5 ms·

Even assuming this is a good idea in theory, it's probably not possible. If you're going to publish a full paper defending an idea, it's going to have enough of

Even assuming this is a good idea in theory, it's probably not possible. If you're going to publish a full paper defending an idea, it's going to have enough of your writing style in it to identify you. Academics early in their careers have to publish a lot in order to advance. With more of that publishing moving to freely available online sources, it will be pretty easy to automate attribution. If this journal is freely available online, there will probably even be a browser extension to add attribution. The only way to publish anonymously is to not also publish under your real name. Even then we'll be able to at least apply a consistent pseudonym to your writing and you'd always be one named publication away from full identification.

Eridrus 8y ago
I disagree. Text analysis only really works when someone attempting to publish anonymously is unaware that someone will do that analysis. At the point where there are simple tools, you can use them to put together a report on what is most identifying, and then fix it, before publication. Or one can even publish false flags. If anyone remembers the "lodestar" speculation from earlier this year, that's a single word that someone could have put in to get the writing completely misattributed in the public sphere.
- daveFNbuck 8y ago
  What you're describing will only work against current publicly available tools. You can't fool the tools that you don't know about as easily. Would you bet your career against the next few years or even decades of AI research just to publish an anonymous paper?
  - Eridrus 8y ago
    The more "sophisticated" a textual analysis tool is, the less interpretable it is, and the more difficult it will be to get people to believe the output. And yes, I think the people involved in this will take that risk.
liftbigweights 8y ago
> it will be pretty easy to automate attribution. Yours is a valid concern, but people can also fool attribution software easily. You could even write software that alters your paper into a few different people's writing style. That will render any attribution process useless. Just like you can disassemble code, you can obfuscate it.
- daveFNbuck 8y ago
  You can fool current attribution software easily. It gets better over time and what you're describing is essentially leads to an arms race of obfuscation software versus attribution software. I wouldn't bet that current obfuscation software would be able to permanently hide all traces of my own composition while keeping the paper coherent.

Eridrus 8y ago

I disagree. Text analysis only really works when someone attempting to publish anonymously is unaware that someone will do that analysis. At the point where there are simple tools, you can use them to put together a report on what is most identifying, and then fix it, before publication. Or one can even publish false flags. If anyone remembers the "lodestar" speculation from earlier this year, that's a single word that someone could have put in to get the writing completely misattributed in the public sphere.

daveFNbuck 8y ago
What you're describing will only work against current publicly available tools. You can't fool the tools that you don't know about as easily. Would you bet your career against the next few years or even decades of AI research just to publish an anonymous paper?
- Eridrus 8y ago
  The more "sophisticated" a textual analysis tool is, the less interpretable it is, and the more difficult it will be to get people to believe the output. And yes, I think the people involved in this will take that risk.

liftbigweights 8y ago

> it will be pretty easy to automate attribution. Yours is a valid concern, but people can also fool attribution software easily. You could even write software that alters your paper into a few different people's writing style. That will render any attribution process useless. Just like you can disassemble code, you can obfuscate it.

daveFNbuck 8y ago
You can fool current attribution software easily. It gets better over time and what you're describing is essentially leads to an arms race of obfuscation software versus attribution software. I wouldn't bet that current obfuscation software would be able to permanently hide all traces of my own composition while keeping the paper coherent.