[ExI] Automated black-box-based system design of unsupervised hyperintelligent learning systems

[Mike Dougherty]

On Mon, Sep 19, 2011 at 7:53 PM, Kelly Anderson <kellycoinguy at gmail.com> wrote:
> Do you have something "intelligent" to say about it now that you've
> had a chance to reconsider your position? Since more intelligent life
> has been spontaneously appearing for the last billion years or so on
> this planet, it seems like a tautology that something more intelligent
> is around the corner. Since we're past genes, and on to memes and
> temes, it seems likely to me that we will have something to do with
> the creation of this "more intelligent" life...

Nope.  I have nothing "intelligent" to add.

In the scope of observing the last billion years (or so), what
interval is considered 'spontaneously'?  10,000 years?  By what
mechanism is intelligence appearing?  mutation or engineering?

"seems like a tautology" is hardly a rigorous proof.  Pointing at a
progression of genes/memes/temes is also not an inherently inductive
process leading directly to the conclusion "... and then the next
thing."

I agree we'll be involved in the next phase of intelligence on earth -
I wasn't trying to suggest otherwise.  I am doubtful that an engineer
is going to be able to draw literal blueprints for building
intelligence that exceeds his or her own.  What I am suggesting should
be a definition of intelligence is not the ability to follow
directions really quickly and accurately.  For example, mechanical
automatons are not intelligent.  Nor are music-box dancers,
variable-settings toasters, thermostats or calculators.  I think
Intelligence must be something more than imitation and repetition.  Is
a parrot intelligent because it can mimic the sounds you make?  Is it
intelligent because it learned how you call the dog by name?

Another example:
http://books.google.com/books?id=OAJW7JA24A8C&q=food-washing#v=onepage
Maybe potato washing was discovered by accident or perhaps invented
through some need.  The application of the principle of food washing
to separate rice from sand is the kind of intelligence I am suggesting
is not available to a calculator.  I do think computers will be able
to emulate this process.  I doubt that it will be built.  I expect
that it will be trained.  Neural Nets can do amazing things - they
have to be trained.  Genetic Algorithms can scour a multidimensional
solution space; evolution takes time.  Can we assemble these (and
other) pieces according to some plan?  Sure, and we Frankenstein's
monster is a similar collection of parts stuck together.  I expect
growing up Frankenstein's baby will ultimately be a more
elegant/scalable solution than starting from monster parts and
smoothing out the ugly bits.

I don't have working code as proof of anything.  Obviously my opinion
is only that.  As long as the few thousand people in a world of 7
billion are actively thinking about and discussing these ideas, I
consider my small contribution (even if it's wrong) to be
participation.  That may be the best I can do for now.