[zbyte64]

Put both in separate rooms you cannot observe and equip them with a two way mic/speaker. Would you be able to determine which was a meat bag and the other was a simulation? If you can then we know the description is inaccurate. If you can't then not only is the description accurate but the substrate that the computation occurs in is not equivalent to what defines the person as a particular agent.

[westoncb]

Two things:

1) The A.I is not just a set of symbols, it's a physical machine, which means you're answering a different question.

2) The structure of the test means that you have stripped down the objects to be compared to the sounds they produce, and yet a person is more than sounds, as is the computer. Whenever we use the relation 'X is a Y' there is an implicit definition of the identities of X and Y; the more abstract it that definition is, the more things your relation will accept; at some level of abstraction the human and the A.I. may be called equivalent—but you'd have to state which one you're talking about.

[zbyte64]

1) You need something to compute the outcome of the symbols in order to test the accuracy of those symbols. I am not sure how to experimentally test your point without doing so. Perhaps that is the crux of the issue, we can't test symbols without tying them to a "physical medium".

2) We can introduce more communication layers as you desire and still keep in the spirit of the experiment. I opted for the strip down version due to simplicity of describing the experiment and the similarity it has to the turing test.