From personal experience (anecdote alert!), errors are also common in the ostensibly stone-cold-hard field of algorithms in computer science. A few years back I went on a string algorithm kick, and started dredging up old algorithm papers from the 80's on which to build Wikipedia articles.
Often, the papers would get the general idea right, but if implemented as described would not work at all or fail on edge cases. The best example I have is an algorithm to find the lexicographically-minimal string rotation[0]. The simplest and fastest algorithm to do this is based on the KMP string search algo, and is tribal knowledge among ACM ICPC competitors. I thought it was pretty neat and wanted to cement this algorithm in popular knowledge, so I set about researching and writing the Wikipedia article.
I found the KMP-based algorithm in a 1980 paper[1] by Kellogg S. Booth. The paper has very detailed pseudocode which does not work. At all. The tribal knowledge version I inherited had similarities in the general idea of the algorithm (use of the KMP preprocessing step) but everything else was different. I scoured the internet for a retraction or correction, but all I found was a paper written in 1995[2] which mentioned in passing errors in the 1980 paper.
I do wonder exactly how common this is. I emailed a professor who co-wrote one of the papers, and he replied that "it seems to me that all
the algorithms (including our own) turned out to have errors in them!" Has anyone done studies into errors in computer science papers?
There is a point of view that says that computer science conferences exist for the purpose of gaming the tenure system. The name of the game is plausible deniability: you're not supposed to submit papers that contain known false claims, but everything else is fair game. And this has become such an integral part of the culture that technical correctness is no longer a necessary condition for accepting a paper [1]. I think in this light it's quite clear why many scientists are happy to leave their papers hidden behind the ACM paywall.
Thank you, that was a fascinating read. It is understandable that technical errors are given a pass, as they aren't the meat of the paper. In the case of the Booth paper, I really should state I do not mean to attack him. The idea of using the KMP preprocess to solve the problem is a wonderful approach and works very well despite the actual implementation being technically incorrect. If I recall, the bug had to do with the termination condition; the algorithm had to run twice as long to terminate correctly. I will say my understanding of the algorithm improved as a result of debugging it!
I think it's pretty common. Here's a note I wrote regarding Luc Devroye's highly-regarded (and excellent!) book on univariate random number generation:
Although not dissenting from the other reviews which tout the comprehensiveness of the treatment and its level of detail, I have to add an unpleasant fact about the algorithms: the codes may not work as written, and if they don't, there's not an easy way to track down the problem. (This is because of the nature of the constructions used in the complex constant-time algorithms -- this opaqueness is not a problem for the elementary algorithms which, alas, may not run in constant time.)
A look at the author's web site (currently at errors.pdf off his main page) shows that, e.g., the algorithm on page 511 [of the book] for Poisson r.v.'s has four serious bugs as originally published. This means that the main algorithm for one of the most important discrete distributions was not coded and tested by the author before the book appeared!
In fact, I believe this algorithm has at least one more bug, because I'm still seeing a small off-by-one anomaly in my implementation. The algorithm for binomial r.v.'s may have trouble as well -- I see problems for N=400, p=0.05. After 10 million draws (i.e., enough to get good statistics) I see deviations of counts in some bins near the peak (i.e. number of integer outcomes of the R.V.) of 8 standard deviations from the expected number of counts. So, be careful, and consider alternate implementations of the more complex algorithms.
There's a lot of detail in the book, and the techniques are valid. But as we all know, implementation sometimes reveals algebra errors!
Implementations of common algorithms are even worse.
The Java binary search implementation had a bug that eluded detection for 9 years, and it was based on a implementation from the 1986 book "Programming Pearls" that also contained the same bug (TL;RD: it's an overflow error that computers in 1986 would probably never have run into - who could imagine having an array with more than 2^30 elements?!).
Even worse: While the first binary search was published in 1946, the first binary search that works correctly for all values of n did not appear until 1962. - and this bug shows it is likely this 1962 version would have failed the same way.
Curiously, most c programs with 2^30 element arrays will probably have 64 bit ints, and will therefore work. In the time the code original code was written, I suppose the ints would have been 16 bit?
This is a problem in computational electrophysiology as well. There are a lot of typos and other simple mistakes in classic papers. Say you want to implement a finite differences model for a certain type of voltage-gated potassium channel. If you go back to the original paper it's not uncommon to find minus-signs omitted, parentheses placed improperly or other unfortunate bugs. It can take a lot of head scratching and wasted time to get to the point that you can reproduce the figures from the paper!
Granted when something has been in the literature for a long time, the derivative papers and popular implementations (in eg Neuron) are usually right, but there is rarely anything in the scholarly record that documents these errors. It's all tribal-knowldege and side-channels.
