NPR had a good segment on this today: The Difficulty Of Enforcing Laws Against Driving While High. Definitely well worth a listen.
I had a few reactions in a comment that ended up being post-length, so here you go.
The major discussion of the segment was two-fold and I think illustrates where policy based on the science can be helpful, even if only to point to what we need to know but do not at present.
The first point was that THC hangs around in the body for a very long time post-consumption, particularly in comparison with alcohol. Someone who is a long term chronic user can have blood THC levels that are...appreciable (no matter the particular threshold for presumed impairment, this is relevant). Some of the best data on this are from the laboratory of Marilyn Huestis when she was, gasp, an intramural investigator at NIDA! There are some attempts in the Huestis work to compare THC and metabolite ratios to determine recency of consumption-that's a good direction. IMO.
The second argument was about behavioral tolerance. One of the scientist interviewed was quoted along the lines of saying the relationship between blood levels, repetitive use and actual impairment was more linear for alcohol than for THC. Pretty much. There is some evidence for substantial behavioral tolerance, meaning even when acutely intoxicated, the chronic user may have relatively preserved performance versus the noob. There's a laboratory study here that makes the point fairly succinctly, even if the behavior itself isn't that complex. As a counterpoint, this recent human study fails to confirm behavioral tolerance in an acute dosing study (see Fig 4A for baseline THC by frequency of use, btw). As that NPR piece noted, it would be very valuable to get some rapid field screen for THC/driving - relevant impairment on a tablet.