Today's NYTimes article on program's designed to thwart machine's and software's ability to "pretend" to be a human reminded me of the entire A.I. field of study from college and years ago when we were trying to sell companies and the government on commercial AI applications.
The article itself is about CAPTCHA and re-Captcha as well as image recognition and how these systems are used by Yahoo, Google and others to keep the "bad guys" from signing up for bogus email accounts and whatever and then spamming us to death. An admirable goal. Even the system's themselves are going good, for example re-CAPTCHA where images that cannot be scanned and OCR'd properly are being "read" by the human populace one or two words at a time as they sign-on or sign-up and the words (actually multiple iterations of the same word from different people in order to ensure fewer errors creep in) are then substituted in the scanned books of libraries and other material and thus improving our ability to read books "at a distance" where maybe only a copy or two exists.
The point of all of these tests is to determine if it's a human behind the "keyboard" rather than a machine cranking out sign-up requests by the thousand's. I have heard that the spammer's themselves were using similar human powered examination of the images to thwart these attempts, in that they rewarded viewers at their porn sites to simply solve the same problem by revealing more and more of a naked lady as the letters of the problem were typed in.
Beyond this is all of the work that has led up to this point... The CAPTCHA acronym itself stands for “completely automated public Turing test to tell computers and humans apart” and is a direct reference to the British mathematician, Alan Turing. (for a fictional counterpoint to all of this, read an excellent novel by Neal Stephenson Crypotnomicon - a 1999 novel which had a side plot about the the Bletchley Park mathematicians in their attempts to crack Axis codes during WW-II.) who has contributed much to be beginning of this field.
The Hollywood machine, with another Terminator movie out, would lead us to believe that much of this is ready, available, and here already. Let's send Jay Leno out to do another "man on the street" interview to see what Americans really know. What he would find would be humorous, because this work is all hard. Even 10 years ago with the beginning of commercialization of this kind AI technology, it's still really hard with only now some specialized things coming out of MIT and other schools/research groups to support some very specific kinds of applications, like an autopilot for a truck or a robot "pack mule". The most recent, big deal AI app I'm aware of was for plane refueling during Desert Storm, where AI's constructed AI's to schedule the mid-air tanker's for all of those fighter jets. None were lost, and at the scale of the operation I'm told we would have been limited or had some flame-outs if the human schedules had continued.
Still, eventually this will all really be commonplace, and probably without the Terminator.
Sunday, May 24, 2009
Subscribe to:
Posts (Atom)