Re: Some thoughts on multi-agent systems and "hyper-economy"

Mark Crosby (
Mon, 24 Nov 1997 13:40:43 -0800 (PST)

On Fri, 21 Nov 1997 01:26:32 -0500 Alexander 'Sasha'
Chislenko wrote:
< I have been interested in multi-agent complex
systems lately, and though I still have to do a lot
more reading and thinking, I would like to share a
few ideas; maybe you'll help me figure things out. As
always, I am looking for all kinds of remarks and
references. >

Along the lines of a general theory of
representations, it's interesting to note the recent
Hayek-L discussion of Barry Smith's "The
Connectionist Mind: A Study of Hayekian Psychology",
particularly the distinction between propositional
(pre-defined) and non-propositional
(dynamically-derived) knowledge:
"The whole resultant process is then not so much a
matter of 'classification' as of 'evaluation' ...
making distinctions of degree as well as distinctions
of kind ... The map's job, from Hayek's perspective,
is to facilitate a certain sort of cognitive laziness
(or what Mach called 'thought economy') ... not,
however that of a passive-memory storehouse, but of
an active-memory competence ... It is remarkable how
little explicit (conscious) knowledge is required by
the agent in order for him to be able to react in
appropriate ways to changes in his circumstances ...
the most essential information is passed on in the
form of abbreviated signals."

Joel Henkel's 971111 response to this essay may be of
interest for the development of effective situational
awareness representations: he notes that what we want
are linkages that are represented "in a mode that is
free from classes, signs or symbols ... a
representation in kind" that uses a connotative
rather than denotative association.
He is probably describing a theory of the human mind,
but we may need something similar for societies of
intelligent agents on the Net.

In other words, rather than try to explicitly
classify and rate millions of agents or sites, it
would be better to implicitly capture the relations
among agents in some semi-anonymous, statistical
manner, much as price signals do in the material
economy. In short, the trick is to find some
self-organizing mechanisms based on ubiquitous signal
exchanges rather than rely on explicit rating schemes
(although such could certainly be supplementary).

Another reference I recommend (for theory) is Robert
Pallbo's "Mind As Evolution and Evolution As Such"
) for its generic description of knowledge
acquisition systems in terms of active components vs.
preservative components, and retention vs. revival
mechanisms, while defining the ways that variation
and selection can be applied to such systems.

While I'm still doing references, Mitchell Porter
wrote about James G. Miller's Living Systems Theory.
There is a paper by Chen and Gaines at
called "A CyberOrganism Model for Awareness in
Collaborative Communities on the Internet" that uses
Miller's theory as a foundation. They have lots of
other related papers one step up at

Next, Sasha wrote:
< Multi-agent systems and action spaces >

Warning: the rest of this is just
off-the-top-of-my-head stuff that may or may not be
addressed in 'the literature' as I haven't done much
research in this area.

Is there a way to publish ones 'action space', e.g.,
web browsing history, in an anonymous way but such
that primitive net agents could access it, whenever a
relevant search or send was initiated? These same
agencies must be able to accept feedback.

< I would envision the future intelligence as a
collection of highly specialized tools, with
high-bandwidth interconnections, with
"personalities", or complex problem solvers
("achievers", "goal engines", "research tools",
"development threads") consisting of limited
contractual relations between multiple
problem-solving, planning, perceiving, and acting
entities, that can at the same time be employed in
many other similar relations. So the entities may use
the same "body parts" on a cooperative or
time-sharing basis. >

To me it seems that 'personalities' are the hardest
part here. Perhaps this could be partially
implemented by tracking my browsing history; better
yet, a diary of my email filtering (saves vs. reads
vs. skips) through some sort of semi-anonymous
'signaling' that would enable personality indices and
pointers to be prepared. The trick, of course, is to
find a way to implement this without compromising
privacy too much.

Sasha again:
< Unfortunately, while the economy provides its
agents with accurate estimates of generic values, it
has little to offer to each individual agent in their
attempts to figure the value of a certain product to
their particular needs, or offer any other
personalized or situational advice. >

This is because each agent's utility functions are
usually known only to that agent (and then only in
some amorphous, embedded analog manner). However, if
my customized search agent knows my interests
(according to some standardized classification
system) and can report a history of my demands to
other service agents, and all demands could be
aggregated by some set of interest indices, then
prospect-mining entities could queue up pointers for
my type of agent (e.g., wants info on memetics), when
I initiate a search, such agents could immediately
recommend various threads (sets of URLs) and rate
them according to my responses (at the most basic,
whether I follow them or not).

< Among the first applications of this approach I
envision large automated collaborative filtering
projects, link exchange programs and targeted
advertising, balancing flows of information in
accordance with diverse and dynamic interests of
creators, consumers, providers, and sponsors of the
Web content. >

My favorite application would be a history of a
browsing session - time-stamped, each link rated by
duration of review, the extent of excerpts
downloaded, any subjective evaluations explictly
entered, automatically stored as a thread
'somewhere'. This seems to be a logical extension of
the more 'unconnected' site rating that is (was?)
available, e.g., with Firefly on Yahoo! I tried this
about a year ago and obtained a few good links but
found that the biggest deficiency, aside from the
effort required to rate a bunch of sites to get
started, was that one could only rate sites that were
in the database and there was no provision to add new
sites on the fly.

If the entities doing the browsing had already
submitted some interest profile, then the recorded
browsing thread could be classified and retrieved by
similar agents when they commenced a related search.
But, ideally we want a way to create the interest
profile without any explicit effort on the inquiring
agent's part.

A payment mechanism is probably essential as an
implicit rating system. Once established, this
'point of sale' data, tracking knowledge acquisition,
could be fed into various indices of interest and
intent that would drive the development of additional
sources and services.

Of course, until it becomes acceptable to charge
(some small amount) for such value-added search
services, and there is a secure, unobtrusive
crediting mechanism in place, it is doubtful that
resources to maintain such a service would ever be
allocated. I DON'T want to buy somebody's service
'whole-hog' but I would gladly pay (if there's some
kind of easy, one-click payment procedure) a few
cents here or a dollar there for value-added hints
and recommendations.

Mark Crosby

P.S. I love the 'What's New' page on your site. One
of my biggest complaints about so many Web page
providers is that they never think to date/time-stamp
their entries. Situational awareness in
collaborative searches is exponentially enhanced by
the availability of such chronological logs. For
example, a quick glance at it tells me that you just
added the Hayek-L list to your Mailing Lists page, so
I know you're probably aware of the "Connectionist
Mind" essay I mentioned in the beginning.

It's probably too much too ask that everyone take
such care with their web sites or publish their
'itinerary' for explicit viewing; but, it would be
nice if anonymous delivery agents were able to track
such information and somehow route information

Sent by Yahoo! Mail. Get your free e-mail at