with our understanding the higher level. We can study both. We can try to understand why the Greeks noticed that humans perceive certain tones as pleasing. We can try to understand what happens when an electrical spike down a particular nerve reaches a particular part of my brain. We can study the parts of the brain that activate when I listen to music, and we can compare those regions of activation to the analogous regions in monkeys. We can ask people to describe in lurid detail what it is that they feel when they listen to A Love Supreme, and we can compare multiple people to understand the differences between them. We can do all of these things, and our understanding of any enriches our understanding of the rest.

Maybe part of the reason that there isn’t a well-developed science of subjective experience is that it’s hard. One of the points that Philip Kitcher makes in The Advancement of Science is that science will often handle idealizations of problems because the problems themselves are temporarily intractable without the idealization. Eventually, after they’ve fleshed out the details of the idealizations, they can relax some assumptions in the idealizations and get a more realistic model of the phenomenon they care about.

It’s also perhaps a question of goals. What questions do we want to answer about the subjective experience of listening to Coltrane? Do we want to be able to predict what John Doe, the subject under investigation, will say when asked for his feelings about Coltrane? Knowing that he’s deaf — a comparatively low-level detail about John, certainly lower-level than the subjective experience of “listening to a Coltrane album” — will tell us a great deal, just as knowing that my friend confuses the colors red and green suggests that his subjective experience of a Monet painting will be different than mine. That experience will be different, presumably, because red and green evoke different emotions for different reasons. The physiology directly feeds into the subjective experience. It doesn’t completely determine the subjective experience, but it does provide some explanatory power. Maybe knowing that I had a traumatic experience with frogs as a child will tell us a great deal about how I view the color green.

My point is this: it’s not clear to me that the only way — or even the best way — to understand a phenomenon is to study it “at its own level” (whatever that might mean). Quite often studying it at other levels gives us insight into the phenomenon. Studying a phenomenon scientifically only detracts from our appreciation of that phenomenon if we take our eyes off the prize (the prize being the phenomenon itself) or if we start to confuse an at-best-approximate model with the reality.

I might have completely misunderstood what Merleau-Ponty has to say; that would make sense, given that the sum of my exposure to him has been, in essence, the paragraph I quoted above. If I should know more, tell me what I should know.

I finished reading The Advancement of Science today. It’s a very solid piece of work. In the final portion, Kitcher sketches a model of scientific consensus-building: assuming that individual scientists behave in such-and-such a way, assign probabilities like thus-and-so, and have utility functions looking like so (with higher utilities for discovering a result before anybody else, higher utilities for discovering something that was previously disbelieved, etc.), when would we expect the community to advance? The answer you get — and the one you’d expect — is, “Well, it depends.” It depends in large part on empirical observation about how science actually proceeds: is it largely filled with people who pursue the truth without regard to their own well-being? Is it filled with those who only hope to advance their careers, the truth be damned?

In all likelihood — duh — the community is a mixture of both. If science only progresses when people are self-sacrificing saints, we have problems. Kitcher sketches a formal demonstration that the community would have to be totally filled with careerists before its epistemic hunt would collapse in the way that some people predict.

Kitcher’s book is valuable in a lot of ways, but I’d single out two:

1. It gets away from the idealization of what Kitcher calls Legend — the belief that scientists inevitably move forward through their steadfast attention to the bare facts of experience, filtered through minds that are free of human foibles (or at least, minds that can be made free of these foibles through rigorous training that all scientists are supposed to have). At the same time, Kitcher moves away from the other side’s equally unrealistic idealization: that science has no greater claim to truth than any other system of belief, and that politics infests it as much as any other belief system. Neither of these idealizations is likely to be true.

2. Whatever the truth is, we’ll need evidence to figure out the way that science actually works. And that’s Kitcher’s second big contribution: turning a question from the philosophy of science into one from the history of science. Again this is in line with my methodological biases: to the extent possible, turn questions about the world (such as “How do scientists do their jobs?”) into empirically falsifiable facts. If modern views of science are supposed to be more realistic, and these views understand that scientists are fallible humans, then it behooves us to model their foibles and understand how those foibles translate into success or failure in the practice of science.

