DOwNEy | 433
We know that Turing used algorithms to generate ‘pseudo-random strings’, strings which
seemed sufficiently random that they worked in the way that a random source should behave.
We also know that Turing attempted to include a random number generator in the Ferranti
Mark I computer, based on a physical noise generator; however, even today it remains an open
question in physics whether the universe can generate ‘true randomness’ or even ‘algorith-
mic randomness’ in a physical device (see Fig. 39.2). Perhaps Turing was concerned only with
pseudo-random strings in relation to algorithms in (for example) cryptography and artificial
intelligence (AI), as outlined in the following section. Certainly, shortly after 1938, the Second
World War intervened, and when it ended Turing did not return to the topic of normality. He
did not mention randomness except in relation to the efficiency and efficacy of algorithms, and
probably never considered the problem of defining an algorithmically random sequence.
Turing on randomness as a resource
It seems clear that Turing regarded randomness as a computational resource; for example, in AI
Turing considered learning algorithms. As he remarked in 1950:^15
It is probably wise to include a random element in a learning machine . . . A random element is
rather useful when searching for the solution of some problem.
Turing then gave an example of a search for the solution to some numerical problem, pointing
out that if we did this systematically then we would often have a lot of overheads arising from
figure 39.2 The ANU (Australian National University) Quantum Random Number generator. It generates
numbers from quantum fluctuations of the vacuum. The process involved is a leading contender for a physical source
of genuine randomness.
Secure Quantum Communication group, Australian National University.