The goal of artificial intelligence is to mimic the properties of human intelligence by means of a computer program. There are many approaches to tackle this challenge. Some efforts start by simulating directly logical inferences in a particular area of knowledge, another approach is to mimic the basic abilities of learning and make the system learn all the rest.
I propose not to simulate human intelligence but to use it as it is and
transplant it into a computer (the information being of intelligence). The
result is not artificial intelligence but human intelligence working in an
artificial breeding ground. Although this seems at first pure science
fiction, I believe it is possible, even with the computers of today.
24.2 Grasping intelligence.
To transplant human intelligence, we must be able to grasp it. The theory
of behavior of information is of great help here. First we have learned
that all knowledge, thus also intelligence is information, so what we want
to grasp is a set of information.
A second thing we have learned is that all used knowledge is also
expressed in the result of its application.
So wherever human intelligence is applied, it is expressed in the result.
This makes it possible to gather the information of human intelligence
without unraveling human brains. Many applications of human intelligence
are expressed in language.
It should be possible to extract human intelligence from this text.
Because there is plenty of written language available in computer readable
form, text seems to be a cheap source of this precious information.
24.3 Extracting structure.
Language is the result of the instantiation of a complex set of abstract
information. Extracting these abstract structures with preservation of
their properties is the most difficult step in the process. Fortunately,
we have plenty of text available allowing to make the proper abstractions.
24.4 Simulating the brain cells.
Our brain is a network of a very large number of connected cells. These
cells, called neurons, have been carefully studied. The logical
functioning of each of these cells seems to be relatively simple.
Since the early days of artificial intelligence (1959), many attempts to mimic the brain have been based on the believe that the structure of the brain results from the operation of a large number of simple neuron cells. Not only the logical operations of the neuron cell have been considered to be simple, the rules to "grow" new connections where also considered to be simple.
Several simulations, with varying complexity have been made (and are still
made today). Some stimulating results have been obtained. However, the
level of comprehension remained very low. The study of these networks has
been valuable for pattern recognition.
24.5 Difficulties.
From the point of view of the theory of the behavior of information, the
physical brain cells are only the most indurated externalizations of
intelligence. The brain is an induration of something which functions at
smaller, minute level. Before any connection between neuron cells grows,
the result of this considered connection has been tested out many times at
a smaller level. A new connection between neurons can only grow because
the cells are "aware" of what is going on. Especially in the development
stage, sophisticated messages are exchanged between the neurons discussing
pro and con of establishing a certain connection. However, it is true
that a developed brain functions mainly by means of the indurated
connections.
24.6 Unlimited complexity?
The heaviest structures in our brain are
externalization of something functioning faster and more stable at smaller
level, let us say at molecular level. The establishment of neural
structures is assisted by the functioning at this molecular level. The
functioning at molecular level is the externalization of a similar
functioning at atomic level. The construction of structures at molecular
level is assisted by the functioning at atomic level. We can go on several
layers more up to subatomic particles (as far we can see).
Simulating subatomic particles in a computer to simulate molecules to simulate cells to simulate the brain to simulate intelligence is of course impossible. Besides these levels of operation, there might be an influence from newer, more intelligent structures of less indurated informatter.
This seems a nice proof that the whole problem is of such complexity that
any attempt to realize real transplantation of intelligence or simulation
of intelligence is impossible. However I strongly believe that it is
possible.
24.7 Limited complexity.
When an organization indurates, the abstract very universal elements
become specialized in repeating always the same function. After being
indurated, it looses all flexibility and it looses even the ability to do
something else. This is also the case for the structures used by
intelligence. Similar to the fact that a stable brain runs mainly on the
almost mechanical reflexes of neurons, the developing brain runs on the
already indurated reflexes of atoms and only partially the flexibility of
molecules. The development of a new body does not involve the development
of all new atoms and molecules. Indurated atoms are used and even some old
molecules are used (otherwise we would have a propagation at more abstract
level).
Because we would already be happy with a copy of intelligence
(information) disconnected from its very abstract origin, we have only to
take a few levels of externalization into account. This limits largely the
scale of the project.
24.8 Memorizing the reflexes of the lowest levels.
The most external structures such as a context described in chapter 22
must be highly flexible and being constructed dynamically (rapidly
changing). Deeper levels are more rigid.
The abstract structures in these levels are limited to almost reflex like
actions.
I suppose it is sufficient to consider only three levels of
externalization. The most internal (abstract, atomic particle) is fully
indurated and a number of such elements are represented by memories of
their reflexes (extracted from text). The next layer is an externalization
of these elements forming an organization and being aware of their place
in the organization (or even several simultaneous considered places in
several organizations). The top layer is then a context structure similar
to the description given in chapter 22.
24.9 Computers.
Although classical computers do only one thing at a time and in a natural
information structures there is plenty of simultaneous activity, they are
at least good in some things:
It is very easy to represent a mapping and a varying distance between structures in a computer. There is no problem with physical space and no pipe like structures have to be build as in our brain. In other words, computer models are not restricted to three physical dimensions.
It is also easy to have many simultaneous instances of an abstract element in a computer memory. All the stable information (knowledge) which is present in every molecule and cell in our brain has only to be represented once in the memory of a computer and can be shared by all the instances.
Another advantage is the possibility to take a copy of the computer memory at almost any moment to create a copy in a separate computer going on to learn from the point reached. In a further stage, experience can be exchanged between the computers to share the findings.
Another possibility is (automatically) analyze the acquired information at
a certain stage and to remove some flexibility to gain in speed. This
process is a kind of compilation of the acquired information translating
it into instructions which can be executed directly by the computer.
This will allow to run fairly intelligent programs on computers of the
size of the personal computers of today.
24.10 New instantiation.
The theory of behavior of information is abstracted from a number of
models of our intelligence I build in the last ten years for the purpose
of artificial intelligence. I needed this abstraction to combine the
structures of several models. It is my aim to start soon a new
instantiation of this abstracted knowledge into a computer model. I hope
to publish the results of these experiments in a separate work.