March 14, 2017


How Algorithms Can Help Beat Islamic State : Hany Farid 'changed the world' by combating child porn. Now his software could suppress terrorists online.. (JOSEPH RAGO, March 11, 2017, WSJ)

[Dartmouth College. Prof. Hany Farid, chairman of the department, is] a founder of the computer-science field known as digital forensics. In the late 1990s as a postdoctoral researcher, he was among the first to recognize that mathematical and computational techniques to authenticate digital images and other media would be useful to society. [...]

In 2008 this research pulled Mr. Farid into another underworld--child pornography. In 2002 the U.S. Supreme Court struck down a ban on "virtual" child porn--computer-generated images that "appear to depict minors but were produced without using any real children." Mr. Farid is sometimes brought in as an outside expert when a defendant claims the material at issue is virtual.

The child-porn industry was nearly defunct by the 1990s, because negatives and videotapes can be confiscated and destroyed. "Then the internet came," Mr. Farid says, "and all hell broke loose."

Supply can create its own demand. Much like jihadists, deviants formed a global community, finding each other online and sharing what are really crime-scene photos. Like ISIS agitprop, material is continuously copied, cut, spliced, resized, recompressed and otherwise changed, in part to evade detection as it is retransmitted again and again.

Mr. Farid worked with Microsoft to solve both problems--detection and replication. He coded a tool called Photo DNA that uses "robust hashing" to sweep for child porn. "The hashing part is that you reach into a digital image and extract a unique signature. The robust part is if that image undergoes simple changes, the fingerprint shouldn't change. When you change your clothes, cut your hair, as you age, your DNA stays constant," he says. "That's what you want from this distinct fingerprint."

The algorithm matches against a registry of known illegal signatures, or hashes, to find and delete photographs, audio and video. Photo DNA is engineered to work at "internet scale," says Mr. Farid, meaning it can process billions of uploads a day in microseconds with a low false-positive rate and little human intervention.

Monitoring by Photo DNA, which is licensed by Microsoft at no cost and now used in most networks, revealed that the nature of the problem was "not what we thought it was," says Ernie Allen, the retired head of the National Center for Missing and Exploited Children. Child pornography was far more widely circulated than law enforcement believed. "Hany Farid changed the world," Mr. Allen adds. "His innovation rescued or touched the lives of thousand of kids, and uncovered perpetrators, and prevented terrible revictimization as content was constantly redistributed."

Mr. Farid linked up with the Counter Extremism Project to apply the same robust-hashing method to extremist propaganda. But this effort has encountered resistance. "The pushback from the tech companies has been pretty strong," the project's Mr. Ibsen says dryly.

U.S. law immunizes internet companies from criminal and civil liability for content that travels over their transoms. Their terms of service forbid abusive content, but they rely on users instead of algorithms to police violations. "It's a very slow and tedious process: You wait for it to get reported, somebody has to review it, they make mistakes," Mr. Farid says. "They take down the Vietnam napalm girl on Facebook."

Liability aside, what about their moral obligations to help prevent death, injury and destruction? "In my mind, we're not asking them even to do something that they haven't said they want to do already. We're saying, hey, would you please do the thing that you promised you would do?" he explains. "I am simply saying, look, for free, you can automate this and make it really efficient and really fast and save you money on the side."

It's just a different deviance.

Posted by at March 14, 2017 11:18 AM