Ugh, that sounds terrible. During a previous internship at a HPC company[0] I implemented a computational electrodynamics FDTD algorithm as given in the Taflove book[1], and I made more than enough errors even without the book containing mistakes! Two fields, each with three components and subtly different equations for all. What a nightmare. Especially since it's impossible to tell what is wrong when watching the EM wave propagate in an impossible oblong fashion during your simulation.
The author says "Petrofsky was a lavishly honored star of the IEEE", but I was unable to find any honors he got besides an award for a paper in an IEEE journal (http://ieee-aess.org/contacts/award-recipient/jerrold-s-petr... -- there could be other awards, but they don't show up in google). I thought "lavishly honored" would be shorthand for IEEE Fellow, but Petrofsky is not on the Fellows list.
Saying his work was prematurely made into a biopic starring Judd Hirsch is not an indictment of science...
Also he seems to lay at the feet of Petrofsky the failure of any further progess in the area and even seems to insinuate Petrofsky misled people intentionally. I don't follow him on point one, it seems a non-sequitur. As for the second, I'm going to quote a section from one of the articles he linked to:
“That was just for that event,” Davis, now Nan Huckeriede, said of her brief, but famous walk at graduation. “It was the computer-controlled electric stimulation, not me.”
Davis had met Petrofsky while she was in college. “I was attending the WSU Lake Campus, and went to a spinal cord society conference there,” she says from her St. Marys home. “Jerry (Petrofsky) was a presenter, and afterwards, I introduced myself and told him I was interested in his research.
“For about a month, I drove back and forth to Dayton to work with him, and then I transferred to the Dayton campus.”
Following her graduation walk, Davis returned to her wheelchair, stayed in Dayton a few years, married, and then returned to St. Marys.
“Jerry moved to California and stopped his research — I think he felt that he had gone as far as he could,” said Huckeriede. “But I still use the equipment he developed to get my exercise.”
Last summer she traveled to Beijing for a procedure to strengthen her back and stomach muscles. “It didn’t work, but I knew it was experimental. It was worth a try.”
Nice find. That kind of testimonial from the subject of the work is really important.
Another relevant item. The magazine he wrote for, The Institute, is a general-interest magazine of feature stories related to IEEE members. It's not a technical journal. It's more akin to the feature newsletters published by universities or engineering schools and sent to their alumni.
The general-interest technical IEEE journal is Proc. IEEE, which is peer-reviewed and contains research articles and research summaries written by the experts themselves.
I remember some of the same overhyped news stories the science journalist who wrote the article submitted here remembers. I especially remember the breathless (and false) reports about a "gene for" this or that human behavioral trait. The science news cycle[1] frustrates journalists, because every new study with an incremental finding (which may not even be replicable) has to be hyped up by research organization press offices, in the interest of obtaining more funding.
The author's follow-up on a famous science story from early in his career is thought-provoking. Indeed, editors are more nervous about publishing stories, even very well reported stories, that question good news and expose hype or even fraud than editors are about publishing stories on the latest science hero.
On the whole, it's good news that more and more scientists and journalists are alert to the possibility that a preliminary research finding may be false and overhyped besides. Here on Hacker News, we can keep one another alert by remembering the signs to look for whenever we read a new research finding news story.[2]
Hacker News readers who want to learn more about how research articles become retracted may enjoy reading the group blog Retraction Watch[3] compiled by two veteran science journalists with lots of help from tipsters in the science community. I think I learned about Retraction Watch from someone else's comment here on HN.
It just would be a very time consuming experiment to run, and the cost of being wrong would be high, but I don't see how it isn't possible to get empirical results about if being a poet would, in fact, make you happy.
It would be easy enough to show that becoming a poet didn't make someone happy. But if they became a poet, and then became happy? Surely its pretty obvious that the connection wouldn't be testable in the slightest.
Actually, the success of science is its ability to predict what will occur in a wide variety of situations. Otherwise science wouldn't have been such a big deal.
If it were possible to specify how much happiness is enough, and how much happiness isn't, precisely, then it would be science. That's the only way to make generalizations from the results. But the threshold of happiness is imprecise, so therefore it's not science.
Reach into your pocketbook, pull out that blank check, write in the figures and a signature, and send it in the mail. Most science research orgs are 501(c)3 organizations - for whatever that's worth - in any case they will be happy to take your money.
you've never met a federal grants administrator, have you? I've met only two non-stupid ones, ever. The first was my neighbor as a child (who was a physics administrator for the NSF who had the foresight to argue to pull the SSC - because of serious problems with the site - to say the least), the second was my former boss, who quit her highly successful position in academia (long story, won't get into her reasons for doing so).
