Hacker News Reader

Comment by mooneater | Hacker News Reader

Some heavy hitters on the signatory list: Jeff Dean, Yoshua Bengio, Volodymyr Mnih, Ilya Sutskever, Geoffrey Hinton, Chelsea Finn, Sergey Levine.

Would like to see Demis Hassabis and David Silver join the effort.

If Deepmind, OpenAI and Berkeley (well represented here) all boycott, that's a huge chunk of key AI researchers.

View on HN

Deep learning, Yann LeCun, Yoshua Bengio & Geoffrey Hinton, Nature volume 521, pages 436–444 (28 May 2015)

Strategy: publish in the high-impact established Nature-family journal, but publicly boycott the no-impact new Nature-family journal. Win-win.

edit: added -family suffix to clarify that I completely understand these are two different journals from the same publisher.

edit 2: I now understand why the editor of the new journal was hitting senior people up for personal meetings at ICRA last week. He didn't want to hear about our work: he wanted hints to help him on positioning.

They're boycotting the new Nature Machine Intelligence journal, not the old Nature journal. It's possible to be against a policy while still taking advantage of the policy for game theoretic reasons.

That's what 'win-win' means.

there's a huge difference between publishing one really general public descriptive article for the whole field and stifling research by having actual research papers published in a closed journal.

one doesn't hinder research, the other does

Their policy: "Nature Research journals support posting of primary research manuscripts on community preprint servers such as arXiv and bioRxiv. Preprint posting is not considered prior publication and will not jeopardize consideration at Nature Research journals."

https://www.nature.com/authors/policies/license.html

Standard practice in the modern AI community is preprint to arXiv first, then out to a journal for review. All the papers are free online before they even go to Nature. So this policy does not stifle or hinder research.

They also do not take copyright - it stays with the authors - unlike other giants such as IEEE.

To my mind the legitimate argument is about publishers taking undeserved profits from unpaid labour while adding little value.

(edit: reordered sentences)

One could argue good editing and curation is of high value. Not sure I buy it, but Arxiv certainly lacks curation or editing...

Conferences can fill that role instead. Their business model doesn't require paywalls or the copyright police.

All the conferences in my field have literal security guards checking the badges that are only handed out after paying hundreds of dollars to register.

Of course, but you don't need to attend the conference to read and benefit from the research, and that's the important thing.

The journal business model usually requires restricting access to the research, while the conference business model still works when the research is free.

There must be some benefit to attending conferences, otherwise why would scientists attend them?

That's the whole point. Conferences can provide a benefit worth paying for without putting the research itself behind paywalls like journals do.

Yeah, but it's usually more about a chance to socialize with other people in the field.

It's also interesting to hear other people's talks that serve as a "teaser" for what's in the paper.

Except that conference proceedings can be exactly like a journal, i.e. you often need to pay to access them.

Sure, if the publisher actually provides that. They don’t. (Or rather, they farm it out to volunteer labor while collecting a fee anyway.)

That just pushes out the argument by one level of indirection. The publishers 'curate and edit' the board of editors for their journals. In turn, these boards of editors curate and edit the articles that go into the journal.

Nope, they don’t. The “volunteer” reviewers are drawn from past published authors, who in turn had their articles selected by previously published reviewers. Turtles all the way down.

This is not true. Reviewers may or may not have been past authors. I have reviewed for many journals in which I have never published.

Nature journals actually do provide professional copyediting services for all articles accepted for publication. For commentaries (from my own experience), the editing process at Nature is quite intensive.

> To my mind the legitimate argument is about publishers taking undeserved profits from unpaid labour while adding little value.

That argument also comes down to hindering research. Funds that could otherwise be used for research are now used to buy subscriptions to this journal.

Not that I'm for journals, but those heavy hitters typically have their own brand in that of Google or Microsoft etc. Without an external 'brand' of Nature the importance of working for the recognised companies when submitting your research papers goes up.

Yoshua Bengio doesn't work for a large brand

For the unfamiliar, Bengio has his own company:

https://www.elementai.com/en/who-we-are

He still spends most of his time at MILA, of which he is the director, which is UdeM and McGill's shared AI research group. source: I study there.

http://mila.quebec/

That's what makes their support so important. If they're not publishing there, then the external brand value of this specific journal goes down, which means that even for less heavy hitters, the best "brand" value is still in the journals that actually make their work as widely available as possible.

Also Ian Goodfellow (GANs) and John Langford (Vowpal Wabbit).

People are forgetting what year we're living in, 1818.

It is not possible to publish something without setting moveable type made of lead. it would be different if there were an Internet or something but that's way in the future.

it seems fewer and fewer people know how to take a joke

Every forum has its own culture. The culture of this forum is against snarky humor that does not contribute substance.

The substance is that in 1818 you can't publish a serious journal to be read by scientists all over the world without expensive typesetting, printing, and so forth. (Setting moveable type, etc.) Typesetting a journal is hard work with high fixed and variable costs.

To give you a sense of the state of our industry, here is an example of an innovation just four years ago and its effect:

>1814 – First cylinder presses

>Friedrich Gottlob Koenig and Andreas Friedrich Bauer build their first cylinder press, which is much faster than the existing flatbed presses. One of the first customers is John Walter of The Times. The first issue of The Times that is printed with the new presses is published in 1814. The press is installed in secret to avoid sabotage by disgruntled pressmen operating the existing Stanhope presses. The machine is capable of printing over 1100 double-sided sheets per hour. In 1817 Koenig & Bauer return to Germany and start building presses in an abandoned monastery in Würzburg.

https://www.prepressure.com/printing/history/1800-1849

You can click that link for some pictures. This press needs to be paid for, as well as having ongoing operational costs that include a lot of meticulous manual labor (mentioned above) to set each and every page.

