The Musical Turing Test: Music instead of Language as a medium for the cognitive evaluation of computing systems
Finding – or developing – an epistemologically safe criterion for distinguishing between intelligent and non-intelligent entities remains a basic aim of the research program known as “Artificial Intelligence” (A.I.). An organized endeavor for the establishment of such an ontological criterion begins with Descartes and reaches to our age where is expressed through various versions of the “Turing Test”. Posing the question “Can machines think?” and being inspired by a popular game (the “imitation game”) played in the parties during the ’40s, Alan Matheson Turing introduced what was finally called “the Turing Test”, an experimental kind of procedure in which a group of humans (“judges”) interrogate the contestant entities (i.e. humans and computers competing to prove their intelligence). The interrogation has the form of a one-on-one conversation in which a judge communicates with one contestant asking questions of various subjects and the contestant tries to respond in an intelligent-like manner. The communication takes place through a net of computers, pretty much like the Internet “chatting”, so that no visual or aural information about the contestant is possible. The Turing Test was meant to be faced by the A.I. researchers not only as a basic experiment-criterion of their research products, but also as a manual guide – or as a steering instrument – for the production of various theories of intelligence. These more than sixty years of applying the Turing Test as a means of the A.I. project evaluation have led to numerous philosophical conflicts and have highlighted a series of significant theoretical errors and weaknesses which in the last decade we have been trying to overcome by introducing various versions of artistic-and especially musical-Turing Tests. In this case Music substitutes for Language and becomes the medium via which the Turing Test is constructed. In the present paper: 1) We try to highlight the fact that by moving from the lingual to the musical medium not only do we fail to avoid the problems of the “classical” version of the Turing Test, but also we find ourselves trapped in the subjective character of any aesthetical judgment. In its turn this aesthetical subjectivity leads to a respectively subjective judgment regarding the intelligence of the contestant entities and harms both the evaluation part of the A.I. research project and the research on Music Cognition. 2) We examine the capacity of the Turing Test to survive as a new research “tool” through a new approach and use in which our interest turns from the contestant to the “judge”. Such a reflective turn might restore the validity of the Turing Test, turning the latter from an ambiguous – even discredited – A.I. criterion into a useful research “tool” for musicologists.