Kitcher reminds me a lot of Judge Richard Posner, or maybe they both just come from the same school of rigorous analysis. Both will take some vague quantity — in Posner’s case, say, “the strength of intellectual-property protection,” in Kitcher’s something like “the unearned authority attributed to someone else in the field” — and sketch some properties for it. Posner, for instance, argues that whatever “the strength of intellectual-property protection” means, we can assume certain things about its effects: increasing it up to a certain point will encourage the production of new artistic works, but there is some maximum past which more protection leads to decreased production. Kitcher speculates that whatever “unearned authority” means, an increase in it probably draws more people into fields containing researchers with lots of unearned authority. Or he might speculate that a new researcher entering a field will scale his probability of believing in a result with the authority of the result’s author. These are reasonable assumptions; Kitcher and Posner pursue them as far as they can, and hope that others will fill in the details later on. They both manage to derive useful conclusions from the logical apparatus that they lay out.

I would have liked Kitcher to have tied his formalism back to his earlier models of scientific progress, where he argues that the force of pre-existing data constrains the models that later scientists can propose. Had Kitcher collected some data, he could have constrained himself nicely: certain of his parameters might be bounded by known data from the history of science, such as the time it took the biological community to converge on natural selection as its accepted dogma. These high-level data must constrain the lower-level models, such as Kitcher’s models of group consensus-formation or even his models of how individual scientists behave. It would have been interesting to see him write something like, “It took chemistry 20 years to get behind Lavoisier, so the parameter k can be no greater than 2.” Or something. Maybe future investigators will run with this.

This parameter estimation points out another good point that Kitcher makes: it’s probably not useful to talk about “science” as an abstract thing. There are many sciences; each have their own standards, their own reputable people, their own methodologies, their own accepted schemata, and their own connections with other sciences. While we know what we talk about when we talk about science generally (peer review, doubly-blinded experiments, repeatability, rigorous mathematics and the rest), Kitcher’s point is that if we want firm answers, we’ll have to focus our questions on the specific science or specific sub-speciality that we care about. The parameters will vary depending upon whether we’re talking about molecular biology, physiology, or physics. That’s one of the advantages of his formal approach: by explicitly using tunable parameters in the model where they’re necessary, Kitcher makes us face up to those times when we’re talking too generally about “science” as a monolithic institution.

Indeed, one of the points that struck me as I read Kitcher was that many of his ideas could carry over to human thought generally. If science brings out cognitive features in humans, and maybe elevates them to a more rigorous level, his results ought to be able to help us say things about, say, human deference to authority generally.

And of course at some level the results are self-swallowing: whatever Kitcher might conclude about scientific deference to authority probably applies to him, and applies to philosophers generally. Do philosophers often fail to chase the truth because their judgment is blinded? Do they refuse to accept someone else’s authority because that person commented negatively on one of their papers? Contrariwise, do philosophers run down particular lines of thought less because those ideas are likely to be true and more because they’re likely to get funding for the research? Sure. As do scientists. And it’s our job as scientists to properly model the systematic irrationalities in any line of research, philosophical or otherwise. Maybe someday there will be a “sociology of philosophy” to match the “sociology of science.” Maybe there already is such a thing.

I took the red line into Park Street today, walked from there to the top of Newbury Street, walked the length of Newbury Street to Mass. Ave., and followed Mass. Ave. all the way up to Davis Square, with some nice stops in between to, e.g., lay under a tree in Harvard Yard and call friends. I had no idea, but Google Maps tells me it was a 5-odd-mile walk. This is a very walkable city. If I move to New York (still kind of a pipe dream, and maybe not the best idea at that), I wonder whether I’d have as much fun walking it.

Maybe before I move there, I’ll take a day to walk the length of Manhattan with Josh.

