-----BEGIN PGP SIGNED MESSAGE----- Date: Mon Jul 17, 1995 5:22 pm GMT From: Timothy C. May Subject: Automated Rant Generators and Letter Generators At 7:35 AM 7/17/95, Martin Hamilton wrote:

MONTY HARDER writes:

| Anyone who has read MAD Magazine could put such together. As an added | bonus, use variable margin settings, and none of the letters would be | exactly the same. Appropriate Imail => FAX software on a puter in DC | (local call that way) with the phone number of the sender filled in on | the top line for ID (izzat legal?) so it doesn't look like a form letter | at all.

Plus - choose the fonts & point sizes at random too ? :-)

Tim May > Cypherpunks could probably have an effect on Tim May > hastening this "denial of service" attack on the Tim May > efficacy of letter-writing by releasing an easy-to- Tim May > use package that does all this letter writing at Tim May > the click of a button....just type in some key Tim May > words, for the topics, and it does the rest. Tim May > An interesting project, actually. Actually, your little project could cause some major problems in a area you may not anticipate --- personnel selection. Specifically, the use of Handwriting Analysis as a tool for personnel profiling. Date: Mon Jul 17, 1995 11:48 pm GMT From: Harry Bartholomew Subject: Re: Automated Rant Generators and Letter Generators Harry > A final step might be to interface the output to old Harry > pen plotters like my HP7470A with an ascii-to- Harry > handwriting program. Akin to the White House Harry > souvenir signature generator, but with a set of Harry > parameters to mimic different "hands". Knuth's Harry > Metafont tricks come to mind. Making the little problem Tim presents, a major headache for somebody else --- handwriting analysts. Date: Tue Jul 18, 1995 1:18 am GMT From: Timothy C. May Subject: Re: Automated Rant Generators and Letter Generators Tim May > Bart's comments about using Knuth's typographic Tim May > work are interesting, to the extent that letters Tim May > need to look handwritten. In the Mac market, it's Tim May > possible to send in some handwriting samples and Tim May > get back a font that emulates the handwriting! Actually, True Type fonts of your handwriting are available, for any platform that accepts that font type. I don't have the URL for them, but there is a pointer to it at HTTP://www.ntu.ac.sg/~tjlow/gclub.html Tim May > I don't think the pen plotter is actually needed- Tim May > - and few people would use it--as most fax can Using it would play hell for handwriting analysts, though. And if it was programmed to change the pen pressure as well --- the possibilities are staggering. Can a pen plotter change pressure? Tim May > be emulated with laser printers (due of course to Tim May > the limited dots per inch resolution). In fact, Tim May > most fax modems can directly fax from any screen Tim May > that can produce printed output. So, the Tim May > combination of handwriting fonts, automated rant Tim May > generators (of varying rabidities), and fax Tim May > capabilities gives a pretty good start. Using lots Tim May > of handwriting samples, various other fonts, and a Tim May > mix of styles in the letters will help. Tim May > Anyway, where this all gets interesting is the Tim May > following: * Can a kind of Turing Test be tried Tim May > here? But of course. Tim May > That is, in this limited domain of "letters to the Tim May > editor/Congressmen," can a letter generator be Tim May > implemented which generates letters effectively Tim May > indistinguishable from letters and fax generated Tim May > by actual human beings? ("Effectively Tim May > indistinguishable" in the sense that a human reader Tim May > could not sort a set of letters into human- and Tim May > machine-generated subsets with statistically Tim May > significant certainty better than guessing). I don't remember the title, but at least one french novel was rumored to have been entirely generated by computer. Tim May > Of course this is also similar to the "style Tim May > detectors" we so often talk about. I don't remember the program name, but there is software available now, that analyzes a document, and figures out who wrote it --- based on the frequency count of the letters of the alphabet. Secondary measures are frequency counts of letter pairs. Words, phrases, sentences etc are totally ignored. So what you'd need to do here, to pass your pseudo- Turing Test is a program that generates different statistical results, for allegedly different people. Tim May > The crypto relevance has to do with detecting Tim May > patterns in letters and rants, in emulating these Tim May > patterns, and (perhaps) in speeding up lobbying. Tim May > (Though I agree that widespread adoption of Tim May > automated letter-writing, such as the direct mail Tim May > folks are already doing, will eventually just kill Tim May > off letter writing as a means of lobbying.) Tim May > This may also hasten the adoption, someday, of Tim May > digital signatures. Congressmen and their aides Tim May > may check incoming letters against databases of Tim May > their constituents who have "registered" with them Tim May > (lots of issues here). Or might just subject all mail to various automations, which accept/reject mail, based on what it looks for. << If it passes the congress person's Turing Test, it is read, as being authentic --- although I doubt half the people in the capital could actually pass a Turing Test to begin with. >> Tim May > Merely counting the "yes" and "no" letters has long Tim May > been problematic, as the Republicans have been Tim May > leading in direct mail campaigns since at least the Tim May > mid-70s (recall Richard Viguerie...). Increased Tim May > automation will just make it even more obvious. Date: Tue Jul 18, 1995 5:23 pm GMT From: Timothy C. May Subject: Re: Automated Rant Generators and Letter Generators Tim May > David Conrad told me he meant for this to go to the Tim May > whole list, but only sent it to me by mistake. So Tim May > here is his post. At 4:14 PM 7/18/95, David R. Conrad wrote:

