[ExI] Some new angle about AI

scerir scerir at libero.it
Wed Dec 30 20:04:11 UTC 2009


[Stathis]
We can compute probabilistic answers, often with high certainty, where true 
randomness is evolved (eg. I predict that I won't quantum tunnel to the other 
side of the Earth), or we can use pseudorandom number generators. I don't think 
anyone has shown a situation where true random can be distinguished from 
pseudorandom, but even if that should be a stumbling block in simulating a 
brain, it would be possible to bypass it by including a true random source, 
such as radioactive decay, in the machine.

#

To my knowledge there are:
-pseudo-randomness, which is computable and deterministic; specific softwares 
are the sources.
-quantum randomness, which is uncomputable (not by definition, but because of 
theorems; no Turing machine can enumerate an infinity of correct bits of the 
sequence produced by a quantum device); there are several sources (radioactive 
decays; arrival times; beam splitters; metastable states decay; etc.)
-algorithmic randomness, which is uncomputable (I would say by definition).

To my knowledge nobody knows if quantum randomness is also algorithmic 
randomness. But there are problems here, because there is more than a single 
quantum randomness, depending on the specific quantum source used to create the 
string. So, it is possible there is a class of quantum randomness, and a class 
of algorithmic randomness. It is possible to say that whether quantum 
randomness satisfies the requirements of algorithmic randomness is uncertain, 
at least.

Comparing (using Borel normality test or other tests) finite (but very long) 
strings produced by pseudo-randomness sources and finite (but very long) 
strings produced by quantum devices, it seems there are no differences FAPP. In 
theory there should be differences in case of infinite strings.









More information about the extropy-chat mailing list