The purely behavioural nature of the Turing Test leaves many with the view that passing it is not sufficient for 'intelligence' or 'understanding'. We propose here an additional necessary computational requirement on intelligence that is non-behavioural in nature and which we contend is necessary for a commonsense notion of 'inductive learning' and, relatedly, of 'intelligence'. Said roughly, our proposal is that a key to these concepts is the notion of compression of data. Where the agent under assessment is able to communicate, e.g. by a tele-type machine, our criterion is that, in addition to requiring the agent's being able to pass Turing's original (behavioural) Turing Test, we also require that the agent have a somewhat compressed representation of the test domain. Our reason for adding this requirement is that, as we shall argue from both Bayesian and information-theoretic grounds, inductive learning and compression are tantamount to the same thing. We can only compress data when we learn a pattern or structure, and it seems quite reasonable to require that an 'intelligent' agent can inductively learn (and record the result learnt from the compression). We illustrate these ideas and our extension of the Turing Test via Searle's Chinese room example and the problem of other minds.

We also ask the follwing question: Given two programs H₁ and H₂ respectively of lengths l₁ and l₂, if H₁ and H₂ perform equally well (to date) on a Turing Test, which, if either, should be preferred for the future?

We also set a challenge. If humans can presume intelligence in their ability to set the Turing test, then we issue the additional challenge to researchers to get machines to administer the Turing Test.

Keywords: Turing Test, Philosophy of AI, compression, Bayesian and Statistical Learning Methods, Machine Learning, Cognitive Modelling.