I never got into philosophy. People arguing about the reality of chairs in the neighboring room and such. I don't have that much imagination.
But today Advisor handed me a commentary in last week's Cell regarding the philosophy of science. The authors (David Glass and Ned Hall) discuss the history of the hypothesis, by which they mean the history of how scientists have approached their work. Do you set an a priori hypothesis, which you then try to falsify with experiments? Or do you use your current knowledge about the world to create a model of how things work, and then do an experiment to test your model?
The first version is called hypothesis testing, and the second is known as induction. At least, I think that's what they said. Every time I hear the word "induction" I think of stovetops, so I may not be a reliable guide here.
The authors discuss the known problems with both approaches, and with these classifications generally (e.g., how do you come up with a hypothesis to falsify if not by the process of forming a model and inducing/inducting/cooking with magnets! from there?)
They conclude by saying that a simpler idea might be to assume that everyone is working within some degree of model framework, and that experiments are best described not as hypothesis falsification or model testing, but as questions about that model.
At least, I think that's what they said. Seriously, I just want an induction stovetop.
Or, as Advisor pointed out, no less an authority than Richard Feynman said, "Philosophy of science is about as useful to scientists as ornithology is to birds."
ps) Biden? Huh. I think he'd be good in office at helping O navigate Senate politics, but selecting him may make it harder to get into office to find out.
15 years ago
8 comments:
I have a book by David Glass called "Experimental Design for Biologists". Frankly, it's got to be one of the more informative books I've ever read on biology (maybe I should suggest it to Nat Blair?). His premise is that in law school, people are taught how to be lawyers. In medical school, people are taught how to be doctors. This does not apply in biology graduate school, where you are taught about biology but not actually how to be a scientist. It goes through some basic experimental designs, what controls are useful, and general advice on how to think about experimental setup.
Not having gone through a PhD program, reading this book has to be one of the most helpful things I've done for myself. He's a very engaging writer and it's easy to breeze through. I haven't read the Cell article, but I suspect it's gotta have a reference to this book in it.
I think I can sum up his "stovetop" idea here - if you wanted to get to Carnegie Hall, would you set up hypotheses to falsify ("I think that if I go down 8th avenue then turn left at 4th street, I will get to Carnegie Hall"), then test each and every one of them to determine which will get you there? Or would you ask the question, "How do I get to Carnegie Hall?", which doesn't require iterative hypothesis testing, get an answer from a New Yorker, then get there.
Apologies for the long-winded comment. I really do think it's a great book, and I bet even PhDs might find it useful.
This does not apply in biology graduate school, where you are taught about biology but not actually how to be a scientist. It goes through some basic experimental designs, what controls are useful, and general advice on how to think about experimental setup.
I don't know whether this book is useful, but this particular point is a total load of fucking bullshit. How to be a scientist is exactly what you are taught in a science PhD program, unless it's a completely fuckwitted one!
And by the way, I'm with Feynman. All this navel-gazing about hpothesis-driven model-based whatthefuckever just makes my eyes glaze over. Shut the fuck up and design properly controlled experiments whose results tell you something about how reality functions.
It is interesting since this discussion seems to come out more firey every time. I wonder if this has to do with the lesser amount of philosophy that grad students read today - and that is existing over all in academia today?
I think there is a vey important part in stating a hypothesis and then test it. However, it will mostly be done by starting out from the "world as we know it today" .. .
I think some research is very much driven by the "we'll try and make up a hypothesis after we've found something". Not always the best thing, imho.
Although, sometimes the discussion about science of philosophy can get very introspective and navel looking...
I'm going based on the book's introduction, PhysioProf, where he indicates that most US grad programs don't have formal instruction in experimental design. It's totally possible that many programs have better courses in designing smart experiments than he suggests. Having not attended grad school, this clearly mapped out method book for experimental design has at least been pretty useful for me.
I'm going based on the book's introduction, PhysioProf, where he indicates that most US grad programs don't have formal instruction in experimental design.
I think both of you are half-right. Learning how to design and assess experiments is the bulk of what you learn in grad school. (If you have a clue -- otherwise you think it's about learning techniques or learning what's already known.) But you typically learn it in an informal, pragmatic way as you slog through failed experiments instead of being told up front how to think things through.
What is really not taught at all in basic biology research is proper statistical methodology, which is conducted with a level of sloppiness and cheating that would lead to legal action against clinical trial statisticians.
What is really not taught at all in basic biology research is proper statistical methodology, which is conducted with a level of sloppiness and cheating that would lead to legal action against clinical trial statisticians.
Unfortunately, this is true, and is a complete fucking disgrace. I cannot tell you how many times I have had to explain to motherfucking POST-DOCS that you can't just do eleventy bajillion separate t-tests to assess paired comparisons in a multi-level experiment. AAGH, even thinking about this shit makes my blood boil.
It also varies a great deal by lab (since much of your grad training comes in lab) how well your mentor teaches you to think about science and design experiments. Some are good at doing this, some are good at doing this but bad at teaching it....
Stats: Should be taught better, but honestly by grad school most of us have reached the stage where we're only going to learn something that's useful and relevant. Stats profs should be encouraged to have students bring in actual data they've acquired, rather than working insipid problems out of the book.
I loved this article.
[as an aside to dgt, he doesn't mention it, but I'm going to check out that book.]
I disagree with PhysioProf and with Feynman and your advisor.
Scientists should care about philosophy. If you can't be self-aware enough to question how you question the world, how thoughtful are you, really?
I think this is actually a major problem plaguing science today. Too many scientists are just plowing along, nose to the grindstone, not bothering to even look where they're going.
None of this article is about experimental design or statisticaly analysis, though.
This article is about how you formulate and frame a question and go about answering it, and the history of philosophy on that topic.
I think it's very important to remember that the concept of hypothesis is not the oldest one in science, although in my field people talk about it like it has existed since before the dawn of time.
I liked the conclusion these authors came to, that both hypotheses and models have some pitfalls and limitations.
Asking a question is much more open-ended and, when done right, avoids all the biases introduced by the way we're taught to frame a hypothesis.
I see this all the time in my lab. My advisor is not self-aware enough to recognize that the lab is too insulated and self-congratulatory, and this leads to giant snowdrifts of bias.
I think the key to understanding is to always question. Question your hypothesis. Question your model. Question where it came from, the data it was based on, what parts are solid and what parts are wobbly.
And question why this is not part of the science curriculum, and why we have so many PhDs running around who not only don't know the difference between hypothesis and model, but who think the difference doesn't matter.
Post a Comment