Intelligence, Artificial?
The brain is the organ of the mind. Anatomists have
described the brain in terms of our evolutionary path. We have old-age,
middle-age and new-age parts, each with different properties. A neuroscientist,
Paul McLean, suggested that the human brain could be viewed as three systems of
different ages - an old reptilian brain, a middle (early mammalian) brain,
topped off with a new, advanced brain, the neocortex. The neocortex allows us to
learn, adapt and create new modes of behavior. The neocortex has the computer
equivalent of random access memory (RAM), allowing the input of new information.
This new information is used to interpret and adjust the operation of
read-only memory (ROM) which is built into old and middle brain modules and
cannot be modified. New babies are not born with the new brain programs. Old
programs are built in and need not be learned. Old programs include some of
the most negative qualities – predatory and territorial aggression, anger and
fighting, for example. Some of our most positive qualities are also innate such
as the tendencies to mate, bond and form social units with altruistic features.
The old brain remains in control of our bodies and our minds.
The central feature of intelligence is the ability to understand what is
really going on out there and to respond to events with successful and adaptive
behavior. Intelligence built from subsystems that sense,
decide, remember and act. It is fashionable to speak in terms of "mental
abilities" and to list a number of different mental abilities in terms of educational concerns, such as reading, writing, math and
music. The brain is modular with a host of different functions contributing to
intelligence. We expect and do find different arrangements of mental abilities
in different people. If you consider the intelligence test of life overall, then
you recognize that there is a range of abilities in any human population. Humans
are born with a somewhat defined intelligence potential. The spread of IQ scores
in any population represents a combination of genetic
determinants that cannot be changed and environmental determinants that operate
in a sequential manner and can be changed. Environmental determinants can be
separated into two groups”
1. Determinants that are
sequence critical
2. Determinants that operate all the
time.
Key nutrients must be supplied as the brain forms in utero on a daily
basis. Deficiency may cause irreversible damage. If the same nutrients are
deficient in an older child or an adult temporary and relatively milder
functional impairment occurs that can be reversed by correcting the nutrient
deficiency. The most common cause, in third world terms, of low intelligence is
iodine deficiency during pregnancy and infancy. Iodine deficiency has
profound implications in terms of economics, politics, human rights and dignity.
Low intelligence populations will not do as well as smarter populations and will
not be capable of fully participating in a technological 21st century. In
affluent populations, children may still be malnourished and suffer from
neglected problems such food excess, nutrient disproportion and food allergy. We
can equate normal intelligence with normal brain function. Not all brains are
created equally and some brains are damaged before birth. The world offers
abundant opportunities to cause abnormal brain function. The overwhelming task
is to avoid foods, drugs, and environmental chemicals that make people less
smart and even demented. Alcohol intoxication for example is a temporary
dementia that becomes permanent if it is repeated too often. Brain injury adds
to the negative effects of using alcohol and other psychoactive chemicals.
Smart people learn faster and learn more than not so smart people. Smart people
also are more curious, seek more diverse experiences and absorb more
information. Intelligence is manifest in the ability to acquire complicated
skills and excel in performance by practice and progressive improvement.
Competent people are smart people who have the discipline to practice and
improve their performance. There is a relationship between being nice person and
being a competent person. In demanding, professional environments the nicest
people tend to be the smartest and most competent. There are exceptions.
Leda Cosmides and John Tooby suggest: “The brain is
a naturally constructed computational system whose function is to solve adaptive
information-processing problems (such as face recognition, threat
interpretation, language acquisition, or navigation). Over evolutionary time,
its circuits were cumulatively added because they "reasoned" or "processed
information" in a way that enhanced survival....our minds consist of a large
number of circuits that are specialized. For example, we have some neural
circuits whose design is specialized for vision. All they do is allow you to
see. Other neural circuits are specialized for hearing -- they detect changes in
air pressure, and extract information from it. Still other neural circuits are
specialized for sexual attraction -- i.e., they govern what you find sexually
arousing, what you regard as beautiful, who you'd like to date, and so on.… you
can view the brain as a collection of dedicated mini-computers -- a collection
of modules… whose operations are functionally integrated to produce
behavior...So it is with your conscious experience. The only things you become
aware of are a few high level conclusions passed on by thousands of specialized
mechanisms: some that are gathering sensory information from the world, others
that are analyzing and evaluating that information, checking for
inconsistencies, filling in the blanks, figuring out what it all means.“
(Leda Cosmides & John Tooby Primer of Evolutionary Psychology: Center for
Evolutionary Psychology University of California, Santa Barbara, USA)
Artificial Intelligence
A new wave of Artificial Intelligence promoters fail to understand human and
other animal intelligence, make exaggerated claims about computing with digital
machines and market AI products as if they know what they are doing. Smart
people have a close look at all these claims and reject the commercial
hype.
