AI-AI-AI!

Once upon a time in Europe, Leibniz (a mathematician) and Newton (a mathematician who liked apples) invented calculus. Generations of liberal arts majors have not forgiven them for this. However, charges of conspiracy to commit mayhem on classics majors were never filed. Why not? (Why not? I cry, remembering what those bastards did to my GPA.) Because, dear hearts, they weren't conspiring. Leibniz and Newton, working independently of each other, each invented the calculus all on his own. What can I say? It was clearly something going around, perhaps a hallucinogenic bread mold. That mold will screw you up.

In other words, stuff happens—sometimes in two places at once. On the other hand, plagiarism happens too. Self-importance happens too. All kinds of wacky things happen, especially in the world of high technology, which is why Bennett Haselton and Declan McCullagh aren't on such good terms today.

AI (artificial intelligence) is the issue here—specifically the AI in BAIR (Basic Artificial Intelligence Routine), a filtering program put forth by Exotrope, a New York-based company. BAIR claims to be able to filter not only the usual naughty words that censorware packages look for but to handle images—in other words, to be able to tell the difference between a naked lady and a knockwurst. Spiffy, no?

Only it doesn't work. The software failed badly in recent tests—blocking Peanuts characters and letting through photos of actual sex acts, for instance — after having received awards from various magazines and sites that claim to evaluate such things. Big news and widely covered, but who figured out that BAIR was bad?

The story broke last week in Wired News under the byline of Declan McCullagh, a well-respected and prolific reporter on the privacy-and-censorship beat. In the story he quotes Seattle's own Bennett Haselton, also a well-respected figure in those circles and someone who has made a career of busting censorware. Problem is, both Declan and Bennett claim to have done the research that brought BAIR down—and Bennett's steamed that Declan's taking credit after he, Bennett, gave him an exclusive on his, Bennett's, findings.

According to Declan, Bennett indeed tipped him off that BAIR had problems. However, Declan claims that his weeklong tests—archived at www.well.com/user/declan/bair—were better than Bennett's, since Bennett didn't in fact set the software to do its work with the AI enabled; Declan says that Bennett's tests (to which he had access) used only the blacklist included in BAIR and not the AI image-recognition capability. Declan says he's surprised at Bennett's anger—"he seems peeved that he didn't get more publicity out of it."

Not true, says Bennett—"it's nice to know that Exotrope got caught in the act no matter who gets credit for it"—but the head of Peacefire.org (now a full-time gig; he quit his job this year to focus on this kind of thing) says that Declan's tests were derived from material Bennett gave him exclusive access to in early June. Bennett acknowledges that Declan's test changed some software settings on Exotrope's suggestion, but says that "we tested it with the AI enabled and some images [were] getting blocked—it was a weaker setting, but it's the default setting that all BAIR customers are using unless they know how to hack the registry," changing settings as suggested on Exotrope's Web site. In addition, the results of the tests were essentially the same—the filtering was "completely random," snorts Bennett—and the same number of porn and nonporn images were blocked. Bennett's findings are archived at www.peacefire.org/censorware/BAIR—and they're dated 6 June, or before Declan's weeklong investigation.

So who busted BAIR? The tests both make the same point, no matter how they were conducted. The question is whether Bennett's findings were treated as central to the story or if the story only became "real" when a representative of the Legitimate Press laid his hands on for the blessing. Much of Bennett's work depends on his ability to get the word out, and that means people like Declan; much of Declan's work depends on his ability to get good leads, and that means people like Bennett. The question is, who really reported the story? The boundary between making the news and reporting the news has rarely looked more like a Mobius strip. (And here I am reporting on the whole dustup. Ain't journalism grand?)

In the words of a very great man, the problems of two little people don't amount to a hill of beans in this crazy world. (OK, Bogey said three people, but fewer people has got to mean fewer beans.) The deal is this: Exotrope's BAIR—an AI image-filtering technique that blocks Snoopy but lets us see orgies—is a mess. Two good people can prove it. Why ask who?

 
comments powered by Disqus