To be less flippant: There's MORE accountability. If you crowd-fund, people will criticize you publically. when was the last time you saw public comment on an NIH grants review? On a DARPA Call for Proposal? If you don't take public money, to a first approximation you get ONE chance to screw things up by dazzling people. The next time, people are more likely to be wise onto your shenanigans.
Think of it this way: Yes, there are spectacular failures on kickstarter. But yet, there are still good projects that get funded. The economic model, despite all of its naysayers, STILL works to produce good content. Think of all the video games that have successfully migrated to the platform with private micropatronage, and contrast with this: http://www.slate.com/articles/business/moneybox/2012/06/_38_...
Now, consider that this may be happening in publically funded academic science (Arsenic Life, for example).
I agree that a kickstarter for science would be an interesting experiment. It would be quite a change for scientists, if they have to propose to the public instead of experts and politicians.
I'm in Germany, so no direct experience with the US system. Nevertheless, the symptoms are the basically the same here.
Well, I'm biased, since that's what I'm doing (details in my profile). I have strong opinions about how it should be done to be honest with your donors.
I think it makes sense to use tax money for science which benefits the general population. The distribution is tricky of course, but it is with all tax money.
Gonon, F., Bezard, E., & Boraud, T. (2011). Misrepresentation of Neuroscience Data Might Give Rise to Misleading Conclusions in the Media: The Case of Attention Deficit Hyperactivity Disorder. PLoS ONE, 6(1), e14618. doi:10.1371/journal.pone.0014618.t003
They argue that a few prominent examples of misreporting in the media come from researchers misrepresenting their results in their abstracts, and journalists rarely read past abstracts.
Interesting. Seems maybe journalists should rely on researchers other than a study's authors for interpretation? I don't mean that sarcastically (well, a little), but it seems that there's a fairly simple solution to this problem.
> I wrote a puff piece about Petrofsky–based primarily on interviews with him and materials supplied by him and Wright State–published in the November 1983 issue of The Institute, the monthly newspaper of the IEEE
My, could this be part of the problem?
> It never occurred to me to question Petrosky’s claims.
Or this?
Maybe, just maybe, it's very hard to get good reporting on something by people unqualified in a subject?
He criticizes himself for this too. But he didn't make the guy famous; Petrofsky was already a star, the subject of laudatory movie, and so forth. He became more skeptical and challenged Petrofsky's claims in print in 1985. I think that's a pretty good turnaround time.
>“Academic scientists readily acknowledge that they often get things wrong,” The Economist states in its recent cover story “How Science Goes Wrong.” “But they also hold fast to the idea that these errors get corrected over time as other scientists try to take the work further. Evidence that many more dodgy results are published than are subsequently corrected or withdrawn calls that much-vaunted capacity for self-correction into question. There are errors in a lot more of the scientific papers being published, written about and acted on than anyone would normally suppose, or like to think.”
I think the fact that people rarely retract can also be more of a practical issue than anything. Some guy wrote his master's thesis 3 years ago and it's in fact wrong? The author's too busy with his real life to have been keeping track of that. Or he's working on some other domain. Or the paper was written 30 years ago.
I've also heard a lot of people say that a lot of research happens in the dark in many domains. People doing research in Haskell will gladly talk about their work on mailing lists it seems, but when it comes to chemistry, apparently it's a whole lotta silence. Pretty depressing.
One of my "pet peeves" (that has actually caused me a fair amount of aggravation and "expense", personally): Absence of evidence is taken (or, insisted upon) as evidence of absence.
I've encountered this particularly in the medical community. As sort of a TL;DR: Many medical practitioners seem at best to be... "technicians" who are not much capable of more that following the "current script" that is handed down to them from blessed authority figures (including particularly if not only the pharmaceutical companies).
P.S. I'll add insurance companies to the mix of authority figures, particularly in the U.S. They categorize and dictate what they will and won't pay for. Better doctors sometimes spend a lot of time finding ways around these restrictions in order to provide the treatment they think is actually appropriate and optimal.
I would understand using statistical evaluation to help determine the best treatment approach. But when the profit motive enters, combined with a more or less fungible population of insured, the number crunching seems often to put cost ahead of outcome.
I did not take this as an anti-science article, but as an article critical of how academic science/engineering has become hype driven. Perhaps it was always hype driven, but I find it to be particularly bad now. Before you could do 'regular' work i.e. work which was scholarly and did not have terms like 'first to show' and which did not appear in the tabloids 'Science' and 'Nature' and still advance your career and get funding. Not so much any more.
