3 ms·

tl;dr: Scientific American published an article about an analog computing technique that the author claims can solve NP-complete problems, but which contains an

tl;dr: Scientific American published an article about an analog computing technique that the author claims can solve NP-complete problems, but which contains an elementary mistake that invalidates the result. Before SciAm published it, an arXiv preprint circulated and commenters on HN and Reddit pointed out the mistake. The editor either did not read these replies, or did read them but decided to publish anyways. This is an embarrassing fuckup on their part, but one that is endemic in mass-market scientific publications, which want to publish sensational results but lack the expertise to recognize them or to summarize results without major distortions. The problem is so severe that I recommend avoiding these sources entirely, because the frequent major errors will effectively leave you less knowledgeable than you started.

j2kun 12y ago
FWIW Quanta magazine (which I think is has a syndication agreement with SciAm) does an excellent job on theoretical CS topics.
pekk 12y ago
The problem is the standard for reviews. It cannot become mandatory to read Reddit before publishing a paper. Reddit has an endemic issue of publishing sensational results with major distortions, Reddit contains frequent major errors which will effectively leave you less knowledgeable than you started. It makes no sense to avoid scientific journals, but then trust whatever you read on Reddit.
- lomnakkus 12y ago
  I don't necessarily disagree, but there's a certain amount of due diligence that editors must do and, as I recall, this idea + rebuttal made a bit of a splash even on non-HN/Reddit media back when the original came out. Even if we ignore that, there's a massive a-priori improbability for the claimed result being true given the current state-of-the-art knowledge by experts in the field. Claiming to "solve" N=NP is definitely a red flag and anyone modestly acquainted with the field, as editors probably should be, would know to consult a larger group of people within the field before publishing this. Let's not forget Sagan's advice to require extraordinary evidence for extraordinary claims. Either way, I think the editors should at least publish a/the rebuttal at this point. (At the very least.) EDIT: I should say: I know this isn't -- per se -- claiming that P=NP, but it's in the same league in terms of boldness since they're claiming that there's a physical machine that can solve NP problems in P, but there's, as yet, no plausible explanation for why no N-TM has yet been realized in physical reality. (Given current QC theory, at least.)
- duaneb 12y ago
  > It cannot become mandatory to read Reddit before publishing a paper. No, but if reddit does better than your review team, your journal is less worth reading than reddit. That's not a good place to be.
DavidSJ 12y ago
From such heights has Scientific American fallen: http://www.federaljack.com/ebooks/Consciousness%20Books%20Collection/Scientific%20American%20-%20The%20Brain%20(1979).pdf http://www.federaljack.com/ebooks/Consciousness%20Books%20Co... (large PDF download ahead)
troymc 12y ago
Even highly-respected peer-reviewed scientific journals make mistakes. The beauty of science (and scientific publishing) isn't that it's always right, but that it's willing to admit it was wrong (given sufficient evidence or reasoning). Scientific American has very high standards. They do extensive fact-checking, even for columns that don't matter as much (such as David Pogue's). Many of the longer articles in Scientific American are written by the scientists who did the work, often stuff that was previously published in peer-reviewed scientific journals. The Scientific American version is written for a generalist audience. Scientific American, the magazine, rarely breaks major sensational science stories. It's not their mission. You say you "recommend avoiding these sources [such as Scientific American] entirely." What do you recommend instead? Reading peer-reviewed journals? Who can read those, other than the specialists? I recommend reading Scientific American, alert to the possibility that something may be wrong. Because there's no such thing as a scientific publication that's always 100% right.

j2kun 12y ago

FWIW Quanta magazine (which I think is has a syndication agreement with SciAm) does an excellent job on theoretical CS topics.

pekk 12y ago

The problem is the standard for reviews. It cannot become mandatory to read Reddit before publishing a paper. Reddit has an endemic issue of publishing sensational results with major distortions, Reddit contains frequent major errors which will effectively leave you less knowledgeable than you started. It makes no sense to avoid scientific journals, but then trust whatever you read on Reddit.

lomnakkus 12y ago
I don't necessarily disagree, but there's a certain amount of due diligence that editors must do and, as I recall, this idea + rebuttal made a bit of a splash even on non-HN/Reddit media back when the original came out. Even if we ignore that, there's a massive a-priori improbability for the claimed result being true given the current state-of-the-art knowledge by experts in the field. Claiming to "solve" N=NP is definitely a red flag and anyone modestly acquainted with the field, as editors probably should be, would know to consult a larger group of people within the field before publishing this. Let's not forget Sagan's advice to require extraordinary evidence for extraordinary claims. Either way, I think the editors should at least publish a/the rebuttal at this point. (At the very least.) EDIT: I should say: I know this isn't -- per se -- claiming that P=NP, but it's in the same league in terms of boldness since they're claiming that there's a physical machine that can solve NP problems in P, but there's, as yet, no plausible explanation for why no N-TM has yet been realized in physical reality. (Given current QC theory, at least.)
duaneb 12y ago
> It cannot become mandatory to read Reddit before publishing a paper. No, but if reddit does better than your review team, your journal is less worth reading than reddit. That's not a good place to be.

DavidSJ 12y ago

From such heights has Scientific American fallen: http://www.federaljack.com/ebooks/Consciousness%20Books%20Collection/Scientific%20American%20-%20The%20Brain%20(1979).pdf http://www.federaljack.com/ebooks/Consciousness%20Books%20Co... (large PDF download ahead)

troymc 12y ago

Even highly-respected peer-reviewed scientific journals make mistakes. The beauty of science (and scientific publishing) isn't that it's always right, but that it's willing to admit it was wrong (given sufficient evidence or reasoning). Scientific American has very high standards. They do extensive fact-checking, even for columns that don't matter as much (such as David Pogue's). Many of the longer articles in Scientific American are written by the scientists who did the work, often stuff that was previously published in peer-reviewed scientific journals. The Scientific American version is written for a generalist audience. Scientific American, the magazine, rarely breaks major sensational science stories. It's not their mission. You say you "recommend avoiding these sources [such as Scientific American] entirely." What do you recommend instead? Reading peer-reviewed journals? Who can read those, other than the specialists? I recommend reading Scientific American, alert to the possibility that something may be wrong. Because there's no such thing as a scientific publication that's always 100% right.