A similar idea was implemented in Lotus Notes (yes, I know. The horror) about 15 years ago. Instead of an icon they used four possibly-different symbols arranged in a 2x2 matrix. It hashed your password and was used as a visual hint to whether you typed your password correctly. It was pretty but not very useful.
links to code and blog post are broken
I couldn’t get this working on Win7/VS2008. It seems CloseFigure() works a bit differently in that it if you have more than one PathPointType of CloseSubPath it throws a “Parameter is not valid” exception. I managed to fix this by removing the gp.ClosePath() from AddPointToGraphicsPath() and putting in patch.ClosePath() in PatchSize set{} after the for loop has completed.
Great job, Jeff. Will you be hooking it up to your blog’s comment support code? Oops. Looks like you need an MT plugin. 
Re Swastika comments, that can be addressed by applying a specialized OCR-like visual analysis to identify all offending codes then crunch them into an effective bloom filter using genetic algorithm. When the filter returns true, a second type of identicon (i.e. 4-block quilt) can be used.
Re usefulness, IP identicon provides imperfect yet reasonably effective means to prevent forgery within a website. But then it’s impossible to please everyone.
I don’t think they’re meant to replace the “avatar” concept, they’re more to help distinguish anonymous posts on something like a blog. For example, I’ve noticed that there’s another “Aaron” who sometimes comments here and even has a similar writing style, but it ain’t me, and the glyph would highlight that (IP addresses are only visible to Jeff himself).
Seems like it could also be used to help identify trolls and sock-puppets, even on sites that require registration. True, some people’s IP addresses do change frequently so it’s not foolproof, but as an heuristic it’s a lot more convenient than visually parsing a list of hundreds of IP addresses. Barcodes are ugly and are probably even more difficult for a human to parse than 12 digits.
I like it. Can’t think of a use for it in my line of work but still very cool!
Matt, you posted the same message on my blog without backing up your claim with an explanation.
An explanation or free pizza and beer will do for me.
Ah, it was just over a misunderstanding. I would have been happy if Matt discovered some major breakage in SHA1 just for excitement sake.
I think that link’s dead. Also how does this 1.3 version compare to the 1.1 Phil Haack’s put on codeplex? http://identicon.codeplex.com