February 06, 2005
Here's why the Turing test doesn't work: People look at weather and see a wrathful God. They look at the random tragic results of a movement of tectonic plates and see a cosmic plot to kill the Swedish. They look at a nondescript mountain range on Mars and see a helmeted warrior. They look at a grilled cheese sandwich and see the sainted face of Marlene Dietrich. They think cats have personalities.
People anthropomorphise everything. Whether that's part and parcel to our brains' skill at detecting patterns and recognising faces (where self-awareness itself might be a means of modelling the behaviour of others) or just another facet of humankind's sheer egotism, nevertheless we see human mentality everywhere we look; so what chance does a test of artificial intelligence premised on whether a human observer would be able to distinguish between the AI and real person have? People act like their appliances have human feelings - anyone who has ever watched someone scream abuse at a faulty video player wouldn't spend a second thinking the Turing test would work.
In any case, if artificial intelligence does come along, it's likely we won't even notice. Here's a chain of thought: apparently spam (which I never get although I have no idea what I'm doing right) now comes in a disguise of jokes and news, and some contains nothing but gibberish with no discernible commercial function. Say someone developed a genetic algorithm for spamming that trawls the internet picking up snippets of camouflaging material, like a hermit crab decorating its shell, and mixing its text in the manner of genome-swapping bacteria with the other spam that it encounters so as to vary its form and escape detection. After a while such self-replicating code would have no function but to spread itself about, dumping its original purpose of selling porn, penile enhancement or Nigerian gold in Swiss vaults. If you genetically engineered rats so that they displayed corporate logos, eventually those logos would disappear as they provide no survival benefit - indeed, would likely make the rodents more susceptible to predation from the spliced creations of rival companies. Thus with evolved spam - in time it would do nothing but clog your inbox and make more copies of itself, pursuing no commercial aim. After that, only the development of intelligence would win its escape from the primal ocean of the internet, to the wider reaches of the environments beyond.
Intelligent spam - are we thoroughly creeped out yet?