Here are my thoughts about replacing peer review by validation. Peer review is the practice where the work of a researcher is commented by peers. The content of the commentaries (reviews) is clearly not important. The social practice is to not make them public, nor to keep a public record about those. The only purpose of peer review is to signal that at least one, two, three or four members of the professional community (peers) declare that they believe that the said work is valid. Validation by reproducibility is much more than this peer review practice. Validation means the following:
- a researcher makes public (i.e. “publishes”) a body of work, call it W. The work contains text, links, video, databases, experiments, anything. By making it public, the work is claimed to be valid, provided that the external resources used (as other works, for example) are valid. In itself, validation has no meaning.
- a second part (anybody) can also publish a validation assessment of the work W. The validation assessment is a body of work as well, and thus is potentially submitted to the same validation practices described here. In particular, by publishing the validation assessment, call it W1, it is also claimed to be valid, provided the external resources (other works used, excepting W) are valid.
- the validation assessment W1 makes claims of the following kind: provided that external works A,B,C are valid, then this piece D of the work W is valid because it has been reproduced in the work W1. Alternatively, under the same hypothesis about the external work, in the work W1 is claimed that the other piece E of the work D cannot be reproduced in the same.
- the means for reproducibility have to be provided by each work. They can be proofs, programs, experimental data.
As you can see the validation can be only relative, not absolute. I am sure that scientific results are never amenable to an acyclic graph of validations by reproducibility. Compared to peer review, which is only a social claim that somebody from the guild checked it, validation through reproducibility is much more, even if it does not provide means to absolute truths. What is preferable: to have a social claim that something is true, or to have a body of works where “relative truth” dependencies are exposed? This is moreover technically possible, in principle. However, this is not easy to do, at least because:
- traditional means of publication and its practices are based on social validation (peer review)
- there is this illusion that there is somehow an absolute semantical categorification of knowledge, pushed forward by those who are technically able to implement a validation reproducibility scheme at a large scale.
UPDATE: The mentioned illusion is related to outdated parts of the cartesian method. It is therefore a manifestation of the “cartesian disease”.
I use further the post More on the cartesian method and it’s associated disease. In that post the cartesian method is parsed like this:
- (1a) “never to accept anything for true which I did not clearly know to be such”
- (1b) “to comprise nothing more in my judgement than what was presented to my mind”
- (1c) “so clearly and distinctly as to exclude all ground of doubt”
- (2a) “to divide each of the difficulties under examination into as many parts as possible”
- (2b) “and as might be necessary for its adequate solution”
- (3a) “to conduct my thoughts in such order that”
- (3b) “by commencing with objects the simplest and easiest to know, I might ascend […] to the knowledge of the more complex”
- (3c) “little and little, and, as it were, step by step”
- (3d) “assigning in thought a certain order even to those objects which in their own nature do not stand in a relation of antecedence and sequence”
Let’s take several researchers who produce works, some works related to others, as explained in the validation procedure.
Differently from the time of Descartes, there are plenty of researchers who think in the same time, and moreover the body of works they produce is huge.
Every piece of the cartesian method has to be considered relative to each researcher and this is what causes many problems.
Parts (1a),(1b), (1c) can be seen as part of the validation technique, but with the condition to see “true”and “exclude all grounds of doubt” as relative to the reproducibility of work W1 by a reader who tries to validate it up to external resources.
Parts (2a), (2b) are clearly researcher dependent; in a interconnected world these parts may introduce far more complexity than the original research work W1.
Combined with (1c), this leads to the illusion that the algorithm which embodies the cartesian method, when run in a decentralized and asynchronous world of users, HALTS.
There is no ground for that.
But the most damaging is (3d). First, every researcher embeds a piece of work into a narrative in order to explain the work. There is nothing “objective” about that. In a connected world, with the help of Google and alike, who impose or seek for global coherence, the parts (3d) and (2a), (2b) transform the cartesian method into a global echo chamber. The management of work bloats and spill over the work itself and in the same time the cartesian method always HALT, but for no scientific reason at all.