A few thoughts prompted by today’s walk:

• The sight of two gay men holding hands walking down Newbury Street is a very tangible (literally “made out of Tang”) sign of progress. It gives me hope.

• It would be pretty great if Boston, Cambridge and Somerville closed off various streets and squares to car traffic. Newbury Street in particular would be a drastically more enjoyable place without cars, as would Davis and Harvard Squares. Cutting them off to cars would, of course, be a traffic headache, but a) I wonder how many people currently route around those streets and b) I wonder how hard it would be to reroute everyone else.

• Does it bug anyone else that when you call someone’s voicemail, sometimes “#” skips the prompt and brings you right to the beep, whereas sometimes “” does that? Whenever the one key skips the prompt, the other is the key that the phone’s owner uses to sign into his voicemail and check messages. So if you press “” to skip the prompt on a service where “#” is actually the right key for that, you won’t be able to record a message at all. Bugs me every time. Can’t the phone companies just standardize on one use for the “*” and one use for the “#”?

I’d really like to hack Google maps to overlay the MBTA subway map atop a Boston street map.

For that matter, I’d like to overlay Google’s street maps atop Google’s satellite images. Or overlay absolutely everything I can find, then create some kind of mini-app that can turn layers on or off.

Some part of this shouldn’t be hard, should it?

I have just started using del.icio.us. I think it’s the worst possible name for a product, but we carry on.

I am now just a little bit closer to buzzword-compliance.

P.S.: Neat: you can get an RSS feed of my del.icio.us links. (Can we stop calling them “del.icio.us links” and instead just call them “shared bookmarks”? Because that’s all they are.)

I’m working on a slightly new design for this site. Let me know what you think. It needs some work, I know. And I’ve been trying to figure out what color links on this site should be; red didn’t do it for me, and now neither is the shade of blue I use. I welcome any suggestions.

I’d like to reiterate that the particular structure of argument that Philip Kitcher is using is tailored to my style. He’s asking how it’s possible that scientists could be petulant schoolboys (read: suffer from normal human failings) and yet still improve their models over time. In particular, what if I cite your paper only because you cited mine? How many scientists would have to behave this way before the entire enterprise collapsed? (To make this question more precise and less rhetorical, assume that an individual scientist’s goals can be categorized as either “epistemic” or “non-epistemic”: those that are focused on increasing knowledge, or those that are aimed at other goals like prestige and so forth. One interesting question, then, is how altruistic scientists have to be — how focused on purely epistemic goals — in order for a scientific discipline to make progress. If the institution is so precariously balanced that it only progresses when everyone acts like a saint, then it’s unlikely that it’ll go anywhere.)

So Kitcher does what I need him to do before I’ll buy it: he develops a model. Assume that scientist A weights scientist B’s judgments according to some weight function w(A, B), which represents the probability that A believes B’s results are sound. Break this down into support that B has earned (because, for instance, he has gathered data that A has reproduce) and support that B has not earned (say, because B has a reputation within the field that A takes on faith). Run with this for a while and see where it takes you. Even if you’ve failed to model the way that actual science works, you’ve started down a path toward precision: suddenly, whether you’re right or wrong, you’ve said something specific enough that people can conclusively reject it. And if you’re good like Kitcher, what you’ve laid down — along with a very honest assessment of your model’s weaknesses — is a firm foundation for future work.

(Note, however, that we’re 72 games into a 162-game season.)

P.S. (June 26, 2005):

I’ve heard a fair number of people arguing against Bayesian methods in statistics and artificial intelligence. Quite often the idea is that humans are simply not Bayesian thinkers — we demonstrably deviate from the perfect rationality (e.g., transitive preferences) demanded by the Bayesian axioms. Since humans have several million years of evolution training us how to make biologically sound decisions, the lack of inborn Bayesianism certainly seems like a good argument.

