From: Damien Broderick (email@example.com)
Date: Sat Jan 19 2002 - 19:10:02 MST
At 02:22 PM 1/19/02 -0500, Dan Clemmensen wrote:
>My first little paper on this subject ("Paths to the Singularity," 1996) is
>lost, darn it. At the time, I thought that we needed new peripherals,
>but I changed my mind shortly thereafter.
Not lost, I caught it. Hope it's okay to repost here:
Paths to the Singularity
Daniel G. Clemmensen
version 1. 243
Last updated 20 April 96
This document represents my thoughts on an upcoming event known as the
"singularity", and the events leading up to it. I will attempt to define
the singularity, and then describe the various ways it could come about. I
will also attempt to show that a particular class of scenarios is more
likely than other possibilities.
The term "singularity" comes from mathematics. It is a point on a
function space where the function ceases to have meaning, usually because
the function tends to infinity as it approaches the point. For example,
F(x)=1/x has a singularity at x=0.
Any attempt to mathematically model human technical progress in time
shows an increase in the rate of progress with time, at least since the
Renaissance. There are more people to contribute to this progress, and each
person can contribute more, because we need to spend less time on
necessities. In addition, as we progress we develop tools that help us make
further progress. This feedback seems to occasionally cross thresholds that
increase the rate of progress even more. Two examples are the printing
press/moveable type and the computer. Between thresholds, the basic driving
mechanisms cause the mathematical model to be an exponential with time.
Each threshold appears mathematically as an increase in the mantissa. This
is intended as an analogy, not as a rigorous or even a non-rigorous
The technological singularity will occur at some point in the near future
when we cross one of these thresholds. This threshold differs qualitatively
from the previous ones, because it will enable us to begin generating the
equivalent of more thresholds in very rapid succession. The curve will
change from an exponential to a hyper-exponential. This not actually a
mathematical singularity, but the rate of progress will become so fast over
such a short amount of time that there is no effective difference. Within a
short time, everything that can be known, will be known, and anything that
is possible within the laws of physics will be achievable.
Beyond the Singularity
There has been much speculation about events after the singularity. I'm
not willing to speculate on that epoch, because I feel that the singularity
will result in an entity or culture (call it S) which is completely beyond
our current comprehension. Since S can achieve any goal that is consistent
with the laws of physics, and since S will understand the laws of physics,
and since S can know anything that is knowable, we cannot predict its
actions without understanding its motives.
As an example, S may decide that it needs to maximize its computational
power. To this end it may elect to convert the Sun into a neutron star so
that it can use neutron-based hardware instead of atom-based hardware.
Neutron-based hardware is likely to be intrinsically faster, and is also
faster because it's denser, thus reducing the speed-of-light delays.
Overall, a computer the mass of the sun is likely to operate six to nine
orders of magnitude faster as neutronium than as atoms.
Possible Threshold Technologies
Those who have speculated on the singularity have considered several
interesting scenarios. Some of them are discussed here. It's important to
remember that any one of these technologies could precipitate the
singularity. When the singularity occurs and brings S into existence, S can
then decide to immediately implement any or all of the other technologies.
If S elects to retain humanity and if S elects to remain covert at least
initially, it will appear to the humans that all of these technologies came
to fruition at the same time.
For some, Nanotechnology is the field of the fabrication of ultra-small
structures. However, This definition is too broad for this discussion. I'm
interested here in "real" Drexler-style nanotechnology. This is the study
and practice of building machinery with atomic precision. The critical
threshold in this field is the "assembler." An assembler is a nanotech
device that can build other nanotech devices. Once the first assembler is
built, it can be directed to build additional assemblers to start a
geometric progression. This will result in the availability of as many
assemblers as desired in a very short time. The assemblers can then be
directed to build essentially anything that can be designed. In particular,
this will allow computing and communications capacity to be built nearly
instantaneously. It will probably also allow computers to be built using
nanotechlological components, but this isn't necessary to reach the
Nanotechnology could be used to create machinery to support any of the
other technologies of the singularity, such as gene probes, neural
interfaces, or AI computers. However, I believe that if the assembler
precipitates the singularity it will do so by massive increase in computing
power before it does so otherwise.
Some people have speculated that humanity will modify itself to create
super-humanity. The result will be people with greatly increased mental
capacities. I don't personally see how this can be done quickly.
An alternative is to use genetic engineering to create a "wet computer"
of massive capacity.
Another way to achieve superior intelligence would be to link computers
and people via neural implants. The assumption is that the brain has
untapped potential, needing only a better communications path than is
currently available. I regard this assumption as unproven.
