With the occasion of the anniversary of the death of Alan Turing, the University of Reading decided to set up an experiment where some chatbots were tested to see if they passed the Turing test.
In the early 50s, Turing proposed a test to see if a program reproduced human behaviour. The test is simply to chat with said program and with humans and then decides which is the person and which is the computer program. If the program convinces to more than 30% of judges that they are conversing with a person, for 5 minutes, this program is said to have passed the Turing test.
This year, Eugene Goostman, a chatbot created by Vladimir Veselov and Eugene Demchenko, is said to have passed this test. It is said to have been the first piece of software to have passed the test but, as they comment in this article, it is not actually true. Some other programs already passed the test. The difference is in the amount of conversations involved in the tests. Eugene Goostman fooled 33% of judges out of 300 total conversations we maintained with them.
It simulated to be a Ukrainian 13 years old boy, and that is one of the reasons he passed the test. It didn´t know some things, and it didn´t have good English grammar either.
Turing foresaw that in 2000 we would be able to create computer programs capable of fooling people making them think that they were human. Is that actually a sign of intelligence? Well, it is not. The program doesn´t think for itself. It´s made of algorithms and data to create the illusion that it is human, but it doesn´t think as humans do.
This news reminded me of the TWINKIES project. Do Twinkies pass the Turing test? The answer if hilarious.
No comments :
Post a Comment