Pei,
I have a few thoughts about your paper.
Your classification scheme for different types of intelligence
definition seems to require that the concepts of "percepts," "actions"
and "states" be objectively measurable or identifiable in some way.
I see this as a problem, because the concept of a "percept" (say) may
well be so intimately connected with what intelligent systems do that we
could never say what counts as a percept without making reference to an
intelligent system.
For example, would it count if an intelligent human "perceived" an
example of a very abstract concept like (e.g.) "hegemony"? Would
[hegemony] be a percept, or would you only allow primitive percepts that
are directly picked up by an intelligence, like oriented line segments?
More precisely, I think that "percepts" like [hegemony] are indeed bona
fide percepts, but they are very unlikely to be defined without making
reference to the systems that developed the concept. So [hegemony] does
not have a closed-form definition and the best we can do is to say that
among a large population of human intelligences, there is a point in
concept-space that is given the word-label "hegemony" but if you were to
look inside each individual mind you would find that the same name is
actually a unique cluster of connections to other concepts (each of
which, in turn, has its own subtle differences among all the different
individuals).
The same story can be told about "actions", and internal "states".
But if there is only a loose correspondence (across individuals) between
terms labelled with the same name, then how can we even begin to think
that the act of comparing states, percepts and actions between computers
and humas (as you do in your paper) would be a good way to dissect the
different meanings of "intelligence"?
The only way out of this problem would be to define some normative
central tendency of the Ps, Ss and As across the population of actual
intelligent agents (i.e. human minds, at this time in history) and then
move on from there. But of course, that would be tantamount to
declaring that intelligence is basically defined by what human minds do.
My point here is not to ask questions about how percepts map onto one
another (across individuals), but to say that the very question of which
things count as percepts cannot be answered without looking at the
chunks that have actually been formed by human beings. To put this in
stark relief: suppose we came across an alien intelligence that did not
use oriented line segments at a very low level of its visual system, but
instead used pairs of blobs, separated by different distances across the
retina. Eveything it perceived above this level would then be various
abstractions of that basic visual building block. This alien mind would
be carving nature along different joints - parsing it very differently -
and it might well be that it simply never constructs high level concepts
that map onto our own concepts, so the two sets of percepts (ours and
theirs) are just not comparable. They might never "perceive" an example
of hegemony, and we might never be able to "perceive" an instance of one
of their concepts either. How would we then - even in principle - start
talking about whether the "same" percepts give rise to the same iternal
states in the two systems? The "percepts" would depend too much on the
actual structure of the two different minds. The 'percepts" would not
be objective things.
I see no way out of this, because I cannot see any way that the abstract
notion of *objective* percepts, states and actions can be justified.
The validity of any proposed objective scheme can be challenged, and so
long as it can be challenged, the notion of percepts, states and actions
cannot be used as a starting point for a discussion of what
"intelligence" actually is.
What do you think?
Richard Loosemore
-----
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/?member_id=8660244&id_secret=85650209-afcd20