Bayesians often retort that while people may not act like rational agents, they should: that certain optimality results follow from acting like a Bayesian. Most famous of these is probably the Dutch Book argument: if you don’t act like a Bayesian, I can get you to take a bet that you’re certain to lose. No one would ever take such a bet, so this is either an argument that you should act like a Bayesian, or that everyone does act like a Bayesian; if they didn’t act this way, they’d always lose bets.

Yet people are not Bayesians. So how do we square the fairly convincing Dutch Book argument with the fairly convincing psychological data?

P.S.: Cosma to the rescue, with a recommendation that I read “Against Conditionalization” and the end of Kitcher. Thanks, Cosma!

Recent experience leads me to suggest that the following would be useful additions to stylesheets:

1. The ability to define text substitutions that get applied to every page using a given stylesheet. So for instance, I would like every page on my site to replace “...” with “ . . . ” — ellipsis dots will always be separated by spaces. (I realize there’s an HTML entity for ellipses — namely … — but I don’t typically like the spacing on it. Pedantic, I realize.) One normally does this sort of thing application-by-application — e.g., I use a Blosxom plugin to do it on this site — but that’s not really general enough. Basically, any page I serve off this site, I want to go through a filter that converts ... to the appropriate ellipsis. You might consider doing this through your web server, but what if you don’t have administrator access to the server? The better idea, it seems to me, is to set up a site-wide stylesheet that every sub-stylesheet on the site imports.

   This might help with another problem I’ve had at one client. Their earlier web-hosting company didn’t allow them to do Server-Side Includes. SSI is damned handy: you can stick a bunch of HTML inside of a file (because I’m creative, call it foo.html), then include foo.html inside an outer file (call it bar.html) with an SSI directive like

   <!--#include virtual="foo.html" -->

   Before bar.html leaves the server, the directive gets replaced with the contents of foo.html; all that the browser sees is a single, complete bar.html file containing foo.html.

   Since my client’s hosting company doesn’t allow SSI, they had to find some way to include bits of external HTML inside of their pages. So they use Javascript. This seems like the wrong approach to me, but it makes sense: there seems to be no canonical client-side way to include other bits of text within an HTML file. CSS is starting to handle something like this using the ‘content’ property.

2. Algebraic rules for CSS properties: e.g.,

   elem1 (elem2|elem3) { statement1; statement2; }

   would apply statement1 and statement2 to any elem2’s or elem3’s that are subelements of elem1’s. This falls under CSS regular expressions, I suppose.

P.S.: Another idea, unrelated to the topic here, but I wanted to add it somewhere: Why is it that it’s not easier to resort a table on the client end within HTML? I should be able to declare a table or a list with something like

<ul sortable="yes">

and thereby turn each column heading into a sortable list. Clicking the column heading would sort by that column; clicking twice would sort in reverse order.

The trickiness in this idea comes when you move beyond basic lexicographic sorting: what if you want to sort the list

• December 21, 1992
• December 23, 2003
• January 1, 2003
• etc.?

Presumably then you’d need to XMLify this somehow — e.g., you’d replace “December 21, 1992” with

 <date> <year>1992</year> <month>December</month> <day>21</day> </date> 

then transform it into a formatted string via XSLT (about which I know nothing) or somesuch. The formatting would come after the sorting, so you’d define how objects tagged with <date> are sorted. Or maybe you could have a lower-tech method to convert “December 21, 1992” into “1992-12-21” and then sort the column lexicographically like you would any other text string.

Somehow you’d want all of this to take place on the client end. I’m not sure how you’d do this. Maybe define a new XML event onSort that could call a script which would reorder the items in the column. Maybe there’d be a default handler for the onSort event which would sort the column lexicographically. And by default all the columns in the table would be resorted to follow the column that we’re sorting.

P.P.S.: Hmm. Looks like there’s already enough XSLT in the browser to do all this?

