Any reasonable mechanical process can be realized on a Turing machine. What’s a reasonable mechanical process? It must be a process that requires only finitely-many steps and finite time to follow through them; it must not require special (human-like) ingenuity, cleverness, or creativity to follow the steps; it must give the right answer every time.
Apart from the challenge of translating a mechanical process either witnessed in operation or represented in some kind of communication (perhaps a verbal description), that is, apart from the challenge of programming a computer, the Church-Turing Thesis states that any reasonable mechanical process that can be described or implemented can be programmed and realized in a computer. So if some expert, such as a medical doctor, can describe how she comes to a diagnosis, then her process (her thinking) can be realized on a computer (or in a robot or whatever). However, after many expert interviews and much programming, there is still something missing in every ‘artificially intelligent’ system produced over the decades.
Thus the challenge to the field of artificial intelligence is not the programming, although at times that can be quite hard. The challenge is not limited computational resources (time and space); those are non-issues if the right algorithm or heuristics (shortcuts) are used. The challenge is rather that we are unable to reasonably mechanize intelligence. We just cannot figure out how to describe a reasonable, mechanical process for recognizing your grandmother.
How do you recognize your grandmother? Do you just match a visual pattern from your memory to visual image of the face of the person standing in front of you, and if there is a high-confidence match you declare, “this is my grandmother” ? What if she turns her head? That visual pattern is useless now. What if you cannot see her, but smell her perfume? What if your grandmother is physically far away, but you just get that sense that she’s nearby? Can all these senses, these forms of intelligence be reasonably mechanized; that is, can they be described simply and clearly, without resorting to vague abilities like “intuition” or “just knowing” ?
The magic of the machine is not in the machine or the machine itself. The machine is just a stupid automaton; it does whatever and only what it is told. The magic, the soul of the machine is rather that what the machine is doing is mechanical at all. If a machine is programmed to recognize your grandmother, or hold conversations about interesting topics with everyday language, the magic is not that the machine does so, but rather that, demonstrably, shockingly, the act of recognizing your grandmother or conversing in everyday language requires no ingenuity, cleverness, or creativity.
Well, that’s the challenge, anyway. Researchers in artificial intelligence seek to find a reasonable mechanical process for recognizing your grandmother, among other tasks. If such recognition requires ‘creativity,’ then the researcher is challenged to find a reasonable mechanical process that does ‘creativity.’ It is not usually relevant whether or not the mechanical process is equivalent to the way a human brain does creativity. However, a researcher does not accept that creativity cannot be reasonably mechanically realized. Why should it be so, that creativity cannot be mechanized reasonably? If the reply is, “because creativity requires inventiveness,” then the researcher is challenged to reasonably mechanize inventiveness. If the reply is, “but inventiveness requires creativity!” then the researcher shall cry, “foul play!”
The machine represents the paradoxes of our very existence as thinking, rational, civilized beings capable of expression. On the one hand, it is a sign of determinism, of the sheer predictability of mechanism, including the mechanism of the models we employ to describe our world and selves that we have grown from empirical evidence. On the other hand it is an expression of our inventiveness and freedom. — David Porush, The Soft Machine: Cybernetic Fiction
