Sunday, August 09, 2009

Turing Test Redux

A recent article in The Economist (“Diagnosing Comas: Unlucky for Some” July 25th, 2009) pointed out that distinguishing between different types of comas is difficult even for specially trained physicians. If someone is in a “persistent vegetative state” they show no signs of consciousness at all. It may be merciful, and legal, to cut off their food and water and let them go.

Other coma patients are in what’s called a “minimally conscious state,” meaning they can sometimes communicate by blinking or moving their eyes in response to questions. That communicative consciousness may be intermittent, displayed only for a few minutes in a month, but it is enough to make a large moral and compassionate difference between the two coma states.

A recent study in Britain found that 40% of patients diagnosed as being vegetative were actually not. Careful and detailed screening tests for communication can show up the difference, but most doctors do not use these tests, preferring to rely on “clinical experience.” This replicates a similar finding from a decade ago.

Unsettling as the finding is, one interesting aspect is the use of what amounts to a Turing Test as the definition of consciousness. In 1950, computer scientist Alan Turing proposed a way to tell if a person (or a computer, for that matter) is conscious. In the now-famous “Turing Test,” you have a conversation with a robot, and a person, both hidden from you by a curtain. If you cannot tell which is which, the robot passes the test and you must, to avoid inconsistency, admit that it is conscious. So the ultimate criterion of consciousness is meaningful communication.

Unknowingly, the researchers whose work was reported in The Economist article were using a variant of the Turing Test to determine if a coma patient is conscious or not. If the patient can communicate, they are conscious. If not, they are “vegetative.”

Is that a criterion we are comfortable with? Are we quite sure that vegetables have no consciousness? Are we perfectly clear on what constitutes “communication?” If I ask a tree how it is feeling and it suddenly bends way over in the wind, has it answered me? Who is to say?

The Turing test has been hotly debated among cognitive psychologists and A.I. researchers for half a century and is by no means universally accepted. It seems odd that the pinnacle of neurophysiological practice would now strive to depend on it.