Now this is an interesting idea: pay $80 per month for unlimited Verizon cellular broadband, then make phonecalls using Skype with your laptop. You get unlimited wireless phonecalls with coverage identical to Verizon’s, but you’re paying a flat fee. $80 per month would be cheaper than my cell bill quite a lot of the time. Plus I’d get reliable broadband Net access nationwide. That’s pretty hot.

I’m happy to see that a product that friends of mine work on made it into the hallowed* pages of Slashdot. Maybe this’ll get them some sales.

* — Tongue so firmly in cheek that it’s poked through.

If your e-mail does not have a Sender ID, Microsoft wants to junk your message. Somewhere after November, MSN and Hotmail will consider it as spam. Sender ID is a specification for verifying the authenticity of e-mail by ensuring the validity of the server from which the e-mail came. Some experts feel that ‘Sender ID’ is not an accepted standard and has many shortcomings. Some also feel that Microsoft is trying to strong-arm the industry into the adoption of an incomplete and not accepted standard.

If true, it’s stupid. Gmail doesn’t do this, and there’s no reason why anyone needs to stick with Hotmail.

I think I’ll suggest to my Hotmail-using friends that they switch to Gmail sometime before November.

I’m reading Philip Kitcher’s very carefully argued, fair, and readable history-of-science book The Advancement of Science right now; I’m about 3/4 of the way through, and the pace is still brisk.

Kitcher’s question is fairly simple: scientists are not automata who process data and spit out theories; how is it, then, that imperfect beings who fall short of logical perfection might be able to produce scientific theories that get closer to the truth as time goes on?

My biases should be obvious by now. I’m fairly convinced that science is moving closer to the truth over time, that there are decent standards for deciding whether one theory is better than another, and that the scientific method is our best method for understanding the world around us. I think Russell fundamentally got it right when he said, “Whatever can be known, can be known by means of science.”

That said, I think Kitcher’s book will do a fair job convincing even those who don’t share my biases that maybe there’s more to science than they initially gave it credit for.

To me, his most convincing device for convincing people is to look at the actual data of how science has progressed. How does actual scientific change happen? When one worldview comes up against an incompatible worldview, how do scientists resolve their disputes? Science seems to come to agreement over basic tenets over time: biologists, for instance, would be laughed out of the room if they didn’t believe in natural selection; is this just power politics codified as truth, or is it that scientists have converged to the right answer? And if they have, why is it that scientific revolutions happen? Why is it reasonable for us to expect that our own views of science will be overturned at some point in the future? If they’re converging, why do their beliefs get overturned eventually?

It’s always seemed to me that the answer is fairly simple: we discover more data, we invent new instruments (say, telescopes), and we thereby notice more phenomena. It shouldn’t surprise us that our old worldviews get overturned as we learn more about the world. And during periods of scientific “revolution” (Kuhn’s phrase, which Kitcher rejects — more on that in a moment), we struggle to find new data that will conclusively falsify one theory over another. As my friend Charlie noted, Einstein would have no choice but to accept quantum mechanics if he were alive today; its predictions are just too thoroughly established for someone with our level of knowledge to plausibly reject it.

There’s a host of criticisms against this view. One is the “underdetermination” thesis: the idea that the data never constrain the space of models enough to conclusively accept one model over the others.

Kitcher dismisses this as a philosophical problem rather than one that scientists care about in practice, after careful thought and argument over decades:

For all my pleading on its behalf, the notion that individual scientists can profitably modify their practices by using eliminative induction is likely to seem (at best) quaint. Surely the idea that there are crucial experiments (or observations) in science was disposed of long ago, specifically by Pierre Duhem (1906). Thanks to Quine’s revival and development of the Duhemian theme (1951, 1960, 1970), the notion that theories are inevitably underdetermined by experience has become a philosophical commonplace. Scientists, however, sometimes greet this allegedly mundane point with incredulity. “It’s hard enough,” they complain, “to find one way of accommodating experience, let alone many. And these supposed ways of modifying the network of beliefs are changes that no reasonable — sane? — person would make. There may be a logical point here, but it has little to do with science.” I think that the complaining scientists are, at least roughly, right. The underdetermination thesis, in its usual guise, is a product of the underrepresentation of scientific practice.

