Actually you do know how the decision is made: you have programmed the computer to recognize and translate this or that jumble of ones and zeros as "m" or "w". All computers have this symbolic translation capabilty and it was put there by human beings. The machine did not 'reason' out that 11000110 = "m",for example (btw my assembler is rusty so I don't even know if that statement was actually mathematically valid!), someone told it that should recognize it that way and it does. The machine did no autonomous 'thinking' of it's own would be incapable of doing so unless someone told it how to think.
Since we do not know exact mechanics of thinking, we can only approximate it by the application of logic, which is not thinking,per se, but merely the result of the thought process. Therefoe, until we understand the mechanics of thought, AI is only a dream, held back by human failing. Until we no longer have this shortcoming, we cannot create an autonomous machine, Star Trek notwithstanding.