[comp.ai.philosophy] simplicity and intelligence

jjewett@math.lsa.umich.edu (Jim Jewett) (10/12/90)

In article <3617@media-lab.MEDIA.MIT.EDU>,
minsky@media-lab.MEDIA.MIT.EDU (Marvin Minsky) writes:
|> In article <1990Oct7.192212.24550@math.lsa.umich.edu>
jjewett@math.lsa.umich.edu (Jim Jewett) writes:
|> 
|> >|> In article <3549@media-lab.MEDIA.MIT.EDU>
|> >minsky@media-lab.media.mit.edu (Marvin Minsky) writes:
|>  
|>   >>(Minsky) ...in an essay of mine --- "Communication with Alien
|>    Intelligence," (is) a cute theory based on some experiments 
|>    with very small Turing machines.  It turned out that many of them
|>    performed operations that could be interpreted as elementary
|>    addition -- while none of them did anything that was "similar" to
|>    addition but not exactly addition!  
|> 
|> >How do you define "similar" to? Do you mean any associative property 
|> >with inverses and an identity? Any accumlator?  It seems that there 
|> >will be no woman similar to Mom because we know Mom so well ... but 
|> >someone for whom she isn't the point of reference may very well see 
|> >her as similar to her sister.  
|> 
|> Consider reading the paper. By "similar" I meant commonsensical things
|> -- like could there be anything like the integers except skipping the
|> number 5?  Or could there be a number system with the symmetry laws
|> holding usually but not always.  Or integers with tree signs, instead
|> of two, etc.

OK ... I may still be missing some of the nuances, but ...

I can see that merely elimating the number 5 and setting 2+3=6,
so that 3+3=7, etc. is just a renaming.  But then you mentioned
trying to figure out a way that you *could* come up with a different
arithmetic.  

You were unable to find a number system with three signs, though
you acknowledged that you weren't able to find one with four either.
(At least not in the time devoted, without just starting from the
Complex numbers.)

And so I wonder how much of your time was, in effect, just refinding
what you had already been trained in.  You did find some systems
with only one sign, and modular arithmetic.  While you didn't find
them useful, I think that they are different ... at least as different
as a third sign.  and so I count three "simple" sets of artithmetic ...
each with different uses.  Modular arithmetic would be more useful for
a view based largely on cycles, and we don't use negative distances.
Perhaps a three-signed system would also have uses, but we haven't
looked for them yet.

The experiment that you mentioned in this group was examining simple 
Turing machines, which meant those with a small set of rules.  I still 
question their simplicity.  To you, they look simple.  They will look 
very simple to me as well ... but we do have more in common with each 
other than with members of another species.

(For those a bit lost by this point, remember that the paper was
in a book on the Search for Extraterrestrial Life, with Artificial
Intelligence assumed to similarly alien.  It was also (the version
I read) in the April '85 Byte.)

For instance, we are genetically based on DNA ... RNA would seem to
work similarly ... but not identically.  Replacing Carbon with
Silicon might also be possible.  On another planet, the 
distribution of "critical" chemicals might be different.  I
don't *think* that that should alter the rules of simplicity ...
but it might have very profound effects on which sorts of chemical
reactions occurred most frequently.

It is possible that on another world, life would use a standard
method that we consider much more difficult, because it would
be more common naturally, and therefore have more chances to 
become useful.  (Just as we have found more uses for water
than for Einsteinium, so would nature.)

On this world, for instance, most people do not understand
1's or 2's complement arithmetic.  Yet to a computer, that
may seem the most natural, or simple method.  Computers
then translate the results into the sort of answers that 
humans like.  If, however, there were no humans, what
would motivate this?  Would 2's complement develop to the
point that we could no longer comprehend it?  And would this
be so basic to any "intelligent" life form that they didn't
think to question the assumption?

As has been pointed out, (most recently by jmc@Gang-of-Four.usenet 
(John McCarthy) in <JMC.90Oct9152336@Gang-of-Four.usenet>), we
don't use the "simplest" methods (logic) ourselves in all cases.
So either complicated methods are sometimes used (and might be
by aliens as well), or we don't really understand simple.

And after all this ... I'm not even sure that the "simple" concepts
are that useful.  You mentioned several that you suspect would have 
to be universal.  Even if they are, I wonder whether they
are sufficient for intelligence, if computers aren't already 
there.

The most basic concepts were

Object-Symbols:		Things, ideas, processes (mostly nouns)
Difference-symbols:	Differences ... including changes
Cause-symbols:		Active agent - that which is responsible.
				(For us, usually a noun, typically 
				the subject of an active sentence.)
Clause-structures:	Allows embedding, so that we can analyze.
				(eg, "The car that I saw at the
				the theatre is going by.", or
				understanding psychology without
				reference to neurochemistry.)

So perhaps a for a computer:

object:		data structure
differnece:	comparisons, or parameters and returns of a function.
cause:		function call  (we ourselves don't always find
			deep causes)  Or maybe the result of a test.
clause:		recursion, or even the concept of calling other
			functions.


And I'm not willing to concede that every program with those
constructs is intelligent.

-jJ 
jjewett@math.lsa.umich.edu       Take only memories.
Jewett@ub.cc.umich.edu           Leave not even footprints.