Applications /

Sex, likely lies and software

By Mark Gibbs
Network World, 01/28/02

"The patterns in people's language change when they are uncertain or lying. We can compare basic patterns in words and grammatical structures vs. benchmarks to detect likely lies."
-- Peter Dorrington, business solutions manager at SAS, quoted in the London Financial Times, Jan. 20.

If I understand this correctly (and there's not a whole lot of additional information available about this product), the SAS software compares your patterns of language use in e-mail messages to norms and can determine if you are telling untruths by looking for deviations. . . . Interesting.

Clues to the kind of algorithmic hocus-pocus under the hood comes from a news item in Newsscan (www.newsscan.com), which cites the following from Aldert Vrij's book, Detecting Lies and Deceit: "For instance, overuse of the word 'or' and too many adjectives can be giveaways."

Apparently the scanning process is very fast and detects not only lies but also uncertainty. Thus, if you know you didn't actually solve the problem of why the CEO's computer keeps crashing, or even if you think you might have but aren't sure, with this software watching your e-mail or scanning your trouble-ticketing system, you will be fingered.

The product, which SAS announced on Jan. 21 under the name Text Miner, can sift through e-mail and text files to "identify trends and create more actionable results for management."

What is really interesting is that it heralds a new age in employee monitoring. To date, e-mail monitoring has been, in the main, about blocking dirty words and verboten URLs. There have been attempts at detecting "adult" pictures in e-mail, aka "fleshspotting," but I've heard of nothing that is really worth using.

Anyway, the whole idea of paying attention to tone, context and intention in e-mail is wild. Usually we've just looked for and acted on the anomalies - you know, the accountant who was running a spam factory sending out a million messages per day whom you detected because he got more than his usual quota of "bounces" in a week.

But now we can know when our people are lying in messages traveling up and down the corporate hierarchy, as well as when we're lying to our customers and when our customers are lying to us. And the guardians who run the detection software will know all kinds of interesting things (IT Tech No. 1: "Hey, you know Bill in Engineering? He's cheating on his wife again - the detector says so." IT Tech No. 2: "Same old - give him the usual blackmail price. He's a good customer.").

Of course, this kind of analysis is subject to false positives. The implications and consequences could be serious. You could conceivably lose your job or status over something you never did.

I can foresee a great business here: There's going to be a huge number of people who want to ensure that their messages aren't flagged as lying or whatever they're trying to avoid (even when they are, in fact, doing exactly what they don't want to be found to be doing).Thus, we set up a service on the Internet to clean messages.

We buy a copy of the SAS software and for a reasonably hefty fee we run people's messages through the mill and determine what will raise warning flags. For an additional charge we suggest corrections, and there's a premium "rush" service for press announcements and urgent responses to your boss.

We'll need a few million to get this off the ground so let me know if you're in and for how much. Send me the money and let's get rockin'! I smell riches beyond my wildest dreams! <<NW Message Scanning System: This message has been found to contain indications of uncertainty and untruth. Further detail may be found in the log.>>

Truth or dare to nwcolumn@gibbs.com.