Re[2]: Preventing AI Breakout [was Genetics, nannotechnology, and , programming]

Matt Gingell (mjg223@is7.nyu.edu)
Wed, 27 Oct 1999 01:19:27 -0400

>> The assertion that qualia are not computable is totally meaningless.
>> Compare against the assertion that baseballs are non-computable or
>> that qualia are NP-complete. Do you have any idea what you're talking about?

>
> Yes, I do. Baseballs are non-computable, although the *important* part
> of their behavior is easily computable. Baseballs are quantum-random,
> but not visibly. NP-completeness is a predicate that describes
> problems, not algorithms, so it can't apply to qualia.

Ok, I've taken a deep breath. Let's try again.

Computability refers to functions - mappings from one set onto another, mappings from input tapes to halted tapes - not baseballs, qualia, or any other non-abstract phenomenon. The problem I have is that you're extending the concept to another domain without being clear what you're talking about, then refusing to explain when queried.

It's unclear what you mean when you refer to the computability of qualia. Do you mean that there is no algorithmic mapping from a symbolic description of the objective universe to a description of subjective experience? Do you mean, as you seem to have implied, that qualia can not exist in a universe who's behavior can be modeled completely by a Turing-computable formal system? Do you mean that qualia have properties and behaviors of their own which can not be described rigorously. Are you saying that a purely computation model of the mind can not account for the existence of qualia? Or do you believe, as you seem to have implied, that Turing/Church is simply incorrect and there exists some more powerful scheme for an abstract computer?

These aren't rhetorical question - I don't know what you mean - and given your refusal to elaborate further the statement is meaningless. I'm interested if you have anything to say on any of these topics, but I'm not going to interpolate between isolated pearls of obfuscated wisdom. If you don't want to go into it then we can let it drop.

Anyway, I apologize for the tone of my previous message - I'm not interested in getting into a pissing contest with you - but perhaps you can compute from whence my subjective experience of irritation arose.

> I might point also point out that truly random processes, strictly
> speaking, require a minor extension of Turing computability; it's just
> that the qualitative behavior of a random process can be simulated by
> pseudo-random processes (or, for finite processes, hidden variables in
> the initial state). The assertion that baseballs are non-computable is
> not only meaningful, it is trivially true. Frankly, your assertion that
> the noncomputability of a process is "meaningless" is so odd that I'm
> starting to question your own understanding of Turing computability.

Yes - Turing machines are totally deterministic. I'm not sure if adding a random number generator extends the set of problems you can solve vs. an arbitrarily good distribution of pseudo-random numbers. It seems like it should, but I can't think of a concrete example off the top of my head.

The baseball example, as is clear from above, wasn't referring to behavior. It was an analogy. Baseballs aren't computable - that statement has no meaning. Their weight may be computable, their color under different levels of illumination may be computable, their wind resistance at a certain velocity may be computable, and so forth. Until you specify want you want to talk about, computability is inapplicable.

But I don't think the behavior is necessarily uncomputable. In the absence of an observer, quantum physics can predict it's behavior perfectly. It's only when we step in and collapse the wave function that there's any uncertainty. It becomes an issue of frame of reference.

You can certainly come up with other problems where a result is computable only in some frames of reference, there's no need to assume a non-deterministic universe. (or get into wacky quantum philosophy stuff.)

>> If you don't want to go into it then provide a direct reference.

>
> No.

Ok. <looks down, checks a box on funding request form.>

>> The above is just juvenile.

>
> Nyah nyah.

Oh dear lord... 'I know you are but what am I' flashback.

I'm reminded of the time I Godelized my little brother's Duplo AI and he disassembled all my Halloween candy in retaliation.

-matt