Learnability thesis, FM-representability and Turing Ideals
Michał Tomasz Godziszewski
We consider the notion of the so-called intuitive learnability and its relation to the so-called intuitive computability. We briefly present and discuss Church’s Thesis in the context of the notion of learnability. A set is intuitively learnable if there is a (possibly infinite) intuitive procedure that for each input produces a finite sequence of yeses and nos such that the last answer in the sequence is correct. We further formulate the Learnability Thesis which states that the notion of intuitive learnability is equivalent to the notion of algorithmic learnability. The claim is analogous to Church’s Thesis. Afterwards we analyse the argument for Church’s Thesis presented by M. Mostowski (employing the concept of FM-representability which is equivalent to Turing-reducibility to $0’$ or to being $\Delta^0_2$). The argument goes in unusual lines – by giving a model of potential infinity, the so-called FM-domains, which are inifnite sequences of growing finite models, separating knowable (intuitively computable) sets from FM-representable (algorithmically learnable) ones via the so-called testing formulae and showing that knowable sets are exactly recursive. We indicate which assumptions of the Mostowski’s argument (concerning recursive enumeration of the finite models being elements of an FM-domain) implicitely include that Church’s Thesis holds. The impossibility of this kind of argument is strengthened by showing that the Learnability Thesis does not imply Church’s Thesis. Specifically, we show a natural interpretation of intuitive computability – containing a low set – under which intuitively learnable sets are exactly algorithmically learnable but intuitively computable sets form a proper superset of recursive sets. Last, using the Sacks Density Theorem, we show that there is a continuum of Turing ideals of low sets satisfying our assumptions. Therefore, if we admit certain non-recursive but intuitively computable relations (namely: some low relations) we are able to consider expanded FM-domains and there are continuum many of such. On the other hand, relations FM-representable in such FM-dommains are still $\Delta^0_2$. The general conclusion is that justifying Church’s Thesis solely on the grounds of procedures recursive in the limit is by far insufficient.
Michał Tomasz Godziszewski is visiting CUNY from September 2016 until March 2017.