Science should be more like this, in current times, yes.
But until much of academia is burned to the ground, or until science can be properly separated from modern academia, this will never be so. The current academic incentives are all wrong: low-quality research is rewarded and results in publications, whereas high-quality research (that takes time, and usually reveals that most exciting publications depend on p-hacking or other highly data-dependent analyses and selective presentations) is not published or actively blocked during peer review.
So instead you get BS arguments about how data can't be released for various privacy concerns (when in reality the vast majority of most datasets are trivial to scrub of identifying factors, and even in more complex datasets where you need to consider k-anonymity, it is still trivial to release data that allows replication of core analyses), and academic science is increasingly irrelevant unless it is tied to tech and industry, where producing junk actually has real negative economic and personal consequences.
I don't know what world this article / post lives in, but it isn't the messy world of actual reality.
> Every result is instantly reproducible. When you read a paper claiming that a new drug reduces symptoms by 30%, you click a link and watch the exact analysis run in your browser. The data processing, statistical tests, and visualizations execute in seconds using the same environment the authors used—preserved perfectly through reproducible containers.
Particularly the "data availability" and "Availability and peer review of computer code and algorithm".
However, in my limited experience, of trying to reproduce certain scRNA-seq processing pipelines, in practice it's never available as just a Github link. I can understand that some/many researcher's code is not in good shape, so I think it'll be quite a stretch to have this available.
I do think it's laudable though, to try and make it available. It would certainly have been very useful for me in the past.
> modern science is synonymous with open source software.
Another problem with reproducibility is the openness of the underlying data. Many academics are terrified of giving away the golden goose and the software is often useless without the data.
However many scientists do work openly, e.g. The Journal of Open Source Software:
Nice read. I believe that traditional software is a great way to showcase the proofs when it comes to physics and mathematics. You can easily code up a theorem in a language of your choice and justify that 'Okay, the output matches the expected value'.
I am particularly fascinated by labs like DeepMind [https://deepmind.google/science/]. The recent advances in their frontier models that are able to predict diseases before they're diagnosed is incredible. This is what AI should be built for and actually do!
Thanks! I appreciate that. Your point about DeepMind and frontier models is spot on. When they "embody"/build on the science done before them we get absolutely mindblowing synergies. But I wonder what happens when the LLMs become way smarter that us: why even loop us in? I guess that's related to the recent field medalist letter https://mathandai.org/
> TL;DR I claim that modern science is synonymous with open source software
That's a strong statement that's not supported by the arguments and IMO misguided.
I don't have a problem with "open", but rather with "software".
Both science and software deal with models, however the focus is quite different. I suspect you conflate theory with models.
The goal of science is to produce and test theories — that's an inductive/abductive process. A model, regardless of whether it's reified into mathematical formulas or software, is a means of making a theory operational enough that its consequences can be derived and confronted with observations.
Software often starts downstream of this: it's a reification of theories, models, algorithms, or findings that are the result of research. Of course software can also be used as part of the research process itself. The distinction is roughly the familiar one between research and development.
I think it is worse than that. Science is a process for resolving disagreements, ambiguity, and uncertainty, and also for discovering abstractions (patterns) among phenomena. It is social and can not be reduced to a binary / digital file, as it is dynamic and ongoing, and, fundamentally, exploratory.
Software is a static program and basically none of these things.
Software development is kind of like science, in some ways, in that you discover abstractions and patterns, and this requires resolving disagreements and ambiguity between you and your users, but in the end, the user demands are usually fairly concrete and specific (though no one may know how to express those demands precisely, initially), and the process is not really exploratory in the way science is.
Thank you for engaging. This is a much more insightful take.
If I'm reading yiur argument right, you're saying that deployed models are downstream versions (reified) of aa theory. Theory, being the actual object of science.
I think this misrepresents science. Science is the ability to build testable knowledge. From that,how would you separate the test from the science? In fact, in an ideal world, why wouldn't you want your theory to be put in a format that's executable? I'm not saying that those things are always the same, my (provocative) title is based on a dream where we can imagine theory and model coexist because software is now a thing.
> Please don't comment on whether someone read an article. "Did you even read the article? It mentions that" can be shortened to "The article mentions that".
TL;DR I claim that modern science is synonymous with open source software. This post explains why, why it matters, and what you can (and should) do next.
Like if CERN published the discovery of the Higgs boson with 99.9997% certainty, but refusing to tell you how they calculated that number, or what equipment they used and how they calibrated it, in order to prevent other labs from copying their methods.
Or like a machine learning lab claiming SOTA on a benchmark, beating a well-known method that they re-implemented, possibly with bugs, on their private dataset, for millions of compute. But you don't get the source to check, and they don't release any intermediate results or ablation experiments. Aka, from the outside you can't distinguish it from corporate marketing.
