Spam, Turing tests, and AI
I received a spam message that slipped right through my Gmail spam filter today:
"Her beaver needs more times a night"
and I was impressed. Most of the prospective audience for the ad will understand the meaning of the title, but it seems very hard for a spam filter to detect. No naughty words or 133T-speak, not much in the way of keywords. Just careful use of innuendo and English slang.
How could you design a spam filter to catch this kind of thing?
It seems like there's an inherent competition between spam filters and automated Turing tests. If you can design a message so that a human can get the meaning but software cannot, then the message can be used as unfilterable spam. If you cannot design such a message, then any automated Turing test designed to pass only humans can be defeated by sufficiently smart software.
On the one hand, this may be frustrating to those who design spam filters and CAPTCHAs. http://en.wikipedia.org/wiki/Captcha
But on the other hand, it may be the best way to improve the state of human-mimicking AI. There's a lot of money in building effective spam filters. Maybe someone will come up with a breakthrough.
"Her beaver needs more times a night"
and I was impressed. Most of the prospective audience for the ad will understand the meaning of the title, but it seems very hard for a spam filter to detect. No naughty words or 133T-speak, not much in the way of keywords. Just careful use of innuendo and English slang.
How could you design a spam filter to catch this kind of thing?
It seems like there's an inherent competition between spam filters and automated Turing tests. If you can design a message so that a human can get the meaning but software cannot, then the message can be used as unfilterable spam. If you cannot design such a message, then any automated Turing test designed to pass only humans can be defeated by sufficiently smart software.
On the one hand, this may be frustrating to those who design spam filters and CAPTCHAs. http://en.wikipedia.org/wiki/Captcha
But on the other hand, it may be the best way to improve the state of human-mimicking AI. There's a lot of money in building effective spam filters. Maybe someone will come up with a breakthrough.