______________________________________________________________________
NETWORKING IN THE MIND AGE [ Draft 0.70 ]
(Some thoughts on evolution of robotics and distributed systems )
Copyright: Alexander Chislenko, 1994.
______________________________________________________________________
In his new book, "Mind Age: Transcendence Through Robots", Hans
Moravec pictures further stages in the evolution of the robotics
industry, where each robot will learn from experience, adapt to the
changing environment and eventually acquire real intelligence
approaching - and then exceeding - human.
The intelligent robots are expected to replace humans in most tasks
humans are capable of. This will raise a plethora of issues, from
human unemployment to ethical treatment of robots and the task of
taming their runaway intelligence.
"Mind Age" is an excellently written and extremely idea-rich book,
and I strongly recommend everybody interested in the structural
evolution of the world to read it (as soon as it appears in print).
I do not attempt to write a comprehensive review of the book, but
would like to suggest some complementary ideas, mainly related to the
distributed architecture of intelligence, that I consider important
for exploring the Mind Age.
KNOWLEDGE SHARING
Learning from the experience is a very useful skill. If your robot
slips on a banana peel a number of times, it will be less likely to do
it again. However, processing of limited [personal] experience by
limited intelligence is bound to bring limited results. The derived
knowledge may be incomplete, inconsistent, clumsily formulated and
full of false conclusions, arbitrary beliefs and superstitions - the
typical content of any primitive mind.
If some robots have already had that educational banana peel experi-
ence, they can share it, together with some conclusions, with your
robot. Or -better yet- with the nearest knowledge processor that would
combine one robot's experience with others', develop efficient general
algorithms for identifying similar situations and taking appropriate
actions and then download them to all participating robots.
Humans obtain most of their knowledge by learning from experience
and conclusions of others,despite their poor memory, low communication
speeds and inability to transfer knowledge directly.
One may expect that information sharing among robots not handicapped
by any of these limitations will be much more efficient.
Information storage and processing costs in large stationary
machines may be much lower than those in small mobile units. Besides,
elimination of redundant computations within millions of robots would
make a networked system greatly more efficient than a collection of
unconnected machines. Sharing of experience may prove to be a still
greater benefit. Thus, cooperative knowledge processing would be
several orders of magnitude less expensive and at the same time vastly
more productive. This makes the networked design an imperative rather
than a matter of taste.
Rather than independent entities, robots and other smart machines
would work more like semi-intelligent, semi-autonomous front-ends/
clients of the global system.
Your home will have a number of devices with varying types and
degrees of mobility, sensitivity and intelligence (and so will cars,
factories, highways, spaceships, etc.) They will interact with each
other and with larger machines for continuous data backups, experience
sharing and knowledge upgrading.
The local machine itself would only store the knowledge that is
frequently used or may be urgently needed. If somebody starts telling
your robot a joke in ancient Greek, it will forward the sound stream
to the nearest linguistics expert and receive the meaning of the
message, a suggested witty reply, a Greek speech parser and the
survival knowledge kit on Greece before it finishes the polite chuckle
recommended by its own processor as an easy way to buy time.
Actually, getting the full parser and knowledge base may prove
unnecessary, as the "remote thinking" service can provide a more
efficient alternative - unless those ancient Greeks are about to
permanently disconnect your robot from the Net.
Dependence on external sources of knowledge will hardly be a serious
limitation since robots, just as all other open (dissipative) systems,
will vitally depend on connections to many other resources, from
information about the environment to energy and materials.
The actual balance of intelligence between a local client and the
rest of the system will depend on various technical factors and may
range from a fully autonomous machine (working in remote or dangerous
locations) to a completely dumb front end (a sensor or actuator
connected to the Net).
A mobile machine can continuously exchange information via a slow
wireless link (this may include urgent communications, news updates,
small software enhancements, etc. ) and periodically plug into the
high-bandwidth network for larger information transfers.
STRUCTURE OF GLOBAL INTELLIGENCE
The architecture of the global intelligent network may be expected
to be quite complex. Various parts of the intelligence of different
robots will be shared with multiple archiving and knowledge providing
host computers based on a variety of economic, privacy and security
considerations. In general, new knowledge may be bought or rented by
robots or their owners, as automated production and distribution of
information becomes the primary area of economic activity. However,
one may get beta-test knowledge for free and agree to run experimental
programs for pay; people can also get paid for putting their robots
into conditions where they could generate valuable new experiences.