Equivalent? In the analogy of the math or physics results it would be a mental model in someone's brain that you can't access or verify. You just hope it's true
You argue that open software is science, which is not at all the same as claiming that science is software. ("is" in this context is not equivalence -- "a poodle is a dog" != "a dog is a poodle".)
I agree the post is muddy about whether the relationship is bijective (equivalent, poodle=dog) or injective (onto, poodle is a dog).
I make it slightly more precise in the statement "I posit that open source software is a necessary condition if we are to science in a computerized world". That's where the "is" comes from in the title. Throughout history, this definitely has not been the case. I'm arguing that's changing.
It's not just "muddy", it's thoroughly inconsistent and impenetrable (which probably has a lot to do with why there is so little engagement here). You say "TL;DR I claim that modern science is synonymous with open source software" which is radically different from your "slightly more precise" statement.
I won't put any more time into this ... good luck in figuring out what it is you really want to claim and presenting a coherent and cogent argument for it.
But until much of academia is burned to the ground, or until science can be properly separated from modern academia, this will never be so. The current academic incentives are all wrong: low-quality research is rewarded and results in publications, whereas high-quality research (that takes time, and usually reveals that most exciting publications depend on p-hacking or other highly data-dependent analyses and selective presentations) is not published or actively blocked during peer review.
So instead you get BS arguments about how data can't be released for various privacy concerns (when in reality the vast majority of most datasets are trivial to scrub of identifying factors, and even in more complex datasets where you need to consider k-anonymity, it is still trivial to release data that allows replication of core analyses), and academic science is increasingly irrelevant unless it is tied to tech and industry, where producing junk actually has real negative economic and personal consequences.
I don't know what world this article / post lives in, but it isn't the messy world of actual reality.
At least some journals have this as a stipulation e.g. https://www.nature.com/nature-portfolio/editorial-policies/r...
Particularly the "data availability" and "Availability and peer review of computer code and algorithm".
However, in my limited experience, of trying to reproduce certain scRNA-seq processing pipelines, in practice it's never available as just a Github link. I can understand that some/many researcher's code is not in good shape, so I think it'll be quite a stretch to have this available.
I do think it's laudable though, to try and make it available. It would certainly have been very useful for me in the past.
Another problem with reproducibility is the openness of the underlying data. Many academics are terrified of giving away the golden goose and the software is often useless without the data.
However many scientists do work openly, e.g. The Journal of Open Source Software:
https://joss.theoj.org/
I am particularly fascinated by labs like DeepMind [https://deepmind.google/science/]. The recent advances in their frontier models that are able to predict diseases before they're diagnosed is incredible. This is what AI should be built for and actually do!
> TL;DR I claim that modern science is synonymous with open source software
That's a strong statement that's not supported by the arguments and IMO misguided.
I don't have a problem with "open", but rather with "software".
Both science and software deal with models, however the focus is quite different. I suspect you conflate theory with models.
The goal of science is to produce and test theories — that's an inductive/abductive process. A model, regardless of whether it's reified into mathematical formulas or software, is a means of making a theory operational enough that its consequences can be derived and confronted with observations.
Software often starts downstream of this: it's a reification of theories, models, algorithms, or findings that are the result of research. Of course software can also be used as part of the research process itself. The distinction is roughly the familiar one between research and development.
Software is a static program and basically none of these things.
Software development is kind of like science, in some ways, in that you discover abstractions and patterns, and this requires resolving disagreements and ambiguity between you and your users, but in the end, the user demands are usually fairly concrete and specific (though no one may know how to express those demands precisely, initially), and the process is not really exploratory in the way science is.
It just really isn't a very good comparison IMO.
If I'm reading yiur argument right, you're saying that deployed models are downstream versions (reified) of aa theory. Theory, being the actual object of science.
I think this misrepresents science. Science is the ability to build testable knowledge. From that,how would you separate the test from the science? In fact, in an ideal world, why wouldn't you want your theory to be put in a format that's executable? I'm not saying that those things are always the same, my (provocative) title is based on a dream where we can imagine theory and model coexist because software is now a thing.
> Please don't comment on whether someone read an article. "Did you even read the article? It mentions that" can be shortened to "The article mentions that".
Or like a machine learning lab claiming SOTA on a benchmark, beating a well-known method that they re-implemented, possibly with bugs, on their private dataset, for millions of compute. But you don't get the source to check, and they don't release any intermediate results or ablation experiments. Aka, from the outside you can't distinguish it from corporate marketing.
I won't put any more time into this ... good luck in figuring out what it is you really want to claim and presenting a coherent and cogent argument for it.