Casey Chesnut: Genius (AND | OR) Evil ?

Casey Chesnut used a neural network to defeat comment-spam CAPTCHAs and used his software to auto-post a polite “Sorry, you've been 0wned, here's how I did it” comment on 90+ blogs. Apparently, some people have their comment system linked to hairs up their asses and Furor Has Erupted (tm). “He proved nothing! Everyone knew that this could be done!” Say some. Ummm... not so much. Recognition of pixel-based distorted shapes against a noisy background is not trivial (he says, being a guy who has reason to believe the U.S. is still using some neural net code of his in the latest generation of mumble-mumble-mumbles).

Neural nets, like most AI techniques, are easy to overestimate. A solution based on neural nets always involves creative preprocessing and an error-prone training and refinement process. Casey proved that it was relatively easy to extract trainable features from a particular, but fairly representative, CAPTCHA implementation. (Note to Casey critics: background grids would be trivial to preprocess out.) That seems absolutely praise-worthy to me.

Did he need to post to 90 blogs to prove his point? Nope. Is what he did unethical? Nope. At most, it was rude and I'm not even sure about that (I don't know who he posted to, but it certainly wouldn't be rude to anyone who had blogged about CAPTCHAs or AI or image recognition).

Keep going, Casey! Hey, why don't you take on continual speech recognition in noisy environments next? Everyone knows it can be done!