In other words, science is hard, predictions are unreliable, scientists are humans and humans make mistakes / are motivated by emotions.
As usual with pieces critical of science they focus too much on a very small number of bad eggs and seem to implicitly assume that scientists should somehow be superhuman.
Even if there are a lot of bad eggs and hype and bullshit, you have to ask yourself what the alternative is. The non-science-based existence we suffered through for millennia? I think not. Excuse us for trying to cure cancer and failing less than 100% of the time.
Part of the reason there is so much hype and bullshit is because, if we weren't cramming it down everyones throats for the 30 seconds they'll pay attention, there would be no money funding science and we'd still be living in our own filth and praying to god that the plague stops.
That's a total misrepresentation of the complaint. The complaint is that claiming to cure cancer or being close to curing cancer to get some funding hurts the credibility of Science as an institution.
Part of the reason there is so much hype and bullshit is because
How will training the public that Scientists are money-grubbing hucksters who are full of crap help the matter any?
A lack of humility and self-criticism is a huge problem in any discipline, especially one that claims to be the best way to learn the "truth".
How will training the public that Scientists are money-grubbing hucksters who are full of crap help the matter any?
Also: It just might teach the public to actually take agency over who gets funded and encourage them to decide for themselves who is or isn't a huckster.
Incidentially, I am trying to cure cancer, and I've set up a nonprofit to do so... And am considering writing an piece explaining why you shouldn't donate to me. (if you can't take the risk of failure, etc.) What do you think? Although I'm being genuine, is it too humblebraggey?
It is valid to ask, if we could do better. I think we can. For example in computer science: Less funding for big projects, but more smaller proposals. Do not consider past publications (aka authority) so much for funding decisions.
But the article doesn't criticize science, in fact, apart from the word, science isn't even mentioned. The article criticizes scientists, science boosters and science journalists. None of those relate very well or consistently to science itself.
The problems discussed in the article, having primarily to do with the difficulties in finding out what is actually so, is why science exists. Science is a (not the)solution to those problems. But to get the solution, one must learn how science works. That isn't even touched on in the article.
I think it's really important to talk about what Science is. It gets bandied about like it's this abstract idea when, in fact, it is a real process that is taking place. A good definition of Science is humanity's current working knowledge of reality based on the process of lots of people utilizing the Scientific method to test hypotheses. I think that's fair definition. On examination of this definition you will notice it has a large human component.
Science is like the stock market. It's lots of people spit-balling about what's happening in the market (in Science's case, the market is "ideas about reality"). In the short term Science can look really ugly, just like the stock market can, in the long term, however, we'd like to think of it as an accurate weighing machine.
I generally think this is a fair assessment, but as I get older and start to see how few people there are who aren't willing to cut corners to get ahead. This worries me, because, like the stock market, Science affects real peoples lives. It's all well and good that over the course of 100 years the Dow Jones will outperform cash or, really, any other investments, but that's of little use to the real people who get left behind in periods of great economic stagnation. Science can go through similar periods of stagnation, and currently, it seems like we might have hit upon one.
It is possible to criticize the current way we have set up the Scientific endeavor without criticizing the abstract notion that human beings will generally discover new things over the long term. I think it is hardly controversial to say that our current way of doing things is not the best, but it may even be bad. Money and time are corrupting factors. Postings that used to require a PhD require a post-doc, positions that required a post-doc, now require two. The immense pressure of publish-or-perish is becoming greater and greater and room for failure, which is an essential part of the Scientific method, is being squeezed out. This is not a good thing and, I think, is a larger reason, among others, why Science is becoming noisier and noisier. When the stock market becomes noisy it benefits insiders, but hardly anybody else. I think Science is, currently, in a similar place, it's efficacy is being diminished by crap.
> I think it's really important to talk about what Science is. It gets bandied about like it's this abstract idea when, in fact, it is a real process that is taking place.
But science isn't defined by its process, it's defined by its philosophical axioms, its foundational rules. The first and most important rule is that a scientific theory must be potentially falsifiable in practical tests -- if there's no empirical testability, there's no basis for falsification, therefore there is no science. The second rule is that scientific ideas cannot ever be proven true, only false. The third rule is that an idea without supporting evidence is assumed to be false, not true (this is known as the "null hypothesis"). The remaining rules are comparatively unimportant -- these are the big three, without which any discussion of science is pointless.
Science's process can change, and from field to field, it certainly does. But the rules stay the same.
> Science is like the stock market. It's lots of people spit-balling about what's happening in the market ...
That is not science. To call that science is like confusing a spacecraft with a conversation in a bar about a spacecraft.