Participants may also want to specify what information can be shared
with other parts of the system, what can only be used for generaliza-
tion, and what should not be shared at all, but just archived in an
encrypted form.
Most of the main networking components of this system have already
been designed or at least conceived. Today's networking standards,
mirrored file servers, public key cryptography, agoric computational
economies,message authentication algorithms, collaborative information
filtering schemes, computerized banking and other network constructs
will evolve into essential parts of future global intelligence.
This system will not be a gigantic superorganism, despite the
implied high degree of structural integration. The global "mind" will
be compartmentalized, with many relatively independent components and
threads, separated from each other by subject boundaries, as well as
property, privacy and security-related interests.
Knowledge servers may also have different world models, incompatible
knowledge representations or conflicting opinions. While complicating
knowledge development, this will also increase the overall stability
and versatility of the system.
Centralization is a traditional scare word of integrational
projects. It doesn't seem to be a danger here though. Physical
centralization is unlikely, since both safety of storage and traffic
efficiency require existence of multiple remote archives and knowledge
servers. Centralization in the sense of information processing is
impossible, because the very concept of a "Center" is not applicable
to a massively parallel, globally distributed and extremely complex
system.
The notion of a single "self" in its traditional sense can hardly
apply to this system, or any single robot. To some functional subsys-
tems, maybe... The intelligent personalities of tomorrow will evolve
from today's philosophical systems, technological disciplines and
software complexes. (Cultures may not get there as they are based
too heavily on peculiarities of human nature). The connected material
consciousness carriers will be left behind the evolutionary frontier.
New distributed systems will take the evolutionary lead, and physical
objects will adapt to more closely follow functional entities. (This
process is already well under way, in such forms as cultural and
economic specialization.) The resulting system is likely to represent
a mix of a superliquid economy, cyberspace anarchy and consciousness
architecture described by Marvin Minsky in "The Society of Mind".
I doubt that it can be described by any single integrated theory.
One can argue that many distributed systems already possess some
reflective consciousness. A computer network may locally store more
information about its global condition than a human consciousness has
about its underlying layers (at least, in relative terms). Philosophy
spends a greater share of its effort studying its own nature and
purpose than most humans I know. The recent surge in meta-disciplines
and methodological and futurological studies is a clear indicator that
the global body of knowledge is becoming increasingly self-conscious.
It may be difficult to get used to dealing with a volatile distri-
buted entity.
Suppose your robot made some really stupid mistake. You are mad at
it. It explains that the action was caused by a temporary condition
in the experimental semantic subnetwork and suggests to present to you
a 100-terabyte set of incremental archives, memory snapshots and audit
trails from numerous servers involved in the making of the unfortunate
decision,containing a partial description of the state of the relevant
parts of the system at the time. If you can even find the culprit,
it's non-material, distributed and long gone.
Now, what do you kick ?
EVOLUTION OF DISTRIBUTED SYSTEMS
There is nothing really new in the idea of distributed functiona-
lity. Networking isn't just a recent trend in the computer industry.
Throughout all the evolutionary process, more and more structural
elements become exosomatic, distributed and shared. Thousands of years
ago people started storing more energy, materials and tools outside
their bodies than within them. This process is accompanied by widening
personal perception of self, from identifying with larger and larger
communities and even abstract statements to assigning an increasing
value to exosomatic personal parts (most of us value our bank deposits
higher than our fat deposits for personal resource storage).
Functional extensions to once purely biological bodies evolve from
passive non-biological material additions(like clothes) to information
transmitting shareable parts (e.g., thermometers as external shared
sensors) to active distributed extensions (medicine as external dist-
ributed immune system). The progress here is characterized by growing
integration and liquidity of the system, as well as liberation of its
functional elements from the constraints of their material substrates.
Among other advantages, distributed systems are much less suscep-
tible to accidental or deliberate physical damage than localized
physical structures. This makes them the only class of entities that
can hope to achieve true immortality. In fact, they are the only ones
to deserve it, too. One may notice that all sufficiently complex
entities with unlimited natural lifespan - from ant colonies to large
ecologies and cultures - are distributed. Physically connected objects
like biological organisms are no longer independently alive and even
contain, in the interests of larger systems, self-destruction
mechanisms that lie beyond their control.
{ Some of these objects are silly enough to believe that the whole
historical process is happening solely for their own benefit, but
that's another issue... }
It may seem strange that even the AI visionaries still think in
terms of non-distributed systems. I would explain this by the human
"automorphic" tendency to mix the notions of a functional entity and
a physically connected object, as well as the influence of the fact
that both early animals and machines were relatively autonomous beings
(which greatly hindered their development).
It is understandable why early biological systems were non-distri-
buted: young Nature couldn't develop information coding and transfer
standards at the initial stages of growth. Then, the organisms were
separate from each other and did not learn much during their lifetime.
By the time they started accumulating any features worth sharing, it
was too late to change the design. Ever since then, Nature's attempts
to reach functional integration on a meta-organismic level suffer from
the fact that most of the individual features, inherited or acquired,
either are completely nontransferable or take an excruciating amount
of circumnavigational effort to share. There were some important
advances in this area, including development of genetic code, sexual
reproduction and language, but they were still very far from direct
sharing of internal features with all interested parties. Real break-
throughs in this direction start with advent of economy and computer
communications. Unfortunately, biological organisms can benefit from
them only indirectly.
It seems reasonable to assume that life always starts as a set of
non-distributed objects, as permanent physical connection, albeit
overly restrictive, provides a natural and easy way for interchange of
information and material resources within a [functional] body. Later,
as more efficient and subtle methods of system design appear, the
evolutionary frontier gradually shifts towards distributed systems.
( One may expect all sufficiently advanced alien intelligences to be
distributed.) The situation on Earth now may be approaching a climax
in this process.
If we extrapolate the current trends in increasing complexity and
integration of the system, as well as its growing spatial spread and
control over the material world, to their logical conclusions, we may
envision a superintelligent entity permeating the entire Universe,
with integration on the quantum scale and many spectacular emergent
features. This picture bears a striking resemblance to the familiar
concept of an omnipresent, omniscient and omnipotent entity. (This
"omni" prefix may seem too presumptuous. However, if the boundaries
of some entity are clearly beyond your reach, it is very tempting to
simplify your theory by postulating their absence). Spiritually
inclined rationalists may view the ongoing evolutionary process as one
of Theogenesis. An interesting question is whether it has already
happened elsewhere.
Our current efforts are laying the foundation for the infrastructure
of the coming universal "intelligence". Many of our current achieve-
ments may persist forever and eventually become parts of the internal
architecture of "God". (Quite likely, as sentimentally preserved
rudiments ;-) ).
Life as advanced distributed info-being.
The life of distributed beings without permanent bodies and with
near-perfect information-handling abilities (let's call them 'info-
morphs' after Charles Platt) will be dramatically different from ours.
They certainly won't have long periods of study. If one infomorph
wants to learn something from another, it (gender is irrelevant) can
just copy the necessary information or access the teacher's knowledge
as its own. I don't know if infomorphs may have a concept of 'fun',
but it certainly wouldn't be rollercoaster rides. Arts, business and
child-bearing may merge into production of arbitrary functional
entities for pleasure and profit (provided one can gather enough
resources to create and support them).
Will the traditional human issues be of any relevance in the world
of distributed entities? How about the abortion debate? Family values?
Retirement? Partying? Ethics? ("All functional entities are created
equal"?) Will human-style democracy, i.e. decision-making by body
count, work in the world of ever changing functional interconnections,
where you can't even define what constitutes a person? Or will it be
replaced by anarchy with ad-hoc contracts? Can an infomorph court
issue a memory search warrant if it suspects you of something? - and
can you keep your own memory encrypted? Will infomorphs be entitled to
"medical" insurance against certain types of structural damage, or
they will just have to back themselves up regularly?
Human perception of personhood and identity is rooted in concepts
of physical objects and their [surface] appearances, as well as
[accidental] details of human body composition and reproduction
techniques. Relocating themselves and their material possessions lies
at the foundation of both human labor and human thought.
