Hacker News Reader

Comment by tombert | Hacker News Reader

I think the title of this article is a bit unfair; pretty much all of our understanding of science is going to be wrong at some level at some point in time, either due to mistakes in the experimentation, or in this case a software bug; good scientists realize this and issue corrections or retract the paper. Doesn't "fake news" imply dishonesty?

View on HN

I really don’t believe it’s fair to call journal articles or letters published in peer-reviewed academic journals, “Fake News” if they turn out to be wrong, or if the authors retract the paper. Part of the larger process of scrutinizing scientific research, is the interactions among scientists, even with competing ideas, in such publications. Where I think we cross the bounds, into “Fake News” territory, is when, immediately upon publication in a Journal, a University PR department, pushes out a release to all the media, often with WILD claims, not representative of the research itself. Or publications such as Scientific American, pick up the paper, and push out a poor Laypersons analysis, without any critical review.

Not necessarily, which is why I've always disliked the term, even apart from how it's used so often as to be meaningless rhetoric. Plenty of people might take "fake" to mean false, regardless of intent. And I would also argue that "fake news" can also be interpreted to mean that something mundane/previously known is falsely insinuated to be shocking "news" e.g. "Barack Obama discovered to have lived with single white teen mom", which is technically true (his mother was white and living alone at the time of his birth), but the intent of the headline is obviously to imply scandal.

That said, I don't think it's normal for a "software bug" to result in such an egregious error as to require a retraction of the study. I'd like to know why the peer review process failed to detect the problem.

>That said, I don't think it's normal for a "software bug" to result in such an egregious error as to require a retraction of the study. I'd like to know why the peer review process failed to detect the problem.

Only because most papers never get the work checked at that level. Software bugs causing errors in studies is extremely common. There's a reason a tiny tiny fraction of researchers share their code without kicking and screaming, and it's not because they'll be scooped on their next paper if they do. (I'm slightly jaded here)

In what field? I work in research and people love to share their software and have you cite their tools in your papers. Software has no chance of being popular or vetted enough to become a standard without sharing it.

That is one good example, and there are countless more. Most people realize how the existence of “fake news” begs the question of what constitutes “real news”, but it’s become frighteningly apparent that very few acknowledge that trust requires more than mutual responsibility. Most importantly, it requires universal forgiveness and a right to interrogate faith. The acceptance of existing “fake news” is not a good sign.

I don't know, I might push back here. While fake news can certainly be interpreted as news that is false, that is not how it is usually used. There is definitely an insinuation, and while the vanilla reading of the definitions of "fake" and "news" don't contain this insinuation, it exists none the less.

That being said, as a writer I might not be able to resist pointing out the irony that the paper on fake news was fake news :). I really can't fault the author.

It's not just an insinuation, the very definition of the word fake includes dishonesty:

/fāk/

adjective

1. not genuine; counterfeit.

"fake designer clothing"

synonyms: forgery, counterfeit, copy, sham, fraud, hoax, imitation, mock-up, dummy, reproduction, lookalike, likeness;

I think the answer is that peer review is essentially broken for finding specific technical errors, and that most errors like this are never acknowledged and don’t lead to retractions, even when they are found which they often are not.

> That said, I don't think it's normal for a "software bug" to result in such an egregious error as to require a retraction of the study. I'd like to know why the peer review process failed to detect the problem.

A Bug in FMRI Software Could Invalidate 15 Years of Brain Research [1]

[1]: https://www.sciencealert.com/a-bug-in-fmri-software-could-in...

fake news is defined as news the speaker doesn’t like and wishes to discredit, so it has nothing to do with the news itself, but rather with the speaker.

That is certainly not the definition of fake news. Fake news is news that is reporting on things which are not factual as though they are. What you are talking about is simply a dishonest speaker, which obviously has everything to do with the speaker.

I think it’s pretty cool that computing science is a unique exception to this rule. Because computing semantics were constructed from first principles, they can be completely understood in a way that things we didn’t construct, like genes, can’t be. Computing scientists are just people so of course they can make errors, but with diligence those errors can be made extremely rare. Knuth is a living existence proof for this claim.

I'd argue that even before computers pure mathematics gave the same level of precision.

Errors are common in mathematics in practice, though, and even expected to a degree because errors in a deep theorem or proof can be extremely subtle. A complex monograph might contain dozens of proofs and lemmas, all of which are completely interdependent so one mistake can be a major problem requiring significant revisions.

Part of the purpose for publishing in the first place is to see if other people can find mistakes in your logic. This is particularly the case at the real frontiers of science and theoretical mathematics, where truly new ground is being broken. Computer-aided theorem provers would like to improve this, but they are currently fairly specialized tools with limited use.

There are certainly some examples. However a lot of older mathematics was actually quite sloppy formally speaking.

It was the demands of early primitive and unreliable computing hardware that forced developers to focus on provable software correctness: it was the best way to find hardware bugs! Dijkstra has a great monograph on developing early interrupt handling correctly. More people should read it, since as anyone who has worked with Unix signals knows, it's easy to get wrong in subtle and nasty ways.

Edit: Also to speak to the pure math point, Dijkstra's official title after graduating was "Mathematical Engineer."

I'm aware, but I'm more talking about theoretical computer science than practical software and systems engineering. Methods for formal verification and design of software are of course interesting, but quite different from formal verification of theory. What I wanted to point out is that even a very formal field (advanced, pure math) is still plagued by very mundane and very human logical errors. Further, these errors are not easily spotted by machine, at least not with current technology.

Regarding Dijkstra, I personally consider TCS to be a branch of applied math or maybe applied logic. What most programmers do in practice, however, is more akin to engineering (or I'd say, is a type of engineering).

"Err and err and err again but less and less and less" https://www.youtube.com/watch?v=v678Em6qyzk

Reply on news.ycombinator.com