DOT watching the Web

Watch what you post on Facebook. The US DOT is watching.

Department of Transportation Undersecretary Derek Kan recently told an industry group, “We are connecting diverse data sets with comprehensive and timely analysis for anybody that’s on social media.”

It seems they consider social media is another source of information, along with things like traffic counts and accident stats.

The quote in Transport Topics was from the recent meeting of the Transportation Research Board where Kan was discussing new tools in the field of predictive analytics – the art of identifying heightened safety risks and preventing crashes.

Is it true? Is the DOT watching people on social media?

I reached out to the DOT multiple times over a week. They did not respond. So it’s pretty clear – to me at least – their nonresponse is deliberate. Maybe it’s supposed to be a secret. Or maybe they just didn’t want to talk about it with a trucking writer.

Either way, we now know they use social media for information, but whose? Truck and bus drivers? All licensed drivers? And what are they doing with that information?

Kan spoke of identifying dangerous locations and conditions with new technology tools and data from 765 sources. According to Transport Topics, such “big data” analysis has revealed unlighted roadways can sometimes be safer than lighted ones and crashes in Los Angeles occur more often in some neighborhoods than in others.

But those conclusions came from observable, measurable data. How does that relate to social media data, which is all about individuals? Has the DOT found something reliably measurable in what people post on Facebook?

At least one company is trying to do just that. Enlistics is a Web-based company that helps carriers hire drivers and screen them using social media. The company looks for phrases it claims can predict the likelihood of success or failure.

Tandem Thoughts featured Enlistics a year ago. Then founder Austen Mance told me, “One such ‘bad’ phrase known to correlate with employee turnover is ‘I’m so drunk.’” He also said research shows people who root for home teams are a better hiring bet than those who root for out-of-town teams. I’m sure he’s statistically right.

The problem arises when an individual who roots for the team back east where he grew up posts a joke with the phrase “I’m so drunk” in it. He might be a great driver, exactly what the fleet is looking for, but the algorithm that analyzed his social media posts does not understand the context. He doesn’t get the job.

We don’t know if DOT researchers scan social media in a similar way – mining words or other information by or about an individual. A mile is a mile. A mouse click is a mouse click. But a word or phrase cannot always be taken at face value. In context, it may mean something different, maybe even the opposite. With all the unknowable factors – including whether or not a post is even honest – any statistics that result will be questionable. So will the conclusions drawn from them. An adage from the early computer days seems to apply here: garbage in, garbage out.

Maybe DOT researchers have a totally different approach to social media, one that bypasses the stuff people post. I doubt it, but I hope so.

At some point – if it isn’t already happening – predictive analytics will be deciding who gets a job and who doesn’t. I’m not looking forward to that day. Far worse, though, is the prospect that it may someday be deciding who gets to drive for a living and who doesn’t.

That kind of technology takes us into the realm of “Minority Report,” the 2002 movie about a future society in which a crime can be predicted before it happens and the potential criminal arrested before he has done anything at all.

In a world where artificial intelligence is now being used to help develop more artificial intelligence, we may get there sooner than anyone thinks.