According to a post on Wired, three MIT students managed to get a computer-generated, gibberish paper accepted for the 2005 World Multi-Conference of Systemics, Cybernetics and Informatics.
This is nothing new in terms of AI. Eliza did this three decades ago and a program called Racter wrote a book of poetry called The Policeman’s Beard is Half-constructed that actually got published in the eighties (I’ve got a copy on my bookshelves somewhere – how geek is that?).
It’s hardly passing the Turing Test but, according to New Scientist, it managed to fool a human reviewer. Or at least they didn’t comment on the proposal.
Peer review and, on a general level, public scrutiny isn’t the perfect tool for sorting out what is rubbish and what is good but it is better than all the alternatives. But it does the review to be done diligently.
Applying this to writing, I have a lot of respect for American magazines (I’ve worked for Wired and Popular Science among others) who have legions of fact-checkers and super-diligent editors. Sometimes it’s a monumental pain having to write an article of 500 words and fact-clarifying emails and notes of 1,500 words but it does mean that the end result is generally accurate and objective.
In my own small way, at Articulate, we send all our work to a third-party sub-editor to check for typos and to improve the copy. It’s amazing what a fresh, well-trained pair of eyes will do.
Related posts:
