This AI Hid its True Identity and Passed Turing Test Successfully

Human observers guessed it wrong 50% of the time unable to differentiate between the human and the AI controller

Variability and unpredictability are the chief characteristics of being human. When we assume robots can be human-like or sentient, it is this character of randomness which is looked for. According to a recent paper, "Human-like behavioral variability blurs the distinction between human and a machine in a nonverbal Turing test" published in Science Robotics, the authors have concluded, that robots do have behavioral variability. The AI computer program was trained by researchers from the Italian Institute of Technology to pass a nonverbal Turing test. "Given how our brains are tuned to detect very subtle and implicit signals of human-like behavior, it is worth asking whether a humanoid robot can ever pass the nonverbal version of the Turing test by embodying human characteristics in its physical actions," the researchers wrote in the study. The test consisted of nonverbal tests such as interactive shape and color-matching games, designed to observe variability in the robot's reaction time. In the typical non-verbal Turing test, humans performed a joint Simon task, when placed by the side of an iCub robot.

The Turing test essentially tests how human-robotic behavior can be. Designed by Alan Turing in 1950, it checks if a machine can think like a human based on human interpretation of robotic responses. The game's rules are similar to the party time 'imitation game', except that the robot is one of the participants. In this particular experiment, the humans and robots are put together in a room and colored objects are displayed on the screen for them to identify them correctly, pressing the right button. The researchers controlled the robot with artificial intelligence and human inputs in a random fashion to train the AI toward human ways of reacting. The AI, taking a cue from the response time the human is taking, randomly varied its response time, giving a human-like impression.

Throughout the game, the human participants were asked to guess if the robotic player was controlled by an AI-powered computer program or a human. Surprisingly they couldn't find out AI's hand in it. The human observers guessed it wrong 50% of the time unable to differentiate between the human and the AI controller. The researchers wrote, "Our results suggest that hints of humanness, such as the range of behavioral variability, might be used by observers to ascribe humanness to a humanoid robot. This provides indications for robot design, which aims at endowing robots with behavior that can be perceived by users as human-like."

Passing the Turing test alone wouldn't qualify for AI to become human because Turing imposes the singularity problem on the question of determining the humanness of robotics. The same argument is put forward by Kevin Warwick and Huma Shah, in their paper "The Turing test doesn't mean the end of humanity". They argue that passing the Turing test has no relationship with human-like intelligence because a machine can be reasonably effective with human-like conversation for a short period of time. This is one of the reasons why the AI industry no more considers the Turing test as a benchmark to determine its sentience. Developers have moved on to newer testing methods like General Language Understanding Assessment (GLUE), or Stanford Question Answer Dataset (SQuaD). The attempts toward putting the Turing test to use in judging AI against human psychology only demonstrate our eagerness toward getting AI as close to humans.

More Trending Stories