Tim May <tcmay@sensemedia.net> writes:

...

Bart's comments about using Knuth's typographic work are interesting, to the extent that letters need to look handwritten. In the Mac market, it's possible to send in some handwriting samples and get back a font that emulates the handwriting!

I suppose the resulting font has only one form for each letter? (Although I understand that when you send them a sample, you send several instances of each letter; a friend was showing me an add for this.) The fact that each letter is the same every time would be a giveaway. We need something like Metafont, or at least choose from a number of different shapes.

...

... So, the combination of handwriting fonts, automated rant generators (of varying rabidities), and fax capabilities gives a pretty good start. Using lots of handwriting samples, various other fonts, and a mix of styles in the letters will help.

Another factor that would make it appear more authentic would be spelling and grammar errors. The grammar errors could be built into the rant generators (an occasional dangling modifier, an incomplete sentence or two); spelling errors could be done by post-processing the output of the rantgens. It's important to take into account the different types of spellos that occur: commonly misspelled words (aquired, beleive); wrong homophone (their, they're, there; two, to, too); transposed letters (transpoesd); near-misses on qwerty keyboards (nesr-mosses); and words left out.

-- David R. Conrad, ab411@detroit.freenet.org, http://web.grfn.org/~conrad/ Finger conrad@grfn.org for PGP 2.6 public key; it's also on my home page Key fingerprint = 33 12 BC 77 48 81 99 A5 D8 9C 43 16 3C 37 0B 50 No, his mind is not for rent to any god or government.

Actually, the usual give away, is in letter and letter pair frequencies --- not spelling mistakes, grammatical errors, etc. However, there a technique called _Scientific Content ANalysis_ that looks at how things are said, to judge their "truthfulness." A good program will not show that the text was randomly generated, nor show that the author is off-the- wall, so to speak. You may have bitten off a bit more than you can chew here. OTOH, a group that tries to crack keys, knowing that the possibility of success is slim to non-existent, can probably pull this one off --- if only because the possibility of success is pretty good. Er, how did the cracking of the key go? Last I read 60+%, and no hints of it being broken. << I almost want to participate, but with a dx25, running NovellDos, I'm not sure what that platform could do. << I''ll graduate to Linux, after I buy some more memory, and a new hard drive for that sytem. >> >> xan jonathon -----BEGIN PGP SIGNATURE----- Version: 2.6.2 iQCzAwUBMAyFtaVRQvz57IB1AQF25wTvQD+eQVxvKOwin+Izb4d5c0u7i6JWWSZR BoY9T3b7BEhiU6EfKgP4BZabi8gHTM742ROCXAvCZQusWAxLfXSOKwjmUs5ieaD7 f6cEB8/D+EZu395qa0bCu28/hLmKslQvXvsWoMpxcHzhjEHJhYs/0BQxHZoZMsrM PrfFLqrhdJzhPYn5iy83nhBB54GlKnCIBgfEqaZnHjjC2hzZJJo= =GyP/ -----END PGP SIGNATURE-----