[ Mini-Organic
Computers | Neural Networks | The Society of Mind ]
ver since
the modern computational computer was invented have people began wonder if the human brain
could be simulated--perhaps so well that the organic brain could be replicated in
a machine! Of course, the human brain is a vastly complex organ that scientists are
just beginning to grasp how it functions. Nevertheless, there are several theories
and experiments to determine how intelligence seems to spontaneously emerge from natural,
physical causes.
The brain. A [dimensions], [weight] pound organ that look like
walnuts. Unlike the other parts of the body, the brain is a highly complexly
organized mass of cells whose only function is to control the entire body from its
development in the fetus to death.(
The special cells the brain is made of are called neurons which are designed to receive and transmit electro-chemical impulses between themselves that
somehow lead to thought and intelligence. In a phenomenon that is often described as
"all-or-nothing", neurons are either "on" when they receive enough
stimuli from sensory cells or other neurons and transmit an impulse or simply off when
they don't transmit anything. There are no "weak" or "strong"
impulses, just whether or not an impulse is produced or not. It is from this
physiological action that have led some AI researchers to see neurons as analogies to the
"on" or "off" of a computers' components operating on boolean logic. Thus, the brain may be
described as a mini-organic digital computer. (For more information, go to "How Biological Neurons Work")
The brain is often called a "neural network" of
neurons whereby individual cells act as processors that share information with other
neurons to produce thought. This idea is the basis of artificial neural networks in
which researchers build specialized hardware and software to simulate the memory and
thought process of biological neural networks.
One neural network project that met some marginal success was called the
perceptron. Designed for character recognition, it consisted of three types of d
evices: sensory
units, associative units, and effector units which are analogous to the sensory,
associative, and effector neurons in humans. As the name implies, the sensory unit
was connected to a light-sensitive device like an eye and fed the information to the
associative units. The associated units begin with random connections but receives
the data from the sensory units and adjusts the paths of the electrical impulses through
trial and error throughout the network using feedback loops. The effector units
interpret configuration of the network and decides what the entire network is actually
seeing. The "random to order" perspective was compared to the human brain
which was thought to be composed of neurons that are randomly interconnected at birth and
organize themselves as a person learns more from experience.
Thus, the artificial neural network enjoyed some success by being able to
learn and identify printed letters, though Marvin Minsky and Seymour Papert's highly
influential book Perceptrons pointed out that there are just some pictures that
perception could not resolve like whether lines were connected or not. Also, the
perceptron could not recognize letters that were distorted by being tilted or changed in
size. Nowadays, it is known that the recognition of the loops and the curves in
letters are in part pre-wired into the brain and are not as random as the study of
perceptrons led early AI researchers to believe. Biological neurons have a
greater capacity of memory and processing power than previously thought which is
reflective of the more sophisticated neural networks using a whole computer to simulate a
single neuron which adds greater parallel-processing.
In 1986, Marvin Minsky wrote The Society of Mind
in which he described his perspective to produce intelligent machines.
In the book, Minsky believed that the mind was composed of many little "agents,"
each with very specific and different tasks. Organized in a hierarchy, agents may
have to break down its tasks to simpler processes which it either calls on existing agents
that specialize in that simpler task or creates new agents to accomplish the main agent's
goals. In the case of image-recognition, lower agents can be used to identify the
features that make up an apple and the agent that controls those lower feature-oriented
agents will make the decision that it is an apple.(Crevier 254-255)
Minsky's agent theory could also explain the fast changes in emotions
babies feel as opposed to an adult. The minds of babies do not have as many agents
in their brains as adults due, so the number of steps it takes to access the agent of
"laugh" to "cry" would only take a few changes. Adults have many
more agents in between "laugh" and "cry" that would have to be
considered and accessed before any significant mood changes occur.
The combined efforts of those agents, Minsky believes, produce
intelligence and the "self"--the feeling of being one person with one
consciousness. Reminiscent of the mind-body
problem, Minsky does not go into detail as to how many little agents relate to neurons
except that neurons seemingly break down thought processes to simpler components nor does
he explain how agents lead to consciousness.
