These meetings were an opportunity for Alan to develop the ideas for chess-playing machines that had begun in his 1941 discussion with Jack Good.
They often talked about the mechanisation of thought processes, bringing in the theory of probability and weight of evidence, with which Donald Michie was by now familiar.
The development of machines for cryptanalytic work had in any case stimulated discussion as to mathematical problems that could be solved with the mechanical aid—that of finding large prime numbers, for instance, was a topic that came up in lunchtime conversations, rather to the amazement of Flowers, the electronic engineer, who could see no point in it.
But Alan's talk went in rather a different direction.
He was not so much concerned with the building of machines designed to carry out this or that complicated task.
He was now fascinated with the idea of a machine that could learn.
It was a development of his suggestion in Computable Numbers that the states of a machine could be regarded as analogous to 'states of mind'.
If this were so, if a machine could simulate a brain in the way he had discussed with Claude Shannon, then it would have to enjoy the faculty of brains, that of learning new tricks.
He was concerned to counter the objection that a machine, however intricate its task, would only be doing what a person had explicitly designed it to do.
In these off-duty discussions they spent a good deal of time on what would be said to count as 'learning'.