[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: computer-go: Proposal for an electronic journal of computer go
You make excellent points! One point that interests me is the one you
say here:
> The first kind of author has a go program that plays *well* but is not
> persuing an academic career. Any kind of paper from this author would be
> interesting, even if it perhaps only describes the program or some method it
> uses, without revealing the details.
It's been my observation that the best writers are not always the
people with the best ideas. When I read ANYTHING, I am looking mainly
for content, not style. I don't care if words are mispelled or
grammer is bad, as long as I can understand the idea. But having a
standard of excellence is still a good objective to have, I would
prefer good style, but I don't consider it a priority. I would like
to say that I think there should be liberal guidelines about which
articles are to be accepted for posting, even if they are not approved
for excellence or do not meet academic criteria such as the example
you mention of the good go progrmmers who just wants to present some
ideas.
What is the objective of such a project? Looking at the web page I
see the following statement at the top:
At the moment, there does not exist any dedicated academic refereed
forum for publishing about computer go. In my opinion this is
unfortunate. It is also quite strange given the applauded belief that
"go will replace chess as the task-par-excellence task in AI".
This is certainly a noble cause in my opinion. But then, in the second
paragraph I see the following:
Nowadays, papers about computer go appear in random places whose
principal topic is not computer go. The authors must therefore see
extra effort to introduce go to the audience, and need to emphasize
the non-go-related principal topic of the particular
conference/journal perhaps artificially. It feels like computer go is
"floating", in a sense, as there is not refereed forum that would work
as the "anchor", being the number one place for publishing about
computer go.
This is by no means contradictory, but it does raise the point that
computer go articles appear in random places. Wouldn't it be nice to
bring them ALL together, even though they may indeed vary
significantly in quality? If this isn't done by this project, then
the statement, "Nowadays, papers about computer go appear in random
places ..." will still be true. It doesn't seem good to me that we
COULD potentially throw out the best stuff simply because it didn't
have enough references in it, or because the author was not a very
good writer. And yet I would still encourage that much attention be
payed to an acedemic version of such a journal.
- Don
> I agree with Julian about peer review here, and I want to try to make this
> point clearer.
>
> Sofar most people who have responded have voiced the opinion that this is
> a great idea but that peer review is unnecessary, but I'd like to argue that
> peer review is a necessity if there is any point in having a journal at all.
>
> In psychology (my subject), it is almost impossible to write a paper that
> goes to print after one round directly because the subject is too complex.
> This is not bad because one can always make the points clearer, answer
> questions that one did not think of until a reviewer did, make references to
> other papers you never heard of until a reviewer told you about it etc.
>
> Peer review is not just a question of which papers are published or not, but
> the quality of *every* paper published. I think the rejection ratio of this
> journal will be rather low, whatever system is used so it will not scare away
> authors, but the main effect of the review is that the quality is increased.
>
> As hinted by Antti there are two kinds of potential authors.
>
> The first kind of author has a go program that plays *well* but is not
> persuing an academic career. Any kind of paper from this author would be
> interesting, even if it perhaps only describes the program or some method it
> uses, without revealing the details.
>
> The second kind of author has a program that barely beats Wally (poor
> Wally, why is everybody so obsessed with beating this program?) thanks
> to some new algorithm that solves some well defined but narrow problem in
> computer go, or perhaps applies some machine learning technique that do
> not make it all the way to a strong program but is interesting in a academic
> sense etc. A paper from this author should fulfill academic criteria: it has to
> be original, relate clearly to other published papers and the details should
> be clear enough to make any go programmer able to replicate the content
> of the article.
>
> If the second kind of author is persuing an academic career, a journal that
> is not peer reviewed is completely uninteresting because it does not
> *count*. The academic world is a culture of "publish or perish". If I in the
> future apply for a job, a paper published on an arbitrary web page do not
> count regardless of the content. If it is in a peer reviewed journal it will at
> least count as one minimum unit of academic credit. Such a person would
> rather publish somewhere else and give the reference to the Go
> Bibliography page of Marcus Enzenberger, which by the way is a goldmine
> already if you have not yet visited it:
>
> http://home.t-online.de/home/markus.enzenberger/compgo_biblio.html
>
> It may be necessary to have two different kinds of papers accepted in the
> journal. Normal ones for the academic people, and special ones for the
> commercial programmers. Both kinds should be peer reviewed and of about
> the same quality but the kind of content could differ a lot.
>
> Summary:
>
> ** Keep peer review, for quality and scientific credibility. Yes, I am selfish :-
> ) in this respect if you wonder.
>
> By the way, biannual publication is enough, judging from the high ;-) traffic
> on this mailing list recently, or are you busy writing papers?
>
> Best wishes
>
> Magnus Persson
>