0
$\begingroup$

I have a two-fold question - theoretical and practical. Where do I ask it?

Thereotical:

Does Rice's theorem imply safe AI is undecideable?

(Meaning "safety", for any reasonable definition of safety, cannot be decided in general for Turing machines, including AI.)

This question could be rigorously proved.

If "yes":

Practical:

Given the safety of AI cannot be rigorously guaranteed (that is, for all reasonable definitions of "safe", you cannot prove AI is "safe"), why isn't there mass concern regarding AI safety?

This is the question I want to ask. Where do I ask it?

$\endgroup$
1
  • $\begingroup$ Your first question can be answered by understanding Rice's theorem. Or rather, you can then see why the question is ill-posed. Here are two older answers I wrote that might help: one, two. $\endgroup$
    – Raphael Mod
    Apr 4, 2020 at 14:45

1 Answer 1

Reset to default
2
$\begingroup$

I believe the main component of the first question is to figure out of what exactly "any reasonable definition" of a safe AI is. You will need to specify more clearly what "reasonable" means here to you when you ask the question, however. I think this question could be asked on Computer Science, but Artificial Intelligence is probably more suited to that question.

I don't really understand the second question. I cannot rigorously guarantee (at least, not with a mathematical proof) that my computer will not explode either and I don't think I should be concerned about that.

In any case, the second question seems to be an open ended discussion question that is not suitable for the Stack Exchange platform.

$\endgroup$
2
  • $\begingroup$ Thanks for the reply! One goal was to prove the statement independent of the definition of safety. That way I don't have to define it. I agree; "reasonable" should be defined. By "reasonable" I meant "non-trivial" as used in the Wikipedia article on Rice's theorem. Thank you for the recommendation to ask on AI! $\endgroup$
    – Jesus is Lord
    Apr 2, 2020 at 18:39
  • $\begingroup$ Discrete lizard♦ Regarding your computer, that seems like a false analogy to me? We don't know the axioms to the laws of physics. We do know the axioms to Turing machines (ZFC?). $\endgroup$
    – Jesus is Lord
    Apr 2, 2020 at 20:40

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .