|
Exploring the Technological Singularity:
Seeking the Universal
Drivers of
Accelerating Change
Preface
John Smart is a leading researcher in
the nascent field of singularity studies, the study of models of exponentiating and asymptotic accelerating
change, from both local and universal perspectives. He
is also chairman of a nonprofit organization, Accelerating.org,
the Foundation for Research in Accelerating Change (FRAC). FRAC is a Web community, conference host and reading group network
that explores universal and local drivers of accelerating
change, and connects lay and professional scholars with
interest in these topics. A professional futurist, he
is passionate about generating greater public awareness
of historically important trends so our society may be
better equipped to act wisely and proactively in what
promises to be the most challenging and exciting chapter
in human history. He is presently writing his first book,
"Destiny of Species," which explores potential
universal mechanisms and the apparent near-term trajectory
of accelerating physical and computational change.
This extended interview, which features
original questions edited for convenience, was conducted
via e-mail between June 2001 and February 2002 and serves
as a preview of Smart's book. It spans a broad range of
ideas and resources that may be relevant to understanding
the technological singularity, including the speculative paradigm of
the developmental singularity that attempts to place the apparently inevitable technological
singularity in a universal developmental context. A few
academic references are cited, but most are accessible
generalist works intended for an educated non-specialist
audience.
Abstract - 1
Technological singularity, the condition
of fully autonomous, human-equivalent machine intelligence,
is coming soon, by many accounts circa 2040, whether we
are ready for it or not. The transition will be a bottom-up
process of evolutionary development within the newly emerging
technological complex adaptive system, or universal computational
substrate.
Nevertheless, this process may seem to us to be largely a top-down, human-guided and human-aided activity.
Significant machine intelligence will mostly be missing
from the human environment for at least another three
decades, but will suddenly explode into high-level, fully
self-catalyzing capacity within a few short years. Most
of humanity, even in technological environments, will
probably not notice the emergence, as this intelligence
will be deeply symbiotic with human ends and largely constrained
to travel much farther down the presently understandable
trajectories of greater intelligence (simulation capacity),
immunity and ethical interdependence already incrementally
displayed by today's technology-aided societies as they
increase their own adaptive complexity. In other words,
the technological singularity will represent only a partial
cognitive singularity--only certain special elements of
the subjective state and goals of intelligent machines
will essentially be unknowable to biological minds. One
dominant feature of post-singularity society will be the
smoothness and rapidity with which remaining human problems
are solved. Another will likely be free and reversible
integration paths, allowing human intelligence to migrate,
effectively unidirectionally after a brief period of bidirectionality,
into the faster, more stable and far more flexible machine
substrate. By all present indications, technological evolutionary
development (cyberspace, nanotechnology and nanocomputation)
promises to be a true superset of the biological experience.
What happens to postbiological intelligence after this
imminent transition is presently a matter of speculative
debate, but I propose that all post-singularity systems
are constrained to create, in a very short period of time
thereafter (centuries? millennia?), a near-instantaneous
computational system capable of outgrowing the physics
of the universe that created it, and then rapidly transcending
to a new multiversal domain.
[Keywords/phrases: logistic growth ("S"
curves), exponential growth and Moore's law, simulations,
matter-energy and space-time efficiency of computation,
hyperbolic (asymptotic) growth, mathematical and physical
singularities, evolutionary development, cellular automata,
evolvable hardware, hardware encoding of environmental
algorithms, autonomous technology and the multi-millionfold
speed-up in technological vs. genetic learning, complexity
and information, computational "substrates",
free energy rate density (Phi), SETI and the transcension
scenario, disposable soma theory, cosmological natural
selection, the anthropic principle, the developmental
singularity hypothesis, a law of locally asymptotic computation,
the oversimplification of technological determinism, self-replicating
systems, self-organization, nanoscience and nanocomputation,
reconfigurable computation, the fluid and incomplete nature
of self, co-evolution of technology and biology, scenarios
for emergent AI, the "cognitive iceberg," information
meta-laws and evolutionary ethics, uploading, the developmental
spiral and subjective immortality, social insulation from
accelerating technological change, the educational challenge
of democracy, singularity studies efforts and conferences,
an overview of the systems sciences.]
Abstract II from first question &
answer of below mention interview:
An Interview with John
Smart for
KurzweilAI.net,
6.2002
Questions by Sander Olson, Jeff Thompson
and Amara Angelica.
1. What does the word "singularity"
and the phrase "technological singularity" mean
to you? Do you think we can presently see somewhat beyond
the coming singularity? If so, why call it a singularity?
What's really being discussed here?
Futurists often discuss the singularity
concept in vague and even contradictory ways. In brief, mathematical and physical singularities
occur in any universal domain where physical law allows an "asymptotic" (tending toward
infinity in a finite time) acceleration to a new environment,
a place where a new and different set of general laws
emerge. Interestingly, physical law allows this kind of
runaway only occasionally. We know of a limited, but growing
number of such domains, and it is worth comparing and
contrasting them. All singularities appear to involve
very special types of computation, if we broadly define
computation as a specific kind of exploration of universal
space and time by physical systems. Let's take a quick
look at a few examples.
The accelerating process of black hole
creation in collapsed stars is one famous place where
singularities arise. We also see them in fluid dynamics,
in the development of vortex tubes, breaking up symmetric
flow and moving into turbulence. They arise in such situations
as the pinching off of a droplet from a capillary column,
perhaps analogous to the way our multiverse cosmologists
suggest new universes may be budding off from our own.
Phase changes, such as the acceleration of a rotating
coin as it spins down onto a desk, offer another example.
Finally, the continual faster-than-exponential acceleration
of computing devices, as physical law allows
them to re-encode themselves into continually smaller,
faster and more resource-efficient forms, appears to be
yet another. This latter example, with its capacity to
create a human-surpassing intelligence in the near-term
future, and the exponential economics that have resulted
from accelerating computational complexity in recent decades,
is the main concern of those who talk about the coming
"technological singularity." But before we dig
further into these fascinating topics, let's take a look
at the more general concept of accelerating change.
To organize our inquiry, at least three important strategic
choices must be made.
First, we need to determine what dimensions
of change we will study. Physical properties (beginning with, but not limited to,
such easily conceptualized properties as matter, energy,
space and time, or MEST), known and suspected physical
law (including universal initial parameters and boundary conditions)
and the computational and informational implications of physical change (including the emergence of new physical
laws or other constraints, if we can discern them) will
be the guiding perspective of our approach. Physical properties,
laws and their computational-informational implications
are unique in that they can be used as frameworks to analyze
physical change, both qualitatively and quantitatively,
at all known levels of observation.
Second,
we must choose to analyze accelerating physical processes
within some universal paradigm of change. Two very common
paradigms that have historically been used to analyze
change are strict Newtonian determinism and classical "random" evolutionary theory. In this book summary, I will propose
the need for a third alternative, evolutionary development, one of several possible useful syntheses
of these two opposing perspectives. The paradigm you use
to analyze universal change will, of course, deeply influence
the way you interpret what you see. For this reason, we
should periodically attempt to use all reasonable paradigms
to explain the data we observe; it is possible each has
applicability only in particular domains.
Third,
we must decide what disciplines of study will be most
useful to understanding the processes of accelerating
change. It is our contention that computation, as both
a local and general universal process, is most useful
when considered from a widely multidisciplinary systems
theory perspective. As we look at a range of disciplines, we also should attempt
to interrelate, if possible, all known processes that
appear to have similar dynamics.
Fortunately, as mentioned above, there is a menagerie
of singularities that we may profitably consider. All
are models of a process of change that accelerates to
some infinite or otherwise unrecognizable, irreversible
point in which new global rules and dynamics emerge.
Briefly, the following six general classes of singularities,
while all apparently related, each have their own semi-independent
literature. Comparing and contrasting this literature
can be quite illuminating.
1. Mathematical singularities (the oldest singularity literature available) are
systems of equations that lead, under certain conditions,
to infinities, uncomputability or irreversible emergences (e.g., those hidden in the mathematics of Isaac Newton and
Albert Einstein, and uncovered by the impressive theoretical
work of Karl Schwarzchild, Roger Penrose and Stephen Hawking,
Mikhail Zak's work with "terminal chaos" in
differential equations, singularities in topology theory,
etc). Only a small subset of possible mathematical singularities
reflect known physical processes. To our knowledge, there
have been no published attempts to explain the coming
technological singularity in the language of mathematical
singularities, though such work is clearly needed.
2. Physical (a.k.a., "dynamical") singularities include symmetry
breaking,
phase changes, self-organized criticality, catastrophe
points, and other rapidly emergent and discontinous non-linear
behavior in real physical systems, including the emergence
of new physical law. Dynamically discontinous systems have been notably explored
by Per Bak, Rene Thom, various complexity and non-linear
systems theorists, and the ecological psychology academic
community. The study of singularities in real non-linear
dynamical systems (e.g., fluid dynamics) is both challenging
and a real frontier of physical theory. The "discontinuity," or rupture, aspect of the physical singularity
is an important component of the technological singularity
proposal, as we will discuss.
3. Cosmological (a.k.a., "astrophysical" or "spacetime") singularities
include black holes (primordial, quasar, stellar, supermassive,
extreme), white holes, Big Bangs and (presently doubtful)
Big Crunches,
incrementally deduced in theory by John Michell, Subramanyan
Chandrasekhar, Robert Oppenheimer, Roy Kerr, Edwin Salpeter,
Yakov Zel'dovich, John Wheeler, Roger Penrose and Stephen
Hawking and, with regard to black holes, experimentally
confirmed by our intrepid astronomical community. Cosmological
singularities may be our best model for understanding
the global attractors (endpoints, in this universe) of
a variety of processes of universal development, including,
in the developmental singularity idea, cosmic intelligence
itself.
4. Computational (a.k.a., "cognitive", "simulational" or "informational")
singularities assume the proposition of the universe-as-computing system,
and a range of different semi-autonomous computing/ universe-simulating
"substrates" operating within it, from atoms to autonomous intelligences, from molecules to
minds. Our still-early ideas of the universe as a simulation
system have been pioneered by such visionary thinkers
as Alan Turing, John Von Neumann, Ed Fredkin, Stephen
Wolfram, many complexity and systems theorists, and speculative
philosophers and physicists, such as Frank Tipler. Computational
singularities occur when a mode of
simulation/computation
used by any discrete adaptive physical system undergoes
an irreversible change, a type of phase transition to
a new regime. Solitary insects simulate their external
world in a particular way. Social insects (such as bees,
ants or termites) add a whole new layer of simulation
complexity. The shift in reference frame between these
two simulation systems represents a computational singularity.
Each operates in a relatively discrete computational domain
and organisms in one domain (say, an ant or a chimpanzee)
cannot understand certain simulations occurring in another’s
domain once the latter's simulation system has become
sufficiently quantitatively or qualitatively different.
Also known as "cognitive" singularities,
these play an important role in understanding
the coming technological singularity.
5. Developmental (a.k.a., "reproductive" or "asymptotically accelerating")
singularities assume continuous accelerating change
as an inevitable process of new substrate creation in
universal development, with elements or variations proposed
by Carl Sagan, Eric Chaisson, Andrei Linde, Valeri Frolov,
Alan Guth, Lee Smolin, Martin Rees, John Gribbin, Edward
Harrison and myself. Like the computational singularity,
this variant expands cosmology to include intelligence/information-processing/computation,
but further proposes that the major features of computation
(past, present and future), including its continuously
accelerating and ever more matter-, energy-, space- and
time-compressed ("MEST-compressed") new local
substrates, are part of a statistically determined physical
developmental process, prespecified in the special initial
conditions of the "seed" (Big Bang) that created
our universe.
Developmental theory also includes the concept of self-organization (cyclic development, incrementally tuned for
future-specific emergent order). This suggests that universal
unfolding may be understood as a chain of mathematical,
physical, cosmological and computational singularities,
and the universe itself may be parsimoniously viewed as
a complex, adaptive, developmental substrate unfolding
within the multiverse. As a result, the developmental
singularity hypothesis might be considered a universal
"singularity of singularities," though it would
still involve strong finite processes and, thus, by no
means be an "ultimate" singularity. The asymptotically
accelerating change component of the developmental singularity
is another important aspect of the technological singularity
idea.
6. Technological (a.k.a., "human competitive" or "effective machine
consciousness") singularities represent the usual
meaning of the term "singularity" when used generically by futurists. The technological singularity
proposal is an amalgam of at least four discrete concepts:
1)
"singular"
human-competitive
AI emergence (AI
"singularity")
2)
discontinuity
(a
property of physical-dynamical singularities)
3)
unknowability
(a
property of computational-cognitive singularities)
4)
instantaneity
(via
continuous acceleration, an aspect of developmental
singularity)
Note that the first word of each of these concepts (singular, discontinous,
unknowable, instantaneous) denotes a property of singularities
in general. As we will discuss later, I believe the last
three concepts are best understood within the framework
of the physical-dynamical, computational-cognitive and
developmental singularity literature and, where possible,
should be addressed as such.
The word "singular" refers to the new, unique and one-time-only
nature of the emergence of any singularity. This is a
simple and an often-overlooked dimension of the singularity
concept. Yet, given the pre-existing singularity literature,
the one unique idea that the technological singularity
proposes is its first one: that general technological
change must soon capture and permanently exceed even the
highest-level features of human biological intelligence
and autonomy. In futurist and transhumanist literature, this has been
called the AI "singularity" component of the technological singularity meme that focuses
on the human-competitive aspect of intelligence development, now occurring in global technological
systems. The
AI singularity idea proposes there will be one singular
time on Earth when technological intelligence
surpasses human biological intelligence as the dominant
form of local computation. It will be difficult to know
exactly when this competitive and permanent event will
have occurred, so various measures have been proposed.
A machine solution to a generalized Turing
Test
of "effective consciousness" (regardless of
the actual inner subjective state of the system) is most
commonly suggested as a useful indication of its arrival.
Note also that the technological singularity meme complex (idea
set) does not require the imminent runaway of local intelligence
must have some universal significance, but leaves that
issue as an open question. Prominent explorers and advocates
of the technological singularity idea have been John Von
Neumann, I.J. Good, Hans Moravec, Vernor Vinge, Danny
Hillis, Eliezer Yudkowsky, Damien Broderick, Ben Goertzel,
myself and other transhumanists and, most eloquently to
date, Ray Kurzweil in his book summary, "The Law of Accelerating
Returns". Those who would like some historical
context for these ideas might also enjoy A Brief Populist History
of Intellectual Discussion of the (Technological) Singularity
at our SingularityWatch.com Web site.
Singularities have been studied for over a century within
the theoretical discipline of mathematics, for mathematical singularities and within
the applied disciplines of general relativity and mathematical physics for cosmological singularities. In the
last three decades, these quantitative efforts have been
expanded to non-linear science
and theoretical computer science
(for physical-dynamical and computational singularities). Evolutionary psychology, ecological psychology and cognitive science are also promising new areas of exploration
for cognitive singularities. The AI singularity can be
explored within the many pastures of artificial intelligence, including, but not limited to, such
disciplines as cognitive science, computer science, information theory and computational neurobiology.
Unfortunately,
neither the full definition of the technological singularity
nor the developmental singularity hypotheses are yet studied
in formal academic programs. This intellectual oversight
is a state of affairs that FRAC hopes will change in coming
years. Nevertheless, there are specialist and generalist
degree programs--future
studies,
science and technology studies,
information studies,
evolutionary and developmental biology, evolutionary and biologically inspired computation, and astrobiology--that would provide valuable preparation
for such work.
About John SMART
http://singularity.xlogs.net/bio_johnsmart.html
E-mail: johnsmart@SingularityWatch.com
|
