Slashdot Log In
Why the Cloud Cannot Obscure the Scientific Method
Posted by
CmdrTaco
on Thu Jun 26, 2008 08:43 AM
from the because-of-science-dude dept.
from the because-of-science-dude dept.
aproposofwhat noted Ars Technica's rebuttal to
yesterday's story about "The End of Theory: The Data Deluge Makes the Scientific Method Obsolete." The response is titled "Why the cloud cannot obscure the Scientific Method," and is a good follow up to the discussion.
Related Stories
[+]
Google Begat the End of the Scientific Method? 387 comments
TheSauce writes "In a fairly concise one-pager from Chris Anderson, at Wired, the editor posits that all of our current (or now previous) models for collecting data are dead. The content is compelling. It notes that we've entered the Age of the Petabyte — where one can collect immense amounts of data that are paradigm agnostic. It goes on to add a comment from the head of Google's R&D, that we need an update to George Box's maxim: 'All models are wrong, and increasingly you can succeed without them.' Have we reached a time where all of our tool-sets are now made moot by vast clouds of information and strictly applied maths?"
This discussion has been archived.
No new comments can be posted.
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
Full
Abbreviated
Hidden
Loading... please wait.
datasource != process (Score:5, Insightful)
Because a datasource isn't a process?
missing link (Score:4, Insightful)
I like the fact that the web and search/aggregate engines may combine vast amounts of data in ways we now
cannot imagine - it expands the field for new scientific research enormously. Replace science? No.
Re:missing link (Score:4, Funny)
What, you mean I can't just google for "unified field theory" and get the right answer? Why does the universe have to be so hard?????
Parent
Bullshit bingo (Score:5, Funny)
Latest addition to bullshit bingo cards:
CLOUD
It's a good rebuttal (Score:5, Insightful)
I'd say that the models are the science. They're how you explain your data. They provide evidence that the experiments make sense, and they guide you by making predictions you can test.
Moreover, SIMPLIFIED MODELS are good science. Understanding which details can be omitted without impacting the predictive ability of your model shows you know which effects are important and which aren't.
I agree, but... (Score:4, Insightful)
What you say is true, Hoplite3. The big issue I see is how people define "model". My guess is that quite a few unfortunately define it as "I got 3 asterisks in the significance test", whether the "model" (say, linear regression) makes sense or not.
I forget where I read it, but I've been studying linear regression, and there was a fascinating example were if they'd have used linear regression techniques on the early "drop the canonball and time it's fall" data, they would have come up with a nice, highly-significant linear regression for gravity.
Then there is the whole issue of explanation versus prediction. Something can be predictive while providing no explanation, and perhaps that's where the petabyte idea is going: who cares about explanation if prediction is accurate enough? (Not my philosophy, BTW.)
Parent
Re:I agree, but... (Score:5, Interesting)
Yes, I think that prediction without explanation is fascinating, but I don't know if it's what I like about science :) Have you ever heard Lenard Smith speak? I saw him at SAMSI, but his MSRI talk is online and is roughly the same. He's a statistician who works in exactly this.
Some fancy-pants technique he has is better at predicting the future behavior of chaotic systems (like van der Pol circuits or the weather) than physical models. But he also points out that these predictions don't tell you what type of data to collect to make better predictions, and that they don't generalize. One nice "model" he has can predict the weather at Heathrow better than physical weather models (from the same inputs: wind speed, temperature, pressure, etc), but it's useless for predicting the weather in Kinshasa until the model is re-trained.
I think these types of data analysis tools will be very important in the future, but they won't replace the explanatory power of models. Just like how scientific computing is useful, but never replaced actual experiments.
Parent
Re:I agree, but... (Score:5, Insightful)
Thank you. Sure, there's a ton of data out there, but how was it collected? What statistical methods were used to analyze the data? How did you select the data set you're analyzing? Nothing I understand about science really applies to data mining a so-called "cloud". Prediction without explanation is just observation. Observation in and of itself is not science. You might have data, but is it the right data?
