How to Create a Mind

The Secret of Human Thought Revealed

Ray Kurzweil • Viking • 2013
We’ve reviewed this title for you as we currently cannot offer a summary.
24

The Role of Intelligence

Futurist Ray Kurzweil cites intelligence as “the Intelligence." As “the most important
phenomenon in the universe,” intelligence lets humans reach beyond their biological limits,
transform the world and change who they are. Human intelligence, Kurzweil says, exists as
it does because the universe’s foundational laws enable people to encode information. That
is a far-reaching, almost revolutionary concept.

Biological evolution produced simple organisms, then more complex organisms with
nervous systems. Finally, says Kurzweil, it produced mammals, which differ from earlier
organisms because they can think hierarchically. The part of the brain known as the
neocortex is responsible for hierarchical thinking. All mammals have neocortices, but the
human neocortex is the most complex. It enables people to embrace abstract thought, use
tools and develop language. It also means that humanity has been able to amass more
knowledge than any single person could learn.

This accumulation, Kurzweil asserts, has reached a stage of exponential growth. The “law of
accelerating returns” (LOAR) now applies as machines take an active role in – and
sometimes active control of – their own evolution. Researchers in this field are now
working to understand the brain for its own sake, but also to provide tools to advance
artificial intelligence.

The Neocortex

The neocortex forms the outer layer of the human brain. This thin, folded tissue has six
layers of varying thickness and function. Kurzweil describes the basic unit of the neocortex
as a “pattern recognizer.” The neocortex uses “pattern recognition modules” to recognize
and process millions of patterns, organizing them hierarchically to solve logical problems.
The brain stores memories in “sequences of patterns.” Because of that, many people can
remember the alphabet from A to Z, but can’t remember it from Z to A. The brain predicts
how incomplete patterns will conclude. These predictions are so powerful that people
experience what they expect to see. Kurzweil sees this as central to the human experience.
He says the main reason people have brains is to “predict the future.”

People’s actions follow patterns. The sequences of patterns that you reinforce with active
use over time will stay vivid in your mind. But if you don’t use them, they will fade. All
patterns contain the input, the pattern’s name and the set of higher-level patterns to which
it belongs. The pattern for the word “apple” includes lines and shapes for each letter, the
term “apple” and the categories into which you’d place an apple. People recognize parts of
letters, then letters, words, and so on. Information flows up and down this organizational
hierarchy.

Kurzweil moves into fascinating territory as he describes brain functions. The neocortex,
he explains, follows a specific algorithm as described by the “pattern recognition theory of
mind” (PRTM). Applying LOAR (accelerating returns) to PRTM (mental pattern
recognition) multiplies the power of the human brain. The PRTM process handles “an
assembly” of roughly 100 neurons as its basic unit. The neocortex contains some 300
million neurons, which communicate with one another through a systematic structure of
connections. Various areas of the neocortex, such as the visual cortex, are also hierarchical,
as are the six layers of the neocortex itself. The neocortex’s plasticity argues for universal
processing throughout this region of the brain.

When sense data enter the neocortex, Kurzweil shows, it recalls similar past perceptions
and uses these memories to predict what the brain will perceive. This triggers millions of
pattern recognizers. Memories are “patterns organized as lists.” A pattern makes sense only
in terms of the information before or after it in the hierarchy. The neocortex recognizes
patterns by anticipating and completing the sense data it receives, storing redundant
copies of patterns, combining lists of details and letting modules encode patterns multiple
ways.

The author explains that people learn and recognize patterns simultaneously. Once they
learn a pattern, they recognize it. The neocortex continually tries to make sense of things.
Once the brain recognizes a pattern, it can synthesize it into a new pattern, creating greater
redundancy. When it recognizes a pattern, the brain sends output to lower levels, creating a
loop. Triggering one pattern in the neocortex triggers others in a “directed” fashion, as
when people try to solve problems, or in a “nondirected” fashion, as when patterns arise in
the mind, like dreams. Kurzweil weighs in on love, which, he says, promotes the emotional
bond and stable environment that support neocortical development and, thus, learning.

“Sparse Coding”

The old brain that guided promammalian (the hypothetical phase of animal development
before mammals) animals remains active, driving people to seek pleasure and avoid
danger. The neocortex moderates and sublimates those drives. The old brain presents
problems to solve, and the neocortex solves them. Thoughts related to emotion happen in
the neocortex, but the old brain influences them.

According to Kurzweil, the neocortex processes information using sparse coding. For
example, humans think they see a steady flow of high-resolution images, but the optic
nerves are sending a stripped down set of compressed data from which people “essentially
hallucinate” the world. This sparse coding lets the neocortex process information without
getting overwhelmed.
Building a “Digital Neocortex”

Irregularly shaped and deeply interconnected spindle cells appear in the neocortex when
babies are a few months old and increase up through age three. People’s “transcendent
abilities” arise from highly developed neocortices.

The evolution of the neocortex massively sped up humanity’s learning process.

Contemporary people can further quicken their learning “by migrating from biological to
nonbiological intelligence.” Once a digital neocortex learns something, Kurzweil says, the
discovery is instantly transferrable. People can use a computer or “artificial neocortex” to
augment their abilities. This process starts with using the computer’s familiar functions:
faster processing, memory and communication.

