You can't really check based on "output" if output is just text. Likewise you cannot check whether a straw in water is bent, or straight, just by looking at it

You can't really check based on "output" if output is just text. Likewise you cannot check whether a straw in water is bent, or straight, just by looking at it -- you have to take the straw out of the water. The "question-begging" part of the turing test, and all modern AI, is that intelligence is reducible to patterns in an extremely narrowed I/O domain where measurement is only permitted on the "output". This to me is a pseudoscientific restriction: no where in science would we tolerate experimenting on a system to determine its properties with such extreme limitations. Limitations which have all the flavour of the stage magician who says, "please stay seated!".