On April 2, 2013, US President Obama launched the BRAIN Initiative to
“accelerate the development and application of new technologies that will enable
researchers to produce dynamic pictures of the brain that show how individual
brain cells and complex neural circuits interact at the speed of thought.” One can
be forgiven for treating megaprojects and lofty goals with considerable
skepticism. The least convincing movement attached to brain science claims that
computers can simulate brain function and will rival human intelligence soon.
This nonsense has gained both popular approval and also corporate funding from
big money corporations such as Google. The issues involve the destruction of the
meaning of intelligence and the great mistake of confusing programmable machines
with living creatures.
Programmers are Intelligent, Not Computers
When you do not know exactly how digital computers work and how programmers
utilize the hardware, it is easy to be fooled into believing that computers are
intelligent or will be soon. When you know how digital computing works,
you are less likely to believe in computers that will develop their own
intelligence. In fact, a programmer knows that he or she has to tell the
computer what to do in precise and annoying detail. Without expert
programming, digital computers are dumb machines.
Much of the polemics written about “intelligent” computers becomes irrelevant
when you realize that the real power of computing lies in the software and not
in the machine. Software is an expression of human intelligence. Computer
software is a new and powerful way of distributing human intelligence. Computer
programs collect and distribute the knowledge and the skills of the smartest
people who are in the minority to large numbers of less skilled users who are in
the majority. Specialized knowledge and procedural understanding can be
programmed in a user-friendly manner.
A simple calculator that costs a few dollars stores
arithmetic algorithms and empowers even illiterate users to do calculations
quickly and accurately. Problem-oriented hand held devices can be programmed to
deploy any number of useful algorithms. This means that a small number of
engineers, programmers, and experts that contribute algorithms can project their
intelligence and knowledge into the world, reaching millions of even billions of
people who otherwise would not be able to solve complex problems. Applications
include communications, navigation, architecture, engineering, music
composition, recording, video production, digital animation, business and
finance, currency conversion, self-monitoring and self-diagnosis. Neither the
machines, nor most of the users have the intelligence and knowledge to program
the algorithms, but the combination of programmer, machine, and user forms a
functional triad that can be reiterated without limitation.
The abstract reasoning that underlies advanced mathematics is more interesting
and is independent of the ability to calculate. Most mathematicians are happy to
do calculations on a digital machine and do not feel the least bit threatened
that some computer will take over their job of abstract reasoning. Digital
computers have no sense of meaning, cannot perceive and are only able to make
simple robotic decisions about the data they receive. They can store images
accurately and will faithfully recall stored data unless a malfunction
intervenes. Output procedures are echoes of input procedures. The biggest
advance in programming involves searching thru large databases to find the right
answers to specific questions. Goggle`s search engines represent state of the
art algorithms, designed to deliver relevant results to search inquiries.
Failure to achieve relevance remains a persistent search problem. Google
requires teams of programmers working full time everyday to monitor and refine
their software.
Neural Networks
The AI dreamers are working on circuits that require less programming and can
self-modify. Neural networks are designed as theoretical simulations
of living neuronal networks, based on the idea that cognition could be simulated as patterns of connections. The mathematical version of the neural network is
composed of processing units, or “neurons”, and they can be either hardware or
software-based. Neural nets have a training phase to build the pattern of
connections that will be applied to unknown data in the future. Neural networks
are helpless when they start out and depend on the trainer, a smart
human who figures out what inputs to select, what training criteria are to be
used and what outputs are desirable. In theory, large amounts of new data can be
processed in parallel by networks to determine the properties of input data.Are neural networks simulating what the brain does? The best answer is
neural networks are doing their own thing, but their operation has been inspired
by a first and crude approximation of how neuronal networks might work. The basic idea is
that learning involves strengthening of some connections and weakening of others
so that inputs get routed more consistently to specified outputs. A neural
network differs from an ordinary electronic circuit because its connections are
modified over time. What is different about neuronal networks? Even the
simplest neuronal network in the brain is more complex than a simulated neural
network; it grew on its own and trains itself. Much of the processing in the
brain is chemical rather than electronic so that no electronic circuit will ever
be a valid simulator. There are different kinds of neurons and some are
specialists in performing specific tasks - size, shape and
connectivity vary with specialized roles. Neurons have multiple inputs and
outputs and integrate the inputs over time using the whole cell surfaces as
topological networks. A neural network designer may be able to cope with a node
with a small number of inputs and outputs but one neuron may have hundreds to
thousands of inputs and outputs. Neuronal signals are sent by a waveform and
then converted into a quantum signals using chemical neurotransmitters. Complex
negotiations occur in the synapses about what signal will be sent for how long
and what changes will occur to the sending and receiving neuron. Neurons are
often spontaneous signal emitters. Unlike the passive nodes in the neural net,
neurons can create signals on their own; their outputs are not always dependent
on their inputs.
