VIDEO TURING TEST(VTT)

What is the Video Turing Test (VTT)? The original Turing Test of 1950, named after its creator Alan Turing, is a test for a machine to prove its ability of intelligent behavior equivalent to or possibly even indistinguishable from that of a human. For this test, an evaluator would receive natural language conversations between a machine and a human (via a text-only channel to not reveal the machine's identity) and had to determine whom of the participants is a machine. If the evaluator could not distinguish the machine from the human, the machine passes the Turing Test.

The Video Turing Test desires to test a machine's ability of intelligent behavior in regards to observing and understanding video input and thereby its video intelligence. A machine capable of this task would prove human-like video understanding capabilities that could open the world of AI to a whole new possibilities in human-like long-term adaptive learning.

GRAND CHALLENGE OF AI

BEYOND THE TURING TEST

The development of a system capable of successfully passing the VTT could also inspire different forms of promotion. An event like this could be of the following contest style based on a popular Korean TV Show.

Hold an "Beyond the Turing Test" event to ensure that the system has video understanding capabilities.

Participants of humans and machines watch a video together and have to answer questions while an audience of about 100 people, who don't know whom of the participants is human or a machine, has to evaluate the participant's answers to find out who is actually a human and who is not (Turing Test similarities). This could be a broadcasted event through various forms of media.

Method

Event Place : A hall that can accommodate all of the participants, evaluators/audience and the event staff.
Such a hall could for example be the MBC Open Hall.

Audience Selection : One hundred randomly selected from the general pulic after applying online.
Considerations of various age groups, educational backgrounds and gender should be taken to ensure a variety of evaluators.

Participants : AI Systems and three human control group participants of people in their 20s to 30s.
Measures are needed to keep the identity of all 6 participants hidden from the audience.

RESEARCH GOAL

The goal of the research is the development of a system that is capable of comprehending video stories and of skillfully solving Question and Answer tasks in order to pass the Virtual Turing Test. The achievement of this target would not create the end to research in this field. Instead, one may confidently say that a system successfully passing the Virtual Turing Test is capable of starting a new era in AI Research.

Development of Q&A Technology
by Understanding Video Stories that can pass VTT

RESEARCH CONTENT

The research behind the goal of teaching a robot with videos to pass the VTT is based on a 5 year plan. The stepping stone for the base start of the research is based Children’s Story Cartoons for they come in great amounts and with a great amount of advantages. They are of multimodal character, as they provide both vision and language to be processed, however, they also tend to follow simple grammar rules and explicit story lines. Furthermore, a cartoon’s advantage can be their simplistic image processing characeristic, their pseudo-realness and their educational content. A cartoon based Question and Answer database may be created to train the system until it may reach the further destinations of VTT Demonstration, TV Drama Learning and VTT Competition.