"I believe that in about fifty years' time it will be possible to programme computers to make them play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning". Alan Turing, 'Computing Machinery and Intelligence', Mind (1950), 442

A Computer Vision Turing Test

Computer vision involves the analyses of images and videos in order to understand the environment. It brings an element of artificial intelligence (AI) as it encompasses fields of study from computer science, mathematics, and engineering, among many others. I believe it would be interesting therefore to consider a Computer Vision Turing Test in which scene analysis results by a computer vision system would be indistinguishable from an analysis by a human.

The Turing test has been proposed for more than 50 years and since then there has since a great level of development in the modeling and simulation of perceptual processes. However, an AI system that passes the Turing test with flying colors remains an illusive goal. This is despite the huge shift in computing horsepower and storage.

Goals of a Computer Vision Turing Test

A Computer Vision Turing Test would be driven by two goals. First, the computer vision system would perform equally well or exceed those of a human subject. This aspect gives futuristic direction and goals for researchers. It is easily noted that today’s vision systems can often err on their analyses of a scene, when an average human would easily recognize and understand the same scene.

Second, the prospect of a Computer Vision Turing Test would serve to humanize a computer vision system. Just as the human perception can be fooled easily with optical illusions and other perceptual trickery, it would be reasonable for computer vision system to fail as well. This basis of comparison might be important, for example, in the field of personal robotics system.

One should notice that these two goals overlap and, at the same time, they are conflicting. This is done on purpose such that the goals are vague and open to interpretation. Under the first goal, researchers would seek to exceed human perceptual capability, allowing freedom to explore new techniques and capabilities. On the other hand, under the second goal, researchers would set the bar lower with an understanding that embedded vision systems are often less than ideal. It is this balance of interpretation, a ying and yang if you will, that maintains and reinforces positive advancement in research.