Musings on the Turing Test


Suppose that intelligent, conscious, etc. persons A and B are taking the Turing test. Person A speaks the language X as his native language and person B is a native speaker of Y. They know no other languages - therefore they do not understand each other. Person A thinks person B is typing nonsense, and vice versa. Thus both parties see each other as failing the Turing test - both consider the other party being unintelligent

In the described case, the Turing test obviously does not function correctly. Now, substitute either person for a computer and re-run the test. The test fails. But is the failure because the computer was unintelligent, or was it because the computer was not capable of speaking the language spoken by the human, or both?

The focus in AI systems undertaking the Turing test seems to be in making the computer capable of outputting some semi-random but linguistically coherent nonsense by reacting to inputs given by the human. This output should then convince the human to believe the computer to be intelligent. However, this does not truly provide a proof that an AI system under test is intelligent. It might appear to be intelligent, although that is all, just appearance - it is not truly intelligent but will instead aim to emulate intelligence up to some level of plausibility. Or, an AI system might appear to be non-intelligent, although it is highly intelligent, in a case similar to the example with persons A and B.

Instead of trying to make computers learn human languages as their "first language", why not instead try to give the computer the capability of creating its own language, whatever it would be, to make it "think" with this language, to form associations, to make it converse with that language with other systems (maybe the whole language could be created through "socialising" among multiple similar systems), to have an internal dialogue of some sort, and so on. This artificial language of the computer could then perhaps be translated to human languages. Perhaps this approach would be offer a path to reaching true machine intelligence and machine consciousness.