Here is another link which I rewrite to the present:

>There are many reasons to celebrate the advent of the steam press in 1814, as well as reasons to worry about it. Steam printing brings the cost of printing down, increases the number of possible impressions per day by four times, and, in a way, we might say that it helps “democratize” access to information. The Times proclaims that the introduction of steam is the “greatest improvement” to printing since its very invention. Further down that page, which itself is “taken off last night by a mechanical apparatus,” we read why the hand press printers might be concerned: “after the letters are placed by the compositors… little more remains for man to do, than to attend upon, and watch this unconscious agent in its operations.”

https://blog.oup.com/2014/12/steam-press-newspaper-london-ti...

As you can see it talks about "the democratization" of information. Still, my point is that the fixed costs of this publishing model are still high, and that it makes a lot of sense to operate on this publishing model for this reason alone. After reviewing my links and the technology in question, I hope you can readily agree! In 1818 typesetting and printing is an expensive, hardware-based, industrial business. It makes sense to have a publishing house behind it.

Is there any part of what I've just written that you disagree with?

See, that contributed snark WITH substance. Gold star.

I’m not sure I totally agree with the parallels, which is probably where some of the initial confusion arose from. The printing press significantly reduced the costs of publication, allowing unprecedented access to the written word beyond the aristocracy. I don’t think anyone disagrees that Nature provides a valuable service and that it “makes sense to have a publishing house behind” peer reviewed literature. Rather, when the cost to produce peer-reviewed content can be taken on by tax-subsidized content production professionals, and demonstrably WITHOUT an older journal that extracts exorbitant fees without producing significant value beyond its brand-name, it makes sense to challenge the value of the brand.

> I don't think anyone disagrees that Nature provides a valuable service

Ooh, ooh! raises hand and waves franticly

Nature does jack shit that couldn't be (and in many cases isn't) done as well or better by the authors' personal or institutional websites.

What they "provide" is a brand name for wasteful status signalling, the same as Goochi purses or sportsball teams.

If you have to pay royalties to use a digital font, it stands to reason you have to pay royalties to the keepers of information. Why that's elsevier et al all I don't know, but they are is the only logical conclusion.

Someone spent time and effort designing that font, and it's reasonable that they should be compensated for it. (Giving them the right to censor any use of the font so they can run a protection racket is a awful way of doing that, but that's not the point I was addressing.) Nature contributes nothing (today - they used to contibute printing presses) that couldn't be accomplished by publishing a first draft and soliciting reviews.

look at what they had to write for that gold star tho, to make the same point! good grief! ppl on HN be self employed, working few hours and all, but still, thats time he aint gettin back ;P

This forum in general values substantive comments like the second comment the OP made and doesn't value snarky one liners like the first one. That's the trade off for the value folks get from this forum. Other forums may have different cultures that may be more suitable for this particular viewpoint.

but there is substance to the joke above. so either its over people's head (as it seems to be on you) which is a shame as HN tends to have fairly intelligent culture, or else its not lost on the audience but they don't appreciate anything inexplicit or non literal, ie, humour, which is also a shame

Just because somebody dosent like a joke doesn't mean they don't get it.

Just because somebody doesn't like a joke doesn't mean there is no substance in it.

Deepmind puts everything on ArXiv already.

they are not mutually exclusive. nature allows for preprints on arxiv according to wikipedia.

also deepmind published in nature justthis month.

but they also seek publications in the main Nature, which is unusual in the field.

Everything? They publish basically extended abstracts, nothing that could be confidently reproduced, nor code ... or I just don't know how to read it and find supplemental material. There's a lot in the bibliography. Really just a journal article, an entry in a log book.

Also many folks from Facebook AI Research (including Yann LeCun)

Not to say that they are wrong in boycotting, but I find it hard to take seriously a moral stance from somebody working for Facebook.

Kind of a hard stance to enforce. If you work for basically any major corporate entity you are probably sacrificing integrity to do so. So it seems you are calling the kettle black here.

Not quite every corp out there has a business model of specifically misusing and abusing personal information, manipulating peoples' emotions, etc., etc.

I work for a rather minor corporate entity anyway, does it mean that I am always off the hook? /s

Equating all major corporate entities is a cop-out.

I clearly left room for some corps to escape my characterization so there's no cop-out, sorry.

https://en.wikipedia.org/wiki/Tu_quoque

Just because their parent company might not be morally in the clear, that doesn't mean that moral criticisms by its employees is automatically or even more likely to be invalid.

Oh, it might be valid, in regards to the magazine, alright. It just smacks of hypocrisy when their research is pretty much bound to be used for unethical purposes.

Literally Who: The List

got some references as to why they are big hitters for the uninformed such as myself?

Jeff Dean is head of AI at Google. He's one of the most prolific computer scientists there, and has worked on many big projects such as MapReduce, BigTable, Spanner, GoogleBrain and TensorFlow.

Yann LeCun is head of AI at Facebook and kinda knows as the father of convolutional networks.

There's a bunch of other ones from other big companies or big ivy league universities.

Reply on news.ycombinator.com