6 ideas
When a question like 'Can machines think?' turns on words too vague to define, replace it with a closely related question that can be settled by observation. The imitation game does this: an interrogator talks by text to a hidden machine and a hidden human, and if the interrogator cannot reliably tell which is which, the machine has passed. Arguing over the meaning of 'think' then gives way to a question that can be tested.
If a machine may be called thinking only when we can be sure it feels, we would need to be the machine to know. The same standard applied to other people would leave each of us certain only of our own mind. We avoid that by adopting the 'polite convention' that everyone thinks, and the objector has no grounds for refusing the same convention to a machine that behaves the same way.
Lady Lovelace's objection is that a machine only does what we order it to do, so it can originate nothing. Turing replies that machines surprise him often, because nobody can foresee all the consequences of a set of premises and instructions. Being fully determined by instructions is not the same as being predictable to the person who wrote them, so the capacity to surprise does not require creative independence.
A digital computer with enough storage and enough speed can be programmed to imitate any discrete-state machine. The question of whether some machine could pass the test therefore reduces to whether a suitably programmed general-purpose computer, with adequate storage and speed, could pass it. The particular hardware matters less than the program, the storage capacity, and the speed of operation.
Rather than program an adult mind directly, build a simple 'child machine' and teach it through something like rewards, punishments, and instruction. This splits the problem into three parts: the initial mechanism, the education, and other experience. Improving the child machine by trial and selection resembles evolution, with the experimenter's judgment standing in for natural selection.
Many objections to machine intelligence rest on lists of things a machine 'will never do', such as enjoy strawberries, fall in love, or make mistakes. Turing suggests these claims mostly come from induction over the limited machines people have seen so far, not from arguments about what machines could be in principle. Asking of each claimed disability what evidence supports it often shows that it reflects current engineering rather than a demonstrated necessary limit.