[ExI] Trilemma of Consciousness

[Stuart LaForge]

John Clark wrote:

​> And after reading Darwin the obvious question to ask is "Evolution can't
> directly detect consciousness any better than I can and yet I know for a
> fact it managed to produce it at least once, how did it do that?". The only
> answer is that consciousness is a unavoidable byproduct of intelligent
> behavior.

I am certain that consciousness and intelligence are correlated. After all
intelligence is a semantic property as well. But they might be seperable.
Like the stroke victim who is aware of everything going on around him but
lacks the motor control with which to express "intelligent behavior".

If one can be conscious but unable to behave intelligently (like the
aforementioned stroke victim), I don't think that it is impossible that
something could behave intelligently and not be conscious. Especially if
it were purposefully designed that way.

> I think "consciousness is the way data feels like when it is
> being processed" is a brute fact so it is pointless to ask how or why.

If you think that fact holds down to the level of a single bit flipping
its value, then you would belong to the "consciousness is trivial" 
school.

> The Turing Test has nothing to do with Turing Machines, the test is
> ​ ​
> agnostic as to how the subject manages to produce the observed behavior,
> ​ ​
> it's irrelevant.

Indeed, that is the way I first envisioned it using Russell's paradox.
That would have been a stronger result that would have applied to all
beings and not just Turing machines. But thank Zermello and Frankel for
screwing up set theory so that Russell's paradox is now off limits and I
am stuck with a result restricted to Turing machines.

> I don't understand, are you saying the
> historical Turing test
> will work for consciousness as well as intelligence, or are you proposing
> some new test that could distinguish between a
​> intelligent conscious being and a intelligent
> non-conscious being?

No, I am saying that the Turing test won't be conclusive for either
intelligence or consciousness with regard to Turing machines unless those
are all or nothing properties of Turing machines. In other words, you run
statistical analysis on the AI's behavior for consciousness, intelligence,
or both and hope your AI doesn't get caught in an infinite loop.


If so then Evolution must have used that test too, but
> then it must be based on behavior because
​> behavior is​
> what improves survival chances not consciousness, but if it's based on
> behavior then it's just the standard vanilla
> Turing Test.

What I am saying is that due to Rice's Theorem, any Turing-like test for
any semantic property (consciousness, intelligence, etc.) applied to a
Turing machine is reducible to the halting problem unless Turing machines
all share that property or are unable to possess that property. Meaning
that all you can rely upon is faith or behaviorial statistics and hope.

> Consciousness theories are a dime a dozen because unlike intelligence
> theories there is no way to prove or disprove any of them, so I have no
> doubt one of those theories could be used to make a consciousness test
>fine tuned to make sure humans passed it but computers didn't (such as
>consciousness theory#93,642: conscious beings must be squishy)

Well people who think that belong to the "consciousness is null" school.
Unfortunately my theorem does not rule that out either.
​
Stuart LaForge
 ​