The Fantasy of Hal
Popular science fiction postulates that digital computers will become
intelligent sentient beings and take over the world. Arthur Clark’s Science
fiction novel and Stanley Kubrick’s movie version of 2001 were exciting in
1968. I was thrilled the sense of motion during the docking of shuttle with the
space station, transformed by Strauss’ Blue Danube Waltz. The spacecraft in the
movie was operated by HAL, the computer. HAL represented the possibility of
computers developing human-like artificial intelligence.
In 1968, anything was possible, but with subsequent developments in computer
science, we now know that living intelligence is so developed, complex and
profound that any success with machine programming is disappointing and
rudimentary. We now know that real intelligence lies well beyond the ability of
present and future digital machines. In AI there is more artificial and less
intelligence.
David Stork, a machine intelligence researcher wrote: “Perhaps a dark side of
HAL’s legacy is to have fixed an anthropomorphic view of artificial intelligence
so firmly in the minds of a generation of researchers… But those idiot savants
(AI programs) did not show even the slightest signs of achieving general
competence. In the subsequent AI winter -- brought on by the end of a military
research spree as well as the inevitable collision between venture capitalists
and reality – only the mechanical cockroaches survived.“
Mark Tildon of Los Almos Laboratories makes small robots from spare parts
derived from discarded portable cassette players. A few transistors in his
robots handle the task of moving limbs and solving problems such as getting past
obstacles or dealing with broken parts. His robots resemble insects and move
like insects. Tildon observes that living brains solve the complex tasks of
surviving as free beings in an ever-changing world by using simple and compact
circuits. He observes that efforts to make free-living robots using digital
computing fail because even simple tasks quickly grow in complexity and require
state of the art computing power.
Digital robots
Robots live in a simple domain with help from teams of humans with PhDs. They may
never compete well with living intelligence even at a rudimentary level. While the work done on robotics
and artificial intelligence is interesting and programmable machines are
everywhere, progress to date informs
us that it will be exceeding difficult to achieve the digital equivalent of the
free-living intelligence of an ant. There is an important difference between the
programmable machines that make mass production and financial systems possible
and real intelligence. Attempts to create AI and self-sufficient robotics helps us to appreciate that
the ant brain is a marvel of computation and miniaturization. We may eventually
progress to computational devices based on different materials and strategies
that are more brain-like and achieve better and unexpected results. At this
writing, no one knows how to do this. The search continues with the study of
animal brains.
Machine intelligence enthusiasts are more visible, vocal
and delusional than ever before.
Their meetings have the giddy feel of a born-again religious revival. One
god-substitute is singularity: ” Techno-Rapture. A black hole in the Extropian
worldview whose gravity is so intense that no light can be shed on what lies
beyond it. … the human mind is not the final word. Someday, human technology
will advance to the point of being able to improve on the underlying hardware
(the brain) - an event known as the Singularity. Depending on how much futurism
people have been exposed to, they tend to imagine different candidate
technologies, “different timescales, and different outcomes for humanity. The
Singularity Institute's favored technology is computer-based synthetic minds -
"Artificial Intelligence" or "AI" - which we think can be developed quickly and
with an outcome favorable to humanity … The Singularity Institute seriously
intends to build a true general intelligence, possessed of all the key
subsystems of human intelligence, plus design features unique to AI. We do not
hold that all the complex features of the human mind are "emergent", or that
intelligence is the result of some simple architectural principle, or that
general intelligence will appear if we simply add enough data or computing
power. “
Fantasy and Delusions
I have been reading essays on machine intelligence from the
Silicon Valley pundits and successful entrepreneurs. These are
people who live in a virtual reality of their own making and invent terms and
phrases such as “superhuman machine intelligence.” They tend to familiar with
digital computer programming and have unrealistic fantasies about the future.
They refer to neural algorithms in the brain that can be replicated
electronically. Neural networks are favored candidates for learning machines
because they can self-modify with a little help form their PhD builders. A neural network is just a bunch of silicon chips
until it is trained by a human programmer. Once trained, it can receive data and
produce output without further intervention. Its output may have some value or
may become gibberish. A human monitor must observe and regulate the computer output, deciding on its value. The advantage of neural
networks for unrealistic speculators is that they cannot see what is going on
inside the machine and hope that something worthwhile is happening. The paranoid speculators believe that evil output
may become the result of an unregulated machine.
