Like so many other terms relevant to AGI, "emergence" has a lot of
different meanings.
Some have used a very strong interpretation that I don't like much... a
meaning like "a property of a collective that is fundamentally
unpredictable based on the components"
According to my interpretation, the attractors in a Hopfield net are
emergent properties of the interactions of the neurons ... but this
doesn't mean it's impossible to predict the attractors that will arise
if one knows about the neurons and their interactions. It just means
that the details of the attractors are **computationally hard** to
predict from the details of the neurons. But the qualitative nature of
the attractors can be understood cleanly by mathematical theory, in this
case.
So in general I will call a property of a collective **emergent** if it
is relatively simple to describe on its own, but computationally very
difficult to predict, in detail, from properties of the components of
the collective.
According to the above definition, it is quite possible to engineer
systems with emergent properties, and to prove things about the
constraints on emergent system properties as well.
-- Ben G
Russell Wallace wrote:
On 3/19/07, *Ben Goertzel* <[EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>> wrote:
Minsky is not big on emergence
This is an interesting point.
I'm not big on emergence, not in artificial systems anyway. It
produced us, sure, but that's one planet with intelligence out of a
zillion universes without it. Emergence is what you get when you
backcast from sentience to the shortest program that produces it; put
more poetically, it's what God uses when His limiting resource is not
time or manpower but improbability.
That doesn't make it a good tool for human engineers, for whom
improbability is no big deal but time and manpower definitely are.
Emergence, after all, basically means you couldn't/didn't predict the
results from the setup; and when a machine does something unpredicted,
it's generally called a bug. (When you bring humans into the equation
it's different - blogs, for example, could be called an emergent
result of the Internet - but then, humans aren't engineered systems,
and we don't look for emergent behavior within blog-serving software.)
Obviously you disagree with this perspective, and I'm wondering if
that's a significant axis for classifying approaches to AGI.
------------------------------------------------------------------------
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?list_id=303
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?list_id=303