In a footnote, Kitcher then writes,

Concerns along these lines were voiced by Stephen Jay Gould. In several conversations in the early 1980s, I tried to convince Gould that Duhemian holism spells doom for any idea of conclusive falsification. He resisted my arguments. I now think that he was correct.

Kitcher goes on to develop a model of how scientific theory revision happens, and — this is crucial to my own acceptance of any argument about “how science works” — tests it against a number of examples from the actual history of science.

His model involves what he calls an “escape tree.” I have a new model that I have to somehow bring into harmony with the rest of my beliefs. As I revise my current theory in the light of new data, I may find that other, more fundamental beliefs (laws of physics, say) need to be revised to fit my new theory. By revising those other beliefs, I may incur certain costs — for instance, I may lose an explanation of a phenomenon that current theories predict very well. But it may be rational to introduce my new theory, because it may buy us other predictions whose value is greater than the lost predictions. And as a scientist, I may propose that my theory will eventually explain the lost predictions as other scientists fill in the details and file off the edges of my theory. In other words, it may very well be rational to lose some explanatory power for a short time, if we think we can regain it in the long run. Newton lost Aristotle’s understanding of the structure of space, but gained an enormous range of predictive power and a simplified theory that Aristotle never had. And a few hundred years later, Einstein developed Newton’s theory to provide an explanation for why objects are attracted to other objects roughly in inverse proportion to the square of their distance and in proportion to the product of their masses. That is, Einstein eventually gave us what Newton lost, while still expanding on what Newton had done; the scientific change was perfectly rational, and there can be no doubt that we’ve made progress. This neatly undercuts Kuhn’s claim that science doesn’t move forward, but quite often in circles; Kuhn asserts that Einstein’s physics is much closer to Aristotle’s than to Newton’s. Kitcher politely notes that you can adopt the best pieces of both, and still salvage a realistic claim that science is making progress.

This process of changing a theory, changing related theories, dropping predictions that we had made, and eventually filling in the details of a given theory (e.g., Wright/Haldane/Fisher’s mathematical reconciliation of Mendel with Darwin) constitutes Kitcher’s description of how theories are revised. And it’s important to note that throughout the process, scientists cannot introduce new hypotheses willy-nilly; they are drastically constrained by pre-existing models. Auxiliary hypotheses — which are supposed to nullify Karl Popper’s theory of falsification — aren’t easy. We may encounter a theory of natural selection, for instance, that contradicts well-known ideas from molecular biology, but it will take a tremendous amount of arguing to convince people that molbio needs to be overthrown; new theories have an enormous hurdle to get over. And it’s enormously difficult to introduce new auxiliary hypotheses that violate no pre-existing scientific models.

Kitcher argues this all very carefully, and provides numerous examples from the history of chemistry and biology to back up his argument. It will, of course, be better left to historians of science to decide whether Kitcher has appropriately described the data. But his argument is clear, makes quite a bit of sense to me, and accords with my intuition that science is constrained by reality. We can’t introduce any hypothesis we want, so it’s unreasonable to believe that there are infinitely many models consistent with a given set of data. Hence the underdetermination thesis is hard to swallow, and Kitcher makes it harder.

Kitcher also makes short (but courteous) work of Kuhn’s idea that scientists speaking across a scientific revolution are using words that are mutually incomprehensible — for instance, that the language of phlogiston is so model-laden that those talking about oxygen are building their world atop a totally different, totally incompatible structure. Again, this is hard to swallow if you realize that both those scientists are trying to understand the world and trying to model the same phenomena: they’re trying to understand why things explode, why the mass of the reactants in a chemical reaction is roughly equal to the mass of the products, etc., etc. They can’t be speaking completely incompatibly, because they’re trying to understand the same world. They’re not playing a word game; they’re playing with the same nature.

