(Digital) Jimmy

Yahoo News reported a few days back that the New York times has discussed creating a Wiki-Leaks style for potential whistle-blowers to submit documents to the site. I think that this gives us a good opportunity to rethink WikiLeaks-style reporting, to keep what we like and to get rid of what we don’t. The controversy over Wikileaks is not just about specific documents on the site, but over the larger question of how much sensitive information should be accessible to the public.

The debate is more complex than simply the tired-old “There should be no secrets” versus “We should trust our government to withhold information for our own protection” battle. For instance, leaked information pertaining to toxic dumping by an agent of Trafigura off of the Ivory Coast was something that we all should be glad has come to light. But in other cases, weighing value of the information versus its potential damage is more difficult. When the names of several Afghan informants were plainly available in documents leaked about the war in Afghanistan, there was cause for significant concern. Although WikiLeaks the Pentagon and activist groups alike to help redact potentially compromising information from the documents, no one was willing to cooperate. By Assange’s own admission, this kind of unfettered exposure could lead to blood on the hands of the group. There must be a way where we can have our cake and eat it too.

It may be no great surprise that I would suggest we “crowdsource” the task of filtering this information. However, this approach immediately presents problems. For one, if individuals are reading the documents to filter them, that means that compromised information may have been leaked in the very act of that filtering. Who is going to do this filtering, and for what reasons? To solve some of these problems, we can borrow some techniques from privacy and security. Let us start with some large document to be redacted. First, split it up into numerous paragraph-sized pieces. When a new user logs in (who is appropriately throttled in terms of number of requests, to prevent spidering) he is given a piece of the document. He reads it and marks both what should be redacted and what should be highlighted as worth further review. He submits this piece and receives the next one. This same piece of the document is given to several users to gain confidence in the accuracy of their assessment. Interspersed with the bits of the actual document are paragraphs from a corpus that has either been randomly generated or taken from existing declassified documentation. The more of these “straw men” there are, the more secure the system is. If the generated bits of document are convincing enough, then the a user won’t know if he is learning genuine, never before seen information, or bogus distraction. If, therefore, he learns something potentially compromising, it won’t be confirmed unless that portion of the document is released.

Of course, even the best crowdsourcing infrastructure is useless without a committed core of users. Why would people be interested in such a system, you may ask. For the same reasons that they are willing to change a wiki page back and fourth to promote their view. WikiLeaks has sparked passionate debate, and if that energy is harnessed in the online world, it could be just the balance of tensions necessary to ensure that important documents are leaked while preserving information that is truly sensitive.