A king is seeking a new advisor, and to do so he invites three wise men to his castle. He tells them he will place a hat on each of their heads that will be either white or blue, and at least one of the hats will be blue. The wise men must work out the color of their own hat they are wearing without talking to each other to become the advisor. After a few minutes of sitting in silence, one of the wise men stands up and guesses correctly.
This riddle (you can read the solution here) is a famous test of logic and self-awareness, and a group of researchers have now recreated a similar test in robots to prove the ability of artificial intelligence to be self-aware – within, of course, limitations.
Three humanoid Nao robots were programmed to think that two of them had been given a “dumbing pill” that prevented them from speaking. All of them were asked “which pill did you receive?” but as two of them were mute, only one was able to answer, saying: “I don’t know.” It then works out that, as it can talk, it must not have been given the pill, so it changes its answer to: “Sorry, I know now. I was able to prove that I was not given a dumbing pill.”
Results of the test, carried out by the Rensselaer Artificial Intelligence and Reasoning (RAIR) Laboratory, will be presented in a paper at RO-MAN 2015 later this year. Selmer Bringsjor from the Rensselaer Polytechnic Institute, one of the test’s administrators, told Vice that it showed that a “logical and a mathematical correlate to self-consciousness” was possible, suggesting that robots can be designed in such a way that their actions and decisions resemble a degree of self-awareness.
Before you start preparing for an onslaught of Terminator-style killer robots, though, it should be noted that this test was obviously rather limited. Nonetheless, it suggests that self-awareness is something that can be programmed, and may open up new avenues for artificial intelligence. Just being able to understand the question and hear their own voice to solve the puzzle is an important skill for robots to demonstrate.
“There are myriad additional steps that need to ultimately be taken,” the researchers write in their paper, “but one step at a time is the only way forward.”