I see all this petabyte stuff as interesting and even as a valuable adjunct to real science, but a basic requirement of science is reproducibility and you can't reproduce the data collection.
Parent
Correlation is not causation (Score:4, Informative)
Re:Correlation is not causation (Score:4, Interesting)
The large scale genetic association studies are a great example. There was a day that you could publish a paper solely describing a correlation between a variant in gene X and its association with disease Y. However, because of the way we do statistics in science, sooner or later you'll find a statistically significant correlation simply due to chance alone. In fact the epidemiologist John ioannidis wrote an article [plosjournals.org] about this (that I believe appeared on Slashdot as well). Now you're often required to show some kind of experimental validation that there is a biological basis that verifies the statistical correlation. The scientific method is not going away anytime soon.
Parent
Re:Correlation is not causation (Score:4, Informative)
Parent
Re: (Score:3, Interesting)
Re: (Score:3, Insightful)
In science, the phrase usually used is "correlation does not imply a specific causation." It does, of course, imply some correlation and most of modern science is noticing correlations and testing for causation.
Re: (Score:3, Informative)
Hey, don't try to pin all that stuff on mathematicians: the original cloud-gushing author, Chris Anderson, says, "background is in science, starting with studying physics and doing research at Los Alamos. [thelongtail.com]"
All models are wrong, but .... (Score:4, Insightful)
All models are wrong, but some are useful.
We still need scientific methods to develop useful models and understand and refine the existing models. When Newton defined his mechanics that was the state of the art in his era, and now we have progressed to quantum mechanics which might be refined tomorrow.
But mere observation of some phenomena is not sufficient to postulate the behaviour in a changed condition. A scientific model and its rigorous application is required for this. Correlations drawn from the cloud cannot substitute it.
gopla
Re:All models are wrong, but .... (Score:5, Insightful)
All models are wrong, to some degree. A better way to put it is all models are imprecise, but some are precise enough to be useful. 'Wrong' is a very flexible word and can easily lead to a misunderstanding in this context.
Parent
Nice rebuttal, bad example. (Score:5, Informative)
In general I'm right behind the rebuttal. However John Timmer chooses a very bad real-life example as his rebuttal champion.
He asks: ...would Anderson be willing to help test a drug that was based on a poorly understood correlation pulled out of a datamine? These days, we like our drugs to have known targets and mechanisms of action and, to get there, we need standard science.
These days we may like our drugs to have these attributes, but very often they don't. There are still quite a few medicines around that clearly work and are prescribed on that basis, but for which there is only the haziest evidence as to how exactly they work.
The good thing about the scientific method, however is it gives us a framework to investigate these drug's actions - even if the explanation is still currently beyond us.
Marketing is not a Science (Score:5, Insightful)
Mr. Anderson was not prescient in any way, he was just speaking his perspective. The only thing is we must be careful to even consider his proposition as a valid reality worth pursuing. Not for true scientists, but from a social perspective, or it will truly be the end of science. There are some in power as it is already attempting to make this happen.
That said, I almost consider responding to yesterday's article as falling for the argument. But, since it hit the
I'm moonlighting in bioinformatics (Score:5, Interesting)
And can back up this rebuttal with a practical example. I am a physicist, I know sod all about blood samples, or proteins, or cancer. I get a pile of mass spec data (about a billion data points or so on some days) and through binning, background subtraction, and a string of other statistical witchcraft I produce a set of peaks labeled according to intensity and significance.
This does not make me a cancer researcher. This data has to go back to the cancer guys and they have to pick out the Biomarkers and thus develop new diagnostic tests, based on principles that I don't understand. I am master of the information but entirely blind as far as the science is concerned. Same goes for google.
Duh! (Score:5, Insightful)
Rise of Engineering over Science? (Score:5, Interesting)
I have always viewed this debate in the context of scientist vs. engineer. That is one who views data as "good and true" vs. "good enough". That's not a slam on engineers (I am one), but a reflection of the balance between the two. A scientist that never applies theory sits in an empty room. An engineer who build things with out science, sits in a cluttered room surrounded by useless objects.
