[lmm]

> There's a solid trend in academia that is "publish early, publish fast".

With an increasing number of researchers isn't that a great thing? If you publish your little incremental innovations then everyone can build on them. If you try to build your giant all-encompassing framework before you publish any of it, you'll have to do it all yourself (if it even amounts to anything).

[Fede_V]

It would be if what got published were plenty of incremental papers. That's the case in certain fields (deep learning comes to mind) - but in other fields, such as the life sciences, you instead get a bunch of results that claim to be groundbreaking and novel - but in practice end up being very difficult to reproduce (like the STAP controversy).

Why? Because small incremental improvements cannot get published in high impact journals, and high impact journals are the currency to scientific prestige/grants/tenure.

[jamesrcole]

You're assuming that all important innovations are incremental. Historically that is clearly not the case.

[lmm]

Not at all clear to me. Examples?

[roymurdock]

There's a popular paradigm shift theory that posits that we tend to adopt frameworks that explain the world, then work incrementally within those frameworks until they cease to make as much sense/have as much explanatory power as they once did.

Then someone (Newton, Darwin, Einstein) comes along and assembles a new framework based on new observations/theories that do a better job of explaining things. Eg the switch from creationism to evolution, or from newtonian physics to relativity.

Here are some more examples:

https://en.wikipedia.org/wiki/Paradigm_shift#Examples_of_par...

[mannykannot]

There is an important truth here, but be aware that the popular history of science and technology over-emphasizes the 'lone genius' narrative (it is the 'lone' part that is exaggerated.)

[roymurdock]

Agreed, but that's the story we like to hear, so it's the one that gets repeated. That nobody lives and works in a vacuum is an implicit assumption.

[lmm]

The switch to relativity seemed pretty incremental to me. Einstein's special relativity work builds hugely on that of Lorentz published one year before.

[jamesrcole]

Could Einstein have justified his course of research as an attempt to incrementally innovate on some existing work?

[lmm]

Sure. "I'm trying to figure out a more consistent way of applying the Lorentz contraction" or something on those lines.

[jamesrcole]

Incremental innovations of the sort you are talking about means extending some pre-existing framework. There are obvious examples of important developments that involved developing different frameworks, which took many years to fully justify with evidence. The heliocentric view of the solar system, Newtonian physics, natural selection, relativity...

EDIT: I would also argue that developments like the printing press and the world wide web would have been difficult to justify as incremental extensions of existing work.

I'm not saying these developments came from nowhere, nor that they didn't build on previous work. I'm specifically arguing against the idea that it's fruitful to just focus on work that would be seen as incremental improvements on existing work.

[jmmcd]

If the printing press or web were proposed as research topics by a new PhD student today, they'd probably be laughed out of the room. But if they were proposed by someone with credibility, who knows what research is like from top to bottom, in a EUR2M grant proposal with a good demonstration of how the new ideas relate to existing work, featuring an appropriate-sized team and a detailed risk management plan up-front -- they'd be funded.

[mattkrause]

This implicitly "punishes" subfields that move more slowly. It can take over a year to train a monkey to perform complicated behavioral tasks. In the same time period, you could breed 50+ generations of fruit flies, or ~25,000 generations of bacteria. Furthermore, the monkey researcher will have fairly little interim data, whereas the bacteria or fruit fly work may have something interesting within a few weeks.

This could be normalized within fields, but in practice it's not. For example, the NIH K-awards have a fixed eligibility period, which seems to keep shrinking.