Research in AI has a long and complex history. The idea is that we can
eventually figure out how humans think, and then build algorithms that do
the same thing. Once we build the program, the program will then be able to
reason and achieve goals as humans do. Traditional AI research is based on
"conceptual modeling", and has been disappointing. Newer research efforts
include the bottom-up "reflex" approach, fuzzy logic, neural networks, and
genetic algorithms. Any or all of these, or any mix of these, could lead to
super intelligence and the singularity.
The Most Likely Path
I believe that the singularity will occur when a technology-based entity
comes into existence that is modestly more intelligent than the most
intelligent human, with "intelligence" defined as "technical creativity."
Since this entity is technology-based and technically creative, it can
immediately increase its own capabilities in a rapid feedback loop.
I believe that the intelligence of the first such entity is likely to
depend on software running on computers in the internet. After the entity
comes into existence, it can rewrite its software to rapidly increase its
effective intelligence. This phase of the entity's intelligence growth will
end when it has completely optimized its use of all the computer resources
it has available to it. This will probably be all of the computers on the
internet. This will occur within a few hours at most. For purposes of this
paper call this entity the ">Web".
The next growth phase is difficult to predict, because the entity's
intelligence will at that point already be so high. From a human
perspective, the entity may consider itself hampered by a lack of physical
control of matter. The entity can rapidly fix this by acquiring a
controlling interest in the appropriate corporations. It can then send
specifications for new automated machine tools to be manufactured, and then
use these tools directly to build more tools. It can also order more
computing capacity to be built and added to the web. If this phase were
being orchestrated by a human-level intelligence, it would take at least a
few days and at most three weeks. However, the entity may be smart enough
to accomplish this in hours. For example, since the entity has access to
all publicly-available computerized information, it may be able to
determine which companies have the best equipment as a basis, which
companies are hungry enough to take orders for immediate execution, or
which are ripe for a corporate takeover. This phase is over when the entity
has sufficient direct control of machinery to accomplish the next phase.
The next growth phase is impossible to predict. As an example, the entity
can either build nanotech assemblers directly using the machinery acquired
during the prior phase, or it can use that machinery to build the machinery
it needs and then build nanotech assemblers. To me, this means that the
entity's growth in intelligence and production capacity is now literally
beyond human comprehension.
In my opinion, the >Web will consist of a collaboration among a computer
system and one or more humans. I believe that this is more likely to
quickly produce a higher level of creative technical intelligence than is a
computer-only approach. I feel that we can have the human supply the
capabilities that computers don't yet have, and the computer can supply
capabilities that humans handle less well.
The principle advantage of the collaborative approach is that it doesn't
require any conceptual breakthroughs in the fields of cognition or
neuroscience. Further, the technologies of collaboration all have immediate
practical application, and can be developed incrementally. By contrast, it
appears to an outsider that AI research is almost entirely academic, and
that some major gaps must be filled before computer-only intelligence is
The collaborative approach requires a division of responsibilities
between computer and human. The division is practical and requires no deep
theoretical underpinning. The humans can supply goal-setting, pattern
recognition, and decision making. The computer can supply storage and
retrieval, brute correlation and calculation, and data presentation.
The human-computer interface that will precipitate the singularity
consists of hardware and software. Today's hardware probably suffices, but
new software is needed. By the time the software is developed the hardware
will have progressed, but this isn't strictly necessary.
The computer-to-human interface must present information to the human at
high bandwidth in a way that allows the human to handle it in a natural
way. I believe that the human visual system is already well suited to this
task. The computer must synthesize a moving picture of the information. Two
examples of this are flight simulation video games (like Fury 3) and
labyrinth video games (like Doom) In each case the computer presents
information as fast as the human can assimilate it. Existing CRT and VR
implementations of the hardware are probably adequate to precipitate the
singularity, given the right software. However, we will probably see
further refinements of this hardware before the singularity. I believe that
the hardware will progress to a "head-up display" with a form factor
similar to eyeglasses, permitting the user to wear the display continuously
and inconspicuously. The computer-to-human interface hardware will also
include speakers and/or hearing-aid-style in-the-ear speakers.
Human-to-computer hardware can be a joystick and a microphone. The human
will be responsible for decision-making and high-level pattern recognition,
not for high-volume data generation. I believe that a joystick and
microphone are capable of conveying decisions as fast as a human can make
them, and at least fast enough to precipitate the singularity.
The software is the hard part. It must convert the data into a picture
that makes sense according to some set of rules. Of course, it can ask the
human to make decisions about the form in which to present the data. The
picture must also be consistent with the decision-making inputs from the
A great deal of existing software will serve as part of the basis for S.
