Eliezer S. Yudkowsky wrote:
> Anders Sandberg wrote:
> > that this process has a time constant shorter than days
> (why just this
> > figure? why not milliseconds or centuries?),
> Actually, I think milliseconds. I just say "hours or days" because
> otherwise my argument gets tagged as hyperbole by anyone who lives on
> the human timescale.
> Centuries is equally plausible. It's called a "bottleneck".
A quibble. If the first successful seed AI gets a x100 speed boost from fast hardware, and its programming is another two orders of magnitude faster than human due to the AI advantage, it still needs almost an hour to do a 1 man-year programming job. Since major architectural changes are likely to run to several man-years of effort at that point, the next few iterations of the enhancement cycle should still be slow enough to watch.
Depending on how hard the software problem turns out to be, its actual speed boost could plausibly be anything from x10^-3 to x10^9 or so. Of course, my bet is on a late takeoff with very fast hardware, which gives exactly the result you predict.
Billy Brown, MCSE+I