Machine Intelligence vs Human…?

Recently, a number of news sources have run stories lauding the ability of AIs to crack prisoners use of “code words” to conduct illicit business. The International Business Times (IBT) declared that “Artificial intelligence has helped detect secret code words used by inmates on prison phone calls”. New Scientist reported that “Prisoners’ code word caught by software that eavesdrops on calls”.

The “secret code words” on the Internet.

As it turns out the code words were “a three-way”. As IBT informs us, “All inmates are only allowed to call a few people on a previously agreed list of numbers. So if the prisoner wants to deal with illegal business and call someone else, for example, gang members on the outside, they need to call a friend or family member on the agreed list of numbers and then ask that person for a ‘three way’, i.e., to dial the person they actually wanted to speak to into the call.”

Apparently, before the introduction of computer transcribing and auditing of the prisoners’ calls, human beings were incapable of realizing that conversations that went, “Hey, Ma, I need a three-way with Lefty.” “Sure, son. <dead air while call is on hold> Hey, bro, what you need?” was suspicious, especially when it led to discussions of gang business. Granted, the gang business might also be conducted using code words and gang jargon even more obscure than calling a three-way call “a three way”, but still, given the large volume of calls, and the apparent frequency of using these code words, one might have hoped.

No one at the prison spotted the code word until software started churning through calls. (New Scientist)

New Scientist tells us that:

The software saw the phrase “three-way” cropping up again and again in the calls — it was one of the most common non-trivial words or phrases used. At first, prison officials were surprised by the overwhelming popularity of what they thought was a sexual reference.

“One of the most common non-trivial words or phrases used," and yet the AI’s human predecessors were unable to detect the use of this common feature in modern telephones. This suggests that either no one was listening to the recordings of the conversations, or no one was who was informed as to the limits on prisoners’ calls and the capabilities of modern phones.

I am a great believer in the potentials of Machine Learning in the area of voice. My interest in voice systems goes back to when I first used voice-capable systems in the R&D department of Digital Equipment Corp. back in 1980. Seeing it mature in the last 3 dozen years to the point where it is now becoming an every day feature is very satisfying. Still, that it takes a state-of-the-art Machine Learning system to break the “three way” code, and that the business and technical press report that with straight faces is really disheartening.

AI of this sort would not be needed nearly so sorely, if more humans behaved as if they were in possession of natural intelligence. One can imagine the prison warden from 1967’s Cool Hand Luke—he of the famous “What we’ve got here… is a failure to communicate.”—who never used anything but land-lines made of actual copper, not knowing about three-way calling, but in the 21st century it is no longer a secret, not if the telephone, cellular and cable companies are doing their jobs. Prisoners still use land-line pay phones, but prison employees are actually allowed to be familiar with modern conveniences. Heavens, by now there are probably prison employees who’ve never used a pay phone.

Even if prison officials are technologically deprived, and due to their jobs are living in an environment where modern phone services are a rarity, I would have hoped that business and science reporters would be better informed. Did they not chuckle or wonder at the notion that “three way” was a secret code? Did they not care that they were reporting an absurdity?