Researchers are pursuing different routes to building a “digital brain.” Some scientists
simulate the entire brains of simple organisms, while others start with a general model of
the human brain and make it more detailed. The Blue Brain project, for example, educates
its simulated brain different ways. For one, it enables the machine to learn as a human
brain does through accumulating data. Science is working on enabling machines to learn
from educated human brains by copying existing neural patterns. Such “uploading”
requires developing a noninvasive scanning process and probably won’t be available until
the 2040s. A third approach calls for creating simpler molecular models to make better
molecular simulations.

The Hidden Markov Model

Russian mathematician Andrei Markov (1856–1922) developed the Markov model that
outlines a system of states of brain activity that are hidden in such a way that observers
can’t see the internal function at each level. Instead, Kurzweil notes, scientists must
compute likely activity based on output. Adding a hierarchy to this model produced
hierarchical hidden Markov models (HHMMs). These self-organizing HHMMs are useful for
challenges like speech recognition.

When people first respond to artificial intelligence (AI), Kurzweil finds they tend to claim
that it will never be able to do X. When AI accomplishes X, people dismiss X by saying it
isn’t that impressive and doesn’t provide a marker of intelligence. This misses the fact that
AI is all around you in today's world. Siri, a commercially available program, answers
questions in natural language. Google Translate teaches its system new data by exposing it
to many examples. The result is a largely accurate translation function. Kurzweil points
out that combining such learning using a small number of coded rules brings impressive
results.

“How to Create a Mind”

The general success of AI is clear enough to indicate how to create a mind, Kurzweil says.
First, build a pattern recognizer. Second, make as many copies as your memory and
computing resources can handle. Organize them to promote “auto-associative recognition.”
Once this system recognizes a pattern, a signal moves up the pattern recognizer’s
“simulated neuron.” The recognizers form connections to the rest of the system.

Kurzweil explains to the uninitiated that scientists would have several mathematical
options for implementing the system; HHMMs would work well. You'd want to design a
digital brain for redundancy, especially of common patterns. Redundancy is essential to
accuracy. Many forms of data storage transfer generate errors or noise. Sending multiple
signals enables computers to compare messages and correct errors, allowing the brain to
recognize and anticipate patterns. Give the brain language recognition software so it can
read text. Add modules for critical thinking and identifying “open questions.”

Consciousness

The author recaps history to explain that ever since people developed computers in the
19th century, they’ve argued about how far computers can progress and whether it’s
correct to view human brains as computers. Is it possible to develop an algorithm that
would let computers function like a brain? Kurzweil answers yes.

He says to think of consciousness as “an emergent property of a complex physical system.”

This provides a spectrum of consciousness: Dogs are conscious, but less so than humans.
Computers have a place on that spectrum. “Qualia” are subjective “conscious experiences”
of some thing or feelings about that thing that are distinct from facts about it. An example
might be the experience colorblind people have of the color red. They might know facts
about red, and even cultural associations like passion or fire, but they may never actually
experience those facts or know what red looks like.

Some people view the relationship between humans and computers as being one in which
computers can know everything about something, but, Kurzweil says, lack fundamental
experience. Since no one can be sure that other people have subjective experiences, this
raises questions of faith and perception – as well as questions about whether people have
free will. When “nonbiological entities” – computers with artificial minds – convincingly
communicate the emotions of subjective states, then, Kurzweil maintains, people will
accept them as conscious.

Applying LOAR: Counterarguments and Objections

The law of accelerating returns shows that information technology follows an exponential
growth path. Computation offers the best example of LOAR. In fact, “once a technology
becomes an information technology,” LOAR applies. That means people can use “molecular
computing” to advance intelligence “trillions-fold.” Increasing computation itself combines
exponentially with exponentially accelerating communication technology. Kurzweil shows
the progression as new scanning techniques allow better understanding of brain structure,
which fuels better AI, which will support breakthroughs in computation, and so on.
People object to the LOAR and its implications for human intelligence because, Kurzweil
says, human intuition about the future is linear, and LOAR is exponential. People might
evaluate the growth of information technology based on Moore’s law alone, rather than
synthesize information and consider a broad array of developments. Other objections
misrepresent AI breakthroughs, such as IBM’s Watson, and undervalue AI achievements.
Some people argue that “strong AI” isn’t feasible because the human brain uses analog
computing, which digital computing can’t replace.

However, “multiple-bit words” allow fine gradations of computation to simulate analog

reasoning. Some thinkers argue that while computers can solve certain problems, they lack
consciousness and don’t understand what they’re doing. However, people also seldom
understand their own actions. Kurzweil suggests that computers will become conscious
because they model the design principles of the human brain. The principles matter, not
what executes them.

Human Intelligence

Intelligence is the tool that humanity applies to its problems. The neocortex is evolution’s
greatest biological achievement. “Truly intelligent machines,” Kurzweil says, will elevate
and extend the principles of the neocortex to allow tremendous new breakthroughs.

Just Plain Smarter

Kurzweil, to state things simply, is just plain smarter than most folks. As he describes how
the brain processes information, for example, you can hear his brain whirring away at a
million miles an hour. Kurzweil juggles, synthesizes, explains and then applies complex
concepts and functional descriptions. He has lived and worked in the commercial realms of
big brains and futurist thinkers for most of his career. His work in voice recognition puts
Kurzweil in a unique position to gain insights into the parallels and incongruities between
human and machine information processing. His brainiac abilities and years spent
dissecting abstruse ideas would, in another person, lead to a love of jargon and of showing
off. Kurzweil stays as down to earth as the ideas he explores permit. If you lack a grounding
in brain or computer functions, some ideas may require heavy lifting. But Kurzweil's
language never adds to the burden. He strives to be understandable and details multi-
leveled ideas with clarity and purpose.