anon (anon.delete@this.ymous.org) on October 11, 2020 1:07 pm wrote:
[...]
> What difference do you make between "unseen before" and "unknown"?

I am not sure if I can come up on the spot with a proper definition, but maybe two examples can illustrate the difference (and a third example will illustrate the difficulty of defining things).

1. The stream prefetcher mentioned before requires very little information: a current address and a stride. This information completely determines the next and all future addresses that will be (pre-)fetched. So those addresses are known, but still unseen by, say, some physical probe on some physical address bus. The pattern is easy to predict, and there are practical computational shortcuts to find, say, the seventh address from now (without needing to compute the six addresses inbetween).

2. Unknowable, in contrast, are chaotic things like Newton's three body problem, or the double pendulum, or the weather. One can argue that the information content here is infinite, as real numbers describing the system's state require unlimited precision. The future configuration of such a system, beyond a brief window of short term immediate future, is unknown.

3. The grey area is something like a picture of the Mandelbrot set. You give the computer a finite set of parameters and let it churn out pixels (using finite precision). The final picture is entirely determined by the parameters. However, certain parts of the resulting picture cannot be guessed beforehand, other than actually performing the full chain of computations. There are no general shortcuts, there is no method of prediction that could save work on all pixels. Many clever optimizations have been devised, and many pixels can be accurately generated with less effort. But not all. So are these problematic pixels known or unknown beforehand? They are fully determined from the start. On the other hand they are not predictable (in the sense of saving effort to obtain them, one cannot "skip ahead" to them).