Video of Lecture at Singularity Summit AU 2012:
David Dowe
dismissed by many as a practical intelligence test. In fact, it is questionable that the imitation game was meant by Turing himself to be used as a test for evaluating machines and measuring the progress of artificial intelligence. In the past fifteen years or so, an alternative approach to measuring machine intelligence has been consolidating.
The key concept for this alternative approach is not the Turing Test, but the Turing machine, and some theories derived upon it, such as Solomonoff’s theory of prediction, the MML principle, Kolmogorov complexity and algorithmic information theory. This presents an antagonistic view to the Turing test, where intelligence tests are based on formal principles, are not anthropocentric, are meaningful computationally and the abilities (or factors) which are evaluated can be recognised and quantified. Recently, however, this computational view has been touching upon issues which are somewhat related to the Turing Test, namely that we may need other intelligent agents in the tests. Motivated by these issues (and others), this paper links these two antagonistic views by bringing some of the ideas around the Turing Test to the realm of Turing machines.
Paper: Turing Machines and Recursive Turing Tests by Jose Hernandez-Orallo, Javier Insa-Cabrera, David L. Dowe and Bill Hibbard