The system described in a paper in the International Journal of Intelligent Information and Database Systems (I link it here) could be used in places where drunk driving and drunk behavior are common problems.
Good and right thing: because there are more than one million deaths worldwide every year due to road accidents. Many of these are the direct result of drunkenness.
A systematic review
Kha Tu Huynh e Huynh Phuong Thanh Nguyen of the Vietnam National University in Ho Chi Minh City explain that the first parameters studied to detect drunkenness were the eyes, the position of the head or other indicators of functional status.
These parameters, however, could be confounded by other factors. The team preferred thermography analysis: it would be ideal for events where alcohol is likely to be consumed and people will drink and drive. Why? Three reasons: Non-invasive, more effective, and above all more reliable.
Drunkenness, first of all precision
The team points out that it is important that a artificial intelligence designed to identify drunkenness in people has a very low rate of false positives and false negatives.
And he's right: with too many false negatives the cars would be full of drunks. With too many false positives there would be too many sober people forced to take taxis (and to lose faith in the authorities).
There will always be a trade-off in any such system, and the newly developed one records a reasonable 93% accuracy.
It will be able to improve further, say the authors of the study, by "learning" from all the data it collects. Anyone who thinks that one day an algorithm can look us in the face and determine our drunkenness with 100% accuracy, well, he's been drinking.