> It is possible to criticize the current way we have set up the Scientific endeavor without criticizing the abstract notion that human beings will generally discover new things over the long term.
Again, that is not science. Science's goal is not discoveries, its goal is to reliably refute ideas that are false, primarily by comparing them to reality. This is why science journalism articles that trumpet breakthroughs, with rare exception do a disservice to both science and journalism.
Science doesn't have such axiomatic rules; did Galileo lay out these these axioms and proclaim the scientific revolution? As I understand it, science is a human enterprise with the goal of understanding principles... with limitations and strengths.
Falsifiability in particular has serious criticisms, in terms of people taking it as a defining part of science. (http://en.wikipedia.org/wiki/Falsifiability#Criticisms) I suspect (and it's just pure suspicion for now which I'm mentioning for no particular reason) it tends to be emphasized in cultures interested in debunking people's claims in a competitive debating way, rather than constructive conversation where both parties aim at coming to new understandings. I don't mean in science, but cultures influenced by science's success.
>Science can go through similar periods of stagnation, and currently, it seems like we might have hit upon one.
Most of the recent criticism of science is related to the biological and medical sciences. I would disagree that these fields have hit on a period of stagnation. On the contrary, I believe that the biomedical sciences are in a period where there is a massive amount of recently generated data and innumerable ways to find correlations among these data. The data themselves are also recorded in a wide number of different formats and data from one paper might be difficult to compare to that from another (giving possibly unwarranted suspicions of non-reproducibility).
In the midst of this maelstrom of data, individual scientists are trying their best to make sure they maintain their publication rate and grant funding. As a consequence, findings are sometimes published "too early". While one might criticize this trend to early publication, the reality is that it is sometimes impossible to know exactly in which direction the "final answer" (if there is one) lies. Retaining data until one had found the complete and final answer to something is also a disservice to other researchers.
I appreciate the public debate on this matter and hope it ends up working both ways, i.e. the public and science journalists get to understand better the constraints in basic research, and the scientists are given a little more support to undertake more thorough studies.
With regards to the latter point, I would refer you to a review on studies involving mouse models of Alzheimer's disease [0]. In that review, the argument is made that almost all studies on this topic are under-powered (in a statistical sense), and most labs should be using 5 to 10 times more mice per experimental condition. This would more than double the cost of each experiment. For constant funding, this would mean halving the number of labs studying Alzheimer's disease, which might not be such a bad idea if the data coming out was more reliable as a result.
Nobody ever got a raise from their Dean or an endowed Chair because their work was celebrated for being careful, thoughtful, measured, balanced and realistic.
Plenty of science that was uber-hyped at the time has turned out to be misguided and/or even outright wrong. Plenty of those scientists have led wealthy, rewarded lifestyles as a result of the hype.
As an academic scientist one has to make a conscious decision to play the game or not play the game. There isn't a lot of room in the middle. You make your choices and you live with the consequences.
You see your colleague making double your salary, you read the press office reports hyping their work, you understand that it's no more innovative, important, or TRUE than your work or anyone else's in your cohort ... but they are playing the game.
Wouldn't you like to take your family to Hawaii for vacation? Wouldn't you like a bigger house? A nicer car? To send your kids to private school? Your University press office is practically going around begging for science stories to promote (i.e. hype). It's difficult to resist jumping in with both feet.
Reading through this article reminds me of why I'm so critical of science journalism: it's in the attention business, not the science business, and directing undue scepticism about so-called "breakthroughs" would kill a lot of great "stories".
Often, the papers would get the general idea right, but if implemented as described would not work at all or fail on edge cases. The best example I have is an algorithm to find the lexicographically-minimal string rotation[0]. The simplest and fastest algorithm to do this is based on the KMP string search algo, and is tribal knowledge among ACM ICPC competitors. I thought it was pretty neat and wanted to cement this algorithm in popular knowledge, so I set about researching and writing the Wikipedia article.
I found the KMP-based algorithm in a 1980 paper[1] by Kellogg S. Booth. The paper has very detailed pseudocode which does not work. At all. The tribal knowledge version I inherited had similarities in the general idea of the algorithm (use of the KMP preprocessing step) but everything else was different. I scoured the internet for a retraction or correction, but all I found was a paper written in 1995[2] which mentioned in passing errors in the 1980 paper.
I do wonder exactly how common this is. I emailed a professor who co-wrote one of the papers, and he replied that "it seems to me that all the algorithms (including our own) turned out to have errors in them!" Has anyone done studies into errors in computer science papers?
[0] https://en.wikipedia.org/wiki/Lexicographically_minimal_stri...
[1] http://www.sciencedirect.com/science/article/pii/00200190809...
[2] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.55.9...