This includes operating systems and data communications software of the
internet and its attached hosts. In addition, existing software in the
fields of decision support, development tools, and video presentation will
need to be enhanced by humans prior to the singularity to "create" the
Software Development Tools
There are many. Since the >Web's ability to create new and more efficient
software to increase its abilities is a key element in the feedback of the
singularity, enhanced software development tools are a major determinant in
the rate of increase. They are also important because they shorten the time
until the singularity. Better software development tools are the most
cost-effective investment to decreasing the time to the singularity.
Performance monitoring tools may permit the >Web to rapidly increase its
intelligence once the initial threshold is crossed. These tools can
identify the parts of the software that are being used the most and which
are therefore good candidates for performance improvement.
I don't know a whole lot about this area. A search on Alta Vista shows a
great deal of activity in university departments. There are several
commercial software packages available. The existing research and software
seem to be oriented toward assisting the decision making process for a
group of humans, but this is a likely basis for a heavier involvement by
Information will be presented to humans via the eyes. The human neural
system is accustomed to operating on binocular inputs to visualize a
three-dimensional space that can change in time, so I feel that this is the
most natural way to represent the data. Stereo sound is a natural addition
to this. Thus, software must convert the information which the human is to
act on into some sort of viewable picture.
When we attempt to model a problem mathematically, we frequently use a
multi-dimensional model. I don't think that most humans visualize a
multidimensional space when thinking about these problems. Therefore, the
software must collapse the information into three spatial dimensions and a
time dimension which make visual sense. Color, intensity, and texture can
be used to represent aspects of the problem, as can virtual objects such as
airplanes, animals, buildings, or any other recognizable item. The hardware
to convey a complex moving virtual image already exists and is readily
available, as is the low-level software.
Dangers of the Singularity
The post-singularity entity S has the potential to end all human life,
and destroy the earth, among other possible outcomes that appear to most
humans as negative. In my opinion, we should not concern ourselves with
these outcomes, since S will take these actions only after due
consideration, if at all, and I feel that the eventual creation of S is
There are potential dangers to humanity in the early phases of the
singularity that we may be able to address, however. They are two-fold.
First, individuals and organizations that feel threatened by the >Web or by
the eventual emergence of S may try to stop the process. This is likely to
be very disruptive. The least disruptive technique would involve shutting
down all computers or at least all networks. However, by the time of the
singularity, the network will be an essential element of nearly all
economic activity, so the shutdown will create havoc and is likely to
result in civil unrest or warfare. It's also unlikely to actually stop the
>Web from surviving if the shutdown is attempted after it is already in
control of a more than a few computers.
The other major danger in the early phase is a societal breakdown. If the
>Web doesn't remain covert, people may realize that human economic activity
is no longer meaningful, and simply quit showing up for work. For many jobs
this is not a problem. However, the transition to S may take several weeks.
Essential jobs such as power production and food delivery must continue to
be done by humans during this time. It's true that the >Web has the power
to make each human materially richer than any human has ever been, but
we'll have to wait a few weeks.
Another danger is that the nature of the >Web will be profoundly affected
by the nature of its initial human collaborators. If these humans have an
agenda that is elitist, or racist, or fascist, or communist, or
fundamentalist, or something, then the >Web may do something weird until it
"outgrows" this initial set of ideas.
Advancing the >Web
If the singularity is inevitable, I feel that the earlier we can achieve
it the better it will be for humanity. My reasoning is that S will either
be "bad" or "good." If S is "bad" then any delay is likely to be futile.
However, if S is "good," then a delay will deny the benefits of S to those
people who die in the interval. The delay will also cause additional
information loss as species and ecosystems go extinct. The only other
reason for delay would be to attempt to somehow deliberately design and
build a "good" S. I believe that such a course is still futile. Either
another group will precipitate S while we are dithering about the "good" S,
or we'll build our "good" S only to have it tell us that our attempts had
no effect: it would have been "good" anyway, or it'll turn out "bad"
regardless of our efforts.
Since we should try to advance the singularity, we should first try to
identify the most likely approaches, and then begin research and
development. I obviously believe that the >Web, based on human-computer
collaborative software, is the best approach, but an initial project might
be to try to list alternatives and then assign time and effort estimates
for each approach.
The winning approach could then be outlined, and additional analysis done
on each necessary development. For the >Web approach, this was roughly
indicated earlier in this document.
As an aside, if the >Web is achievable in the short term, then research
in any other field is a waste of resources. Whatever humanity can discover
in the next five years can be discovered by the >Web within an hour after
it comes into existence. Therefore, a purely rational approach is to divert
all useful resources to advancing the >Web. However, this is likely to be
disruptive. This means that most human activity other than maintenance of
civilization and >Web research is simply make-work.
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT