Drunk driving is a big problem in America, accounting for up to 31% of all traffic fatalities in the country. For years, education and awareness campaigns have tried to address the issue, but could the real solution lie in teaching cars themselves to spot intoxicated drivers? The National Highway Traffic Safety Administration has been wondering the same thing and recently sponsored a study to find out.

Study methodology

The study was conducted at the University of Iowa's National Advanced Driving Simulator. A team of researchers evaluated 108 participants on simulated drives from an urban bar to a suburban neighborhood. The drive included city, highway, and rural roads with varying speed limits and a variety of intersections, exit lanes, and so on.

The drivers themselves were classified as "moderate" to "heavy" drinkers, ranging in age from 21 to 68. They were each tested at three blood alcohol concentrations: 0.0%, 0.5%, and 0.10%.

The study used onboard computer systems to analyze the drivers' habits and determine whether or not they were impaired. The computers were set up to evaluate a range of behaviors, including:

* control inputs (e.g., steering wheel and brake pedal movement),

* vehicle state (e.g., accelerometer and lane position),

* driving context (e.g., speed zone information and proximity of surrounding vehicles), and

* driver state (e.g., eye movements and posture)

The highlights

The researchers' report to the NHTSA is something of a doozie -- you can download the full 98-page PDF here -- so we'll summarize some of the high points.

* On the whole, the study's computerized systems were slightly less accurate in judging impairment than the standard field sobriety test (SFST), matching the SFST results between 73% and 86% of the time.

* The most accurate means of evaluating driver impairment wasn't her/his use of control inputs (e.g. how regularly s/he used turn signals), but long-term vehicle state (e.g. whether the driver consistently kept the vehicle in its lane).

* When the computers used the full range of factors to assess drivers, it took approximately 25 minutes to determine whether the drivers were impaired.

* When the computers focused on specific information -- namely, key actions by the drivers, paired with information about the individual -- the time dropped to eight minutes.
Those numbers might improve in real-world situations, since the majority of alcohol-related accidents -- roughly 66% -- involve drivers with blood alcohol concentrations of 0.15% and higher, meaning that their performance could be easier to spot.

Our take

This study raises a lot of interesting facts -- and even more intriguing questions. Here are three of the major thoughts that crossed our minds while reading the document:

* Computerized systems like this are inevitably part of the auto industry's future. Until vehicles become fully autonomous (hurry up, Google), this kind of technology will occupy increasingly larger portions of our cars' onboard computers. It's not too different from the collision-avoidance technology that's already found on an increasing number of vehicles, and it's not a huge jump to get from here to Ray LaHood's vision of vehicles that monitor driver's whole health, either.

* The stats unearthed by the study point to something scary: fatalities from drunk driving have remained constant, even as cars have become safer. If alcohol-related traffic accidents are keeping pace with safety technology, that may mean that the frequency of drunk driving is on the rise, too.  

* Are computerized monitoring systems like this the answer to the problem? Why not just add breathalyzers to ignition systems? Apart from the outcry that might come from certain libertarian groups, breathalyzers alone can't solve the problem. Humans are pretty clever when it comes down to it, and addressing the problem of drunk driving requires a similarly clever, multi-pronged solution.

Whatever the solution to drunk driving may be, it's high time to find it.

NB: The study cited in this article was initially released in August 2010 but was recently updated with additional information.