Re: The engine of intelligence. (was, Loosemore's Collected Writings on SL4 - Part 6)

From: William Pearson (wil.pearson@gmail.com)
Date: Thu Aug 31 2006 - 05:17:11 MDT


On 29/08/06, Heartland <velvethum@hotmail.com> wrote:
> > We know AI is possible because natural intelligence exists. So what are the
> > features of natural intelligence you would need to extract and what can be
> > replaced with something nature could not (or did not) evolved like a rotating
> > prop?
> >
> > Keith Henson
>
> I'm always disappointed when important questions like this are being ignored,
> especially by those who claim to know the answer. It's a single most important
> question in AI, still, after 50 years from the birth of the field.
>
> What is intelligence, exactly? What functions are necessary and sufficient for
> creation of a thought? What is a mechanism of thought?

I'll have a brief shot at the first question. The second two I shall
leave well alone for the while. At its most basic an intelligence is
something that transforms its input channels into an output and also
extracts information from the input about how to change its state to
better transform the I to the O and extract information to change the
state. The better-ness implies some form of goal or purpose. It is a
problematic concept, but not too vital for seeing what intelligent
system actually does. Better can mean anything from less processing
power or energy used for the transformation, to creating a more
accurate transformation or just a quicker one.

There are many ways to extract information from the input stream from
collecting statistics, to parsing the input linguistically. All depend
upon making some assumptions about the world. The latter is the
fastest way of getting bits of information, assuming a friendly
communicator, as you can saturate your input stream with information
about how to change your state.

Of course my definition is over broad for human type intelligences. To
get anything like thinking or thought, the assumption about how to
extract information from the environment would need to be something
like: a, potentially unfriendly, source of linguistic information
about how to change your state. In that case you have check new
linguistic information against previous information gained from other
sources, internal consistency of the new information and whether the
state changes suggested will actually help your goals.

Also necessary would be other means of extracting information, to
cross check against the linguistic and also to fill in the holes where
the linguistic information is not available. For example the
linguistic information in how to create an AI, is not available yet ;)

 Will Pearson



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:57 MDT