What exactly does the Turing Test Test?
In the Turing test as Turing proposed it, the interrogator communicates with the contestants with a teleprinter machine. As noted in a previous post, the test is asymmetrical in that the human contestant also has a teleprinter and this is connected to the interrogator’s machine, but the computer contestant is wired directly into the interrogator’s teleprinter. All the computer gets from the interrogator is voltages along a wire – not words (on a screen or printout).
The computer responds to the voltages in the wire because a human (the programmer) had arranged the causality of the computer (by programming it) to send voltages back to the interrogator’s teleprinter that are then converted into words. That is, the Turing Test as Turing proposed it tests the intelligence of two humans, the human contestant and the human programmer of the computer contestant.
This has been a ginormous problem for all of computational AI’s inconsistent history and still is today. For example, in the deep learning image data sets used for self-driving vehicle systems, humans annotate images by putting boxes around trees, people, signs, cars, … and adding the annotation of “tree”, “pedestrian”, “stop sign”, “car”… AI is still producing systems that embody the intelligence of the human. And how much human-like genuine intelligence does the computer itself have in today’s computational AI systems?