I do find interesting though that the advent of "google data" may indicate a flip in order of the two disciplines. Historically (IMHO) science has led engineering. A theoretical breakthrough, provable by the scientific method, may take years to give birth to a practical application. Now, with enormous piles of data and the knowledge that "good enough" is often good enough, we may be creating useful objects that will take science many years to explain and model.
The biggest issue and omission in both of these pieces is that this "cloud" of data does not represent "truth" (as the scientist may seek), but rather a summation or averaging of the "perception of truth" as seen by the individual authors. The cloud, therefore, is only as useful as human's ability to divine truth without the scientific method.
My two cents. :)
Re: (Score:3, Insightful)
I have a theory that some of the best engineers are scientists, and some of the best scientists are engineers.
Scientists often need to build crazy stuff to figure things out, and engineers often need to figure things out to build crazy stuff. Because they are each result oriented, they don't get hung up on the things that someone in field would.
knowledge != understanding (Score:4, Insightful)
I have a problem with the google generation, sure, they can parrot facts and find things in an instant, as can any slashdotter I'm sure, but knowing something is not the same thing as understanding something.
I coworker asked me yesterday "how do you call a C++ class member function from C [or java]?" The question is an example of pure ignorance.
If they "understood" computer science, as a profession, this would be a trivial question, like how do I or can I declare a C function in C++. The second question is what google can help you with while having to ask the first question means you are screwed and need to ask someone who understands what you do not. Not understanding what you do for a living is a problem.
How programs get linked, how environments function, virtual machines vs pure binaries, etc. These are important parts of computer science, just as much as algorithms and structures. You have to have a WORKING knowledge of things, i.e. an understanding.
Google's ease of discovery eliminates a lot of the understanding learned from research. Now we can get the information we want, easily, without actually understanding it. IMHO this is a very dangerous thing.
science-open , clouds-? (Score:3, Insightful)
Science and openness go together.
Without openness, we all are reinventing private wheels, which we destroy the plans to when there is no profit.
If you work in software, consider for a moment how scientific your work is, considering the work of other companies doing similar work.
This Clouds thing is the "billion monkeys/humans typing on keyboards" model.
Yes, it really can work (with humans).
But, as with science, the chaos development model only works with openness.
Of course, organized science along with a little chaotic development work work even better.
There are forces in our society that do not like any open model. The Microsoft's, the MPAA, the RIAA. These type of organization thrive from closed models. More copyright controls, more DRM, longer copyright and patent terms.
These forces would prefer to own,control and close science and clouds of data. They are unaware of the inevitable impact of such actions.
In a free capitalist society, we are naturally driven my contrary forces.
A desire to hide discoveries, to maximize profits, even at the expense of innovation.
A desire to share discoveries, to contribute to society and for credit.
While it is possible to profit when ideas are shared,
It is more difficult to contribute to society by hiding information indefinitely.
Re:Don't blame the author's incompetence (Score:4, Insightful)
Truly what yesterday's article was saying is that causation or correlation is meaningless if you have a mimic of the real world in the form of a collection of data. You don't need a model that is accurate or valid or anything. You just need to run the data in the exact replica of reality. This is the simulacrum. The first problem is that data does not just run itself. At the least it needs an algorithm to be processed to a result. Thats the model, without its just useless data, which has been mentioned already yesterday in comments. But second, the problem with even ATTEMPTING such an idea is that you lead yourself into a situation where you "predict" the future and then operate to become that future thus destroying the creative nature of humanity and become the self-fulling prophecy of machine code!
Keep in mind i speak mostly of social sciences that try to pattern human behavior. For hard sciences, etc., all you have done is created a simulation of reality, but it tells you nothing about the reality. It merely mimics it. There is no insight into creating a map the size of the United States, at best it is a work of art.
Parent