There is room for fantasy and speculative thinking; however, no-one needs to
take the AI view or timetable seriously. Some of the worst future
predictions claim that digital circuitry is becoming faster, denser and less
expensive and therefore “supercomputers’ will soon emerge that have greater
processing power than the human brain. Some even suggest that massive parallel
processing is superior to brain computational abilities.
I would be grateful for continuing programming improvements and more
insightful system designs. I would also be grateful if computer games smart
phones and dystrophic movies all disappeared. I would keep simple cell phones.
Smart phones have been mislabeled. They are portable computers that render users
addicted and dumb.
There is no knowledge that allows anyone to assess brain processing ability and
no basis to compare living brains with digital computers. One of the aspects of
“futuristic speculations” that amazes me is the lack of knowledge about the
present. Another aspect that concerns me the most is the ignorance of life
processes. I doubt that any machine will soon display free-living competence.
Ant brains are amazing but digital robots are disappointing. The challenge for
future computer designers is to make robots that do as well as an insect in a
free-living competition. This task will require a new computing technology, lots
of money and the rest of this century to achieve. Unless, of course, some genius
discovers and copies brain circuitry that underlies insect competence.
I do not believe that digital computers even of great speed and complexity will
attain consciousness, nor do I believe that robots controlled by digital
computers will ever come close to achieving the self-organizing, free-living
intelligence of any animal or humans.
I am concerned about human treachery, but have no concern about machines
independently developing destructive intentions that could rival or match their
human makers. Evil is a human invention. Humans already make world-destroying
machines. This is not a future scenario. Once launched, a world-destroying
machine such as an intercontinental ballistic missile carrying hydrogen bombs is
self-sufficient. The ICBM is a dumb robot that after launch can find its way to
its target without further assistance from human programmers. A bevy of dumb
ICBM robots with hydrogen bomb warheads can destroy human civilization. The
combination of bad and dumb humans and dumb robots is to be feared. This is
history and no one has to wait for future malevolent robots to be constructed.
Attempts to emulate brain function
Zahran reviewed the challenges facing engineers attempting to emulate
brain function: ”The majority of computers follow the Von Neumann model, where the machine fetches instructions from the memory and
executes them on data in the central processing unit. The Von Neumann model has undergone numerous enhancements over decades
of use, but its core architecture remains fundamentally the same. The Von
Neumann model puts a lot of restrictions on how much we can learn from the way
the brain works. So how about exploring non- Von Neumann models? First, we do
not fully know, at least at this point, how the brain works. We have many pieces
of the puzzle figured out, but many others are still missing. The second issue
is that we may not need an actual replica of the brain for computers to be
useful. Computers were invented to extend our abilities and help us do more,
just like all other machines and tools invented by humanity. We don’t need a
machine with free will—or do we? The answer is debatable, and this is assuming
we can build such a machine!
“But what many would agree on, or at least
debate less, is that we need machines that do not require detailed programs. We
need machines that can accumulate experience. We need machines that can continue
to work in the presence of a hardware failure. The Von Neumann model is perfect
for many tasks, and, given the billions of dollars invested in software and
hardware for this model, there is no practical chance of moving immediately and
fully to a non-Von Neumann model. A good compromise may be to have a hybrid
system, similar to the way digital systems and
analog systems are used together. For instance, a Von Neumann machine executes a
task; gathers information about performance, power efficiency, and so forth; and
submits that information to a non-Von Neumann machine that learns from this
information and, in the next execution of the Von-Neumann machine, it is
reconfigured to best execute this piece of software. This is just one scenario,
but the potential in this direction is very high.
“The brain is
decentralized. This is not yet the case in a Von Neumann model or in the whole
design of the computer system. Decentralization has an effect on reliability
and performance. Even though we have several cores working in a parallel
computer system, they are all under the control of the operating system that
runs on some cores. A computer needs to be able to detect a failure and then
move the task to another part to continue execution. This has been implemented,
to some degree, with what are called thread-migration techniques. But can we
implement these on the whole computer system? Current computers are precise and
have a finite memory. The brain has a large memory that is approximate. If we find a new memory technology that provides
a huge amount of storage relative to current state of- the-art memory. Can we
design software that makes use of this memory?” [i]
[i] Mohamed Zahran Brain-Inspired Machines. What,
exactly, are we looking for? Engineering in Medicine and Biology Society (EMBS)
of the IEEE March 14, 2016