What happens if our artificial intelligence systems cannot be manipulated enough to pass the Turing test?
The Turing test postulated by Alan Turing, the famous mathematician who cracked the German Enigma Code, states that for artificial intelligence to really be that, it must be able to fool a human into thinking that the human corresponds to another human and not to another human. a machine. That’s setting the bar pretty high for computer scientists, isn’t it? It surely is, but I have another thought on all of this, as chatbots with very little AI can already do this online. Let’s talk.
You see, there was an interesting article on BBC Science News on May 9, 2015 titled; “A Question of Computers and Artificial Intelligence” by Peter Day, a global business correspondent that really got me thinking and asking a question. You see, it is inevitable that AI will soon surpass human intelligence in all respects and then never look back, or it will, our future Artificial Intelligence, will find it necessary to fool itself, pretend to be as stupid as humans to gain our trust. . to serve us? The article said:
“When machines can outthink humans, it makes a lot of headlines. But those closest to us are wary of claims made by experts like Ray Kurzweil, Google’s chief engineer, that the human race will soon be eclipsed by intelligent machines. Mr. Kurzweil has long been convinced that one year (perhaps in 2050) computers will have evolved to be as smart as we are. Two years later, following Moore’s Law, they will be twice as smart.” .
Granted, if all of the AI will soon be smarter than even the smartest human, it’ll be pretty easy to spot them, so the only way to get past this and trick a human into thinking they’re another human would be to severely cheat. at the human cognitive level. In fact, chatbots are already doing this, deliberately misspelling words and using bad grammar and slang. Turns out this works. In the future, the same methodology will be used by AI to fool humans, because humans will program the next generation of AI which will then learn what works, and since this strategy works so well, it will be implemented more frequently.
Perhaps passing the Turing test will require a bit of “playing stupid” from our future Artificial Intelligence. Please consider all this and think about it.