3 ms·

life sciences is just as bad if not worse -- you can read news stories about many cases of outright photoshopped images in biology papers by respected researche

life sciences is just as bad if not worse -- you can read news stories about many cases of outright photoshopped images in biology papers by respected researchers in major journals. physics has a good reputation, but then you hear of things like apparent fraud in that quantum computing paper about Majorana fermions... computer science actually seems relatively okay as far as avoiding outright fraud, probably because there's such a strong norm of sharing code. plenty of questionable research practices though.

mike_hearn 3y ago
Life sciences is beyond the pale :( Lots of papers coming out of China are entirely fake and made to order by professional paper forging shops. Doctors have to be published to get promoted there, even if they are just regular hospital doctors and not professional researchers, so it leads to really bad incentives. Even the Chinese government's head of research integrity was found to be engaging in research fraud! But, nobody can do anything about it, as journals aren't interested in becoming research audit organizations. So they just publish floods of fraudulent work discussing non-existent interaction effects on random proteins and stuff. Public health is in even worse shape. During COVID I was forced to conclude nobody in power was actually reading the papers that were driving policy. They'd regularly have major errors in them, like model code that was full of bugs or where things were stated about the real world as if they were clearly 100% true but the underlying paper stated the models had unrealistic assumptions! (same people making both statements). Universities didn't care about any of this. I never saw outright fraud in CS but I've seen things that were getting close to the line in cryptography research. Some years ago there were a spate of research papers on Intel SGX that kept claiming they'd broken the security. When examined closely they all used an identical version of GNUtls in the enclaves for their proofs of concept. I got curious about why this specific version of this not very popular cryptography library kept cropping up as the reason for using it never seemed to be mentioned in the papers. Turns out that it's an ancient version, the last version before side channel mitigations were added. The techniques they were claiming broke SGX only worked if you ignored basic cryptographic best practices and used obsolete libraries. And last year there was a non-SGX case where the researchers claimed to be able to extract private keys over the network via side channel attack. The demonstration required the use of a very specific algorithm nobody actually uses, and required the target server to not be doing anything else, and for it to be saturated with traffic that did nothing except invoke this specific obscure algorithm and even then it took days. So this wasn't a realistic scenario anyone needed to worry about but it was presented as if it was.

mike_hearn 3y ago

Life sciences is beyond the pale :( Lots of papers coming out of China are entirely fake and made to order by professional paper forging shops. Doctors have to be published to get promoted there, even if they are just regular hospital doctors and not professional researchers, so it leads to really bad incentives. Even the Chinese government's head of research integrity was found to be engaging in research fraud! But, nobody can do anything about it, as journals aren't interested in becoming research audit organizations. So they just publish floods of fraudulent work discussing non-existent interaction effects on random proteins and stuff. Public health is in even worse shape. During COVID I was forced to conclude nobody in power was actually reading the papers that were driving policy. They'd regularly have major errors in them, like model code that was full of bugs or where things were stated about the real world as if they were clearly 100% true but the underlying paper stated the models had unrealistic assumptions! (same people making both statements). Universities didn't care about any of this. I never saw outright fraud in CS but I've seen things that were getting close to the line in cryptography research. Some years ago there were a spate of research papers on Intel SGX that kept claiming they'd broken the security. When examined closely they all used an identical version of GNUtls in the enclaves for their proofs of concept. I got curious about why this specific version of this not very popular cryptography library kept cropping up as the reason for using it never seemed to be mentioned in the papers. Turns out that it's an ancient version, the last version before side channel mitigations were added. The techniques they were claiming broke SGX only worked if you ignored basic cryptographic best practices and used obsolete libraries. And last year there was a non-SGX case where the researchers claimed to be able to extract private keys over the network via side channel attack. The demonstration required the use of a very specific algorithm nobody actually uses, and required the target server to not be doing anything else, and for it to be saturated with traffic that did nothing except invoke this specific obscure algorithm and even then it took days. So this wasn't a realistic scenario anyone needed to worry about but it was presented as if it was.