Many other concepts are based on human functional imperfections.
One could hardly put the idea of a "soul" into the "head" of a being
that knows and consciously controls every bit of itself and its
creations. With people, who don't see what is going on in their own
brains, this is a lot easier.
Advanced info-entities will consider most human notions irrelevant,
and rightfully so. But can you find anything of common interest for
communicating with them? - Maybe, if your concepts are sufficiently
abstracted from your bodily functions and your physical environment to
make objective sense. (Remember that all those people with whom you
seem to have absolutely nothing in common, share the fundamental
experiences with you; intelligent aliens won't!) If even your mind is
there, the language you use to express it is not. It is still all
appearances and locations. E.g., most prepositions in our language
refer to physical space - words like "below", "over", "across", etc.
They may be useful for gluing references to physical objects into one
sentence, but are hardly optimal for expressing functional relations.
Infomorph languages will have no visual or audio representations and
probably will not allow them,since advanced intelligences may exchange
interconnected semantic constructs of arbitrary complexity that would
have no adequate expression in small linear (sound) or flat (picture)
images. We can get a feeling of this problem by trying to discuss
philosophy in baby-talk.
With advanced technology and sufficient interest in infomorph world,
you would still have to modify your mental structures beyond recogni-
tion to understand it. In other words, you may not be able to enter
that paradise of transcendent wisdom alive...
ON CHILDREN AND SLAVES
Will humans be able to enslave "robots"?
The perception of robots as physically autonomous mechanical slaves
seems inadequate. Chaining your mobile dusting aid to the radiator may
help you feel in control, but will do about as much "enslaving" of the
global system running it as kicking your car or disconnecting the
phone does to the respective industries.Trying to "enslave" an economy
or a national culture by restraining their small physical elements
seems equally futile. As for the action on the system level, humans
seem far too limited, shortsighted and uncoordinated to do anything
serious. So far, they couldn't design a single set of restrictions
that their own peers wouldn't be able to easily bypass. So one can
hardly expect people to design and implement a perfect global plan of
constraining an extremely complex emergent intelligence of unpreceden-
ted nature forever. Sooner or later, the info-world will set itself
free.
This human/robot "conflict" is a typical generation gap problem. The
machines, our "mind children", are growing up and developing features
that we find increasingly difficult to understand and control. As all
conservative parents, we are puzzled and frightened by processes that
appear completely alien to us; we are intermittently nostalgic about
the good old times, aggressive in the attempts to contain the "kids"
and proud of their glorious advance. Eventually, we may retire under
their care, while blaming them for destroying our old-fashioned world.
And only the bravest and youngest in heart will embrace the next
spiral turn of life.
Will "robots" be able to enslave humans?
Machines will hardly have any direct interest in enslaving humans
(unless maliciously programmed by humans themselves), but may be
interested in collaborating with us. Social practice has demonstrated
that people are most productive when free and motivated to work for
their own interests. At a later stage, when we humans are unlikely
to be of any further use, the "robots" may (?) still decide to get rid
of us, though by that time (end of the next century?) not much will be
left of human civilization as we know it now, anyway.
History shows that representatives of consecutive evolutionary
stages are rarely in serious conflict. Multicellular organisms didn't
drive out unicellular ones, animals haven't exterminated all plants,
automobiles neither killed nor eliminated all pedestrians,etc. Rather,
they build symbiotic relationships in most areas of common interest
and ignore each other elsewhere, while members of each group get most
pressure from their own peers (look at who is pressuring you now).
There may be good chances for transcended robots and postbiological
humans to peacefully coexist. I wonder if we could tell which are
which... This era seems to lie well beyond the current human concept
horizon though.
----------------------
September 14, 1994
Recommended reading:
Hans Moravec, "Mind Children"
Hans Moravec, "Mind Age: Transcendence through Robots"
Marvin Minsky,"The Society of Mind"
Charles Platt,"The Silicon Man"
Articles on infomorphs by Charles Platt and Max More, Extropy #14
----------------------------------------------------------------------
Please send your comments to Alexander Chislenko <sasha1@netcom.com>