4 ms·
Form and Content in Computer Science (1970)
- ScottBurson 9y agoSection 2.2 is particularly interesting to me. Marvin said (in the last paragraph of the section), There is no practical consequence to the fact that the program-reduction problem is recursively unsolvable, in general. In any case, one would expect programs eventually to go far beyond human ability in this activity The first sentence is ridiculous as stated -- here it is 47 years later, and we still have nothing like the automated program-transformation system Marvin envisioned, and uncomputability has to be considered a major reason for that -- but nonetheless, I share his intuition that this barrier "should", in some sense, be surmountable. After all, as he goes on to point out, we humans do manage, albeit with considerable effort and many mistakes, to write programs that are at least correct and efficient enough to be usable. It seems there should be some way to characterize the mental processes by which we do that well enough that we could program a computer to do the same thing -- and even to do it better than we can, as Marvin suggests. Yet this has turned out to be one of the most difficult problems in AI. I don't even know that one could say that much progress has been made since Marvin gave this talk.