5 ms·

Incentives that work are easy. Do research in corporate labs. That's all it takes. Now there is an incentive to do work that replicates (because it is intended

by nvm0n2 3y ago

Incentives that work are easy. Do research in corporate labs. That's all it takes. Now there is an incentive to do work that replicates (because it is intended to be used to build useful things), there are people responsible for detecting and resolving fraud (managers), they are incentivized to do so by a mix of carrots and sticks some of them legal. Fact is, corporate R&D doesn't have this relentless problem with reproducibility. It's academic output that does, because academics only care about getting a paper published and don't expect that anyone will use their results. Often they don't even make their code or data available at all because it's not to their advantage for others to be able to replicate their work, as that would yield fewer papers. But this is all wrong. Science exists to be used in technology, not for its own sake.

jampekka 3y ago
There are corresponding responsibilities in academia. Perhaps even more so. But like a business culture can, and often does, rot the incentives and responsibilities, so has the academic culture. I've worked for two corporate labs and have collaborated with several. There was less rigor if anything. Corporate lab work is quite cozy and stable. In corporate lab your job doesn't end if you don't (pretend) to get a new major discovery every year. They don't make anything available, often not even internally. Science is about a lot more than technology.
- BurningFrog 3y ago
  > like a business culture can, and often does, rot the incentives and responsibilities, so has the academic culture. Maybe the problem is that there are many corporations, with different cultures. But there is mostly just one academic culture across the whole system.
- nvm0n2 3y ago
  Hmm I've had the opposite experiences. Corp labs release cool and useful stuff all the time. On the front page right now is Seamless, a very useful thing released by a corporate lab. And corps have performance evaluation and management programmes designed to encourage high performance output. And of course the competent ones manage to bring research into production regularly. Whereas with academics, you get a paper. You might get data and code, or might not, depending on field and temperament of the researchers. If you're really lucky that data/code might actually be correct, match the paper and be usable for something, but it really depends a lot on the field. You almost certainly won't get products.

by nvm0n2 3y ago

jampekka 3y ago

There are corresponding responsibilities in academia. Perhaps even more so. But like a business culture can, and often does, rot the incentives and responsibilities, so has the academic culture. I've worked for two corporate labs and have collaborated with several. There was less rigor if anything. Corporate lab work is quite cozy and stable. In corporate lab your job doesn't end if you don't (pretend) to get a new major discovery every year. They don't make anything available, often not even internally. Science is about a lot more than technology.

BurningFrog 3y ago
> like a business culture can, and often does, rot the incentives and responsibilities, so has the academic culture. Maybe the problem is that there are many corporations, with different cultures. But there is mostly just one academic culture across the whole system.
nvm0n2 3y ago
Hmm I've had the opposite experiences. Corp labs release cool and useful stuff all the time. On the front page right now is Seamless, a very useful thing released by a corporate lab. And corps have performance evaluation and management programmes designed to encourage high performance output. And of course the competent ones manage to bring research into production regularly. Whereas with academics, you get a paper. You might get data and code, or might not, depending on field and temperament of the researchers. If you're really lucky that data/code might actually be correct, match the paper and be usable for something, but it really depends a lot on the field. You almost certainly won't get products.