And empirically, it is unreasonable to believe that these scientists are speaking in mutually-incomprehensible tongues, because we have historical data to show how they spoke with one another. And we have data showing that formerly unmoved scientists became convinced as more data came in. Lavoisier performed crucial experiments that led his opponents to realize that their options were drastically limited: only a few models could possibly explain the data, and their own preferred models were not among those available. Claiming that scientists can’t speak across a divide ignores the fact that they do speak across the divide and that they change their minds on the basis of new data.

(Note that this once again shows Popper’s wisdom: we may not be able to say which theory is right, but we certainly know which ones are wrong. Popper was good at pointing out these asymmetries; his observation that maximizing happiness is not the same as minimizing suffering, and that the two goals demand totally different policies, still rings true.)

Kitcher addresses as many arguments against the progress of science as he can find. Given Adam Rosi-Kessel’s taste for Bruno Latour, Kitcher’s response to Latour struck me as particularly interesting. Now I’ll have to read Latour; Adam should read Kitcher.

Kitcher has much to say, and says it quite well. I highly recommend his book.

P.S.: My obsession with a formal model theory of scientific models is based on the same intuition that Kitcher runs so far with: formalizing the idea that the data constrain the space of models a great deal. Kitcher’s formalization seems to be based on the subjective expected costs and benefits of changing a model (which makes sense, given that we’re dealing with scientists who, like all humans, have cognitive limitations), whereas I’m looking for something more formal. Both have predictive power, but Kitcher’s actually  . . .  you know  . . .  exists.

That is all.

One of the great joys of reading blogs is that every now and then you read something about a subject you knew precisely nothing about, and you feel like you’re getting a much better taste of world news than American newspapers or television ever give you — a report on the Chiapas region of Mexico, for instance. Well worth reading.

The Boston Globe, reporting on the final game of the Sox’s three-game-series sweep of the Indians, notes that “the defending World Series champions are 11 games over .500 (41-30) for the first time this season.”

They are thereby buying into terminology that I’ve never quite gotten. There have been 71 games this season. If the Sox were playing .500 baseball, they’d have won approximately 35 games — meaning that, with 41 wins, they are 6 games over .500, not 11. The Sox have won 11 games more than they’ve lost, but that doesn’t make them 11 games over .500.

Am I missing something?

 . . . is on a bit of a roll recently, and this week’s “The Onion 2056” is really pretty brilliant. I’m particularly a fan of the Infographic, listing which cybernetic implants are most popular. Among them my favorites are “TestiCubes Stackable Testicles,” the “Ikea Rüump,” and “I Can’t Believe It’s Not Lymph!™.” Plus I love that “Breast implants” are included in there, totally straightfacedly.

Faced with that literary troublemaker TS Eliot, a York University academic called in the FBI and now claims to have cracked the case of how The Waste Land was written.

Lawrence Rainey, of the university’s English department, spent two years travelling across Europe and the US to sort out the sequence in which Eliot wrote the poem.

A sheaf of rough drafts for the poem surfaced in 1971 and Prof Rainey compared them with the letters and other writing that Eliot was producing in the years before its publication in 1922 — a task calling for forensic as much as literary investigation. He examined more than 1,200 leaves of paper, including 638 pages of letters, Eliot had written between 1912 and 1922, visiting 22 international libraries and several private collections in his two-year journey.

(Via The Bookslut)

Julian Barnes from Flaubert’s Parrot:

Why does the writing make us chase the writer? Why can’t we leave well enough alone? Why aren’t the books enough?

Bill Clinton writes today for the New York Times, describing the relief effort in the tsunami-stricken areas, and explaining his role as the U.N. special envoy.

And it strikes me: can you imagine George Bush, in the years after 2008, playing any kind of diplomatic role? Can you imagine sending him to another country as a gesture of goodwill?

He’ll spend his time on the board of ExxonMobil, in all likelihood, and I hope we never hear from the fucker again.