|Search the Mindcreators site:
3.0 Research Purpose And Philosophy
1. Conversion of a Functionalist
It should be obvious by now that I strongly disagree with the theories of functionalism,
and that I feel that most of the work done in connectionism is going down the wrong
path. However, this was not always true of me. I have been interested in the field
of Artificial Intelligence and robotics since I was a little boy and saw my first
Star Wars movie. I remember thinking how cool it would be to build my own C3PO to
do all of my dreaded chores. From that time on I was hooked. As I grew up I taught myself
to program the computer and wrote a number of simple AI applications. When I was a
teenager I wrote a expert-system based application (lots of if-then rules) that could
play poker. It had a simple system to use its past experiences to become a better player
(choose which rules to follow based on the current hand and past experience).
Eventually it was able to beat me and my friends a decent amount of the time. I also did a
lot of reading on the subject, all of it preaching the functionalist faith. I did not
understand a whole lot of the explicit details at that time, but I did come away with
a strong belief that the brain was a computer and that all you had to do to build
thinking robots like C3PO, Data, or Daneel was to be able to write a thinking program.
I began work on a new project my senior year in high school. I wanted to build a
handwriting recognition system. I just knew I could build a system that would work incredibly
and be the first step to making really intelligent machines. For the first time I was
really forced to think about how we think, instead of just reading about what other people
said should work. After a lot of hard work I discovered that something that seems so simple
that we just take it for granted, like recognizing a written symbol, is really incredibly
complex. And that while something like playing a poker game can be characterized by formal
relations and if-then logic fairly well, it is horribly difficult to do that with mental
processes like recognition. There are simply to many possibilities and exceptions for it
to be programmed in explicitly. I began to doubt the prevailing dogma that thought can be
explicitly programmed. I began searching for other ideas on how the brain works, and by
shear luck I stumbled across the book Neural Darwinism By Gerald Edelman. I only
understood a fraction of the book at the time. If you have ever read anything by him then
you know that he is very difficult to understand. But it was enough to set me on a new path
in my quest to create intelligent machines. He believes that understanding the structure and development
of the brain is critical in understanding intelligence. Studying his books then led me
to study Searle's books, and to learn about genetic algorithms, chaos theory, emergent
behavior, etc. The more I read, the less realistic the idea of functionalism seemed.
2. Research Philosophy
I do not have any set philosophy really. I believe that there are a few different philosophical
ideas out there that are going in the right direction, but will require a lot of work to either
confirm or disprove. The two I like best are Searle's Biological Realism and Edelman's Neural
Darwinism. They are similar in several ways. Searle focuses more on the philosophy, and Edelman
is more concerned with the physiological details of how things like consciousness arise.
Some of my core beliefs on the mind can be summarized as follows:
The mind is a physically emergent process created by the highly complex and non-linear interactions
of billions of neurons in the brain. In order to understand things like perception we must come
to understand how these processes emerge from the interactions of the underlying physical structure.
We have to study the working examples that nature provided us. We can not get a high-level functional
view of the mind until we understand the lower level details of what causes the mind to emerge.
Discussing philosophy on things like consciousness, internationality, language, and qualia are all fine
for passing the time in a good debate, and getting some ideas on possible future research.
But we still have a lot to learn before we can really talk about these concepts in an intelligent manner.
We still have never even created something with the intelligence of a roach yet. A creature that is capable
of great feats of adaptation and survival, but in no way has intelligence even approaching the complexity
of humans or larger animals. It is not capable of consciousness as we think of it in human terms. How
can we ever create something with the intelligence of a human if we can not even create something with
the intelligence of roach? As with everything else, you must start small and work up incrementally.
Once we are able to create a robotic system with the intelligence of an
insect, then we can move up
the evolutionary ladder and eventually have the knowledge we need to talk intelligently about what
The mind can not be explicitly programmed. And I believe that a neural network that causes a
realistic mind to emerge can not be explicitly built.
It is simply too complex. The insect simulator system described here uses less than
200 neurons. And even though such a simple system can produce astonishing
results, this system only reproduces the barest outline of how a true insect's
brain works. Even a simple organism like the roach has between 250,000 to 500,000 neurons.
The only way we will ever be able to create machine organisms with this order of
complexity is to use the same tool that created the original: Evolution. While I do
not believe it is feasible to explicitly build a neural network of this complexity, it
is imminently realistic to believe that we can harness evolution and make it
build such a system. After all, that is how the originals were created.
That leads to the next big question. Why? Or more accurately "What is it good for?" Why do I spend my
time doing all of this studying and working to build this stuff? My answer is that I believe that if
we can truly get machines that even have the intelligence of insects, it can cause dramatic changes
to every part of human existence that will be even more helpful and exciting than the computer revolution.
There is virtually no human endeavor that could not be made more efficient, less costly, and less polluting
with the addition of some machine intelligence. Everything from construction, agriculture, mass production,
sanitation, recycling, the military, and keeping your yard maintained and your house clean could benefit
greatly from even a simple machine intelligence. I will not go into all of the possible uses that I
have fantasized about building. But I would like to mention a few of the possible uses in some more
detail in the following two pages. The first talks about some ideas for
a subject I am very passionate about,
space exploration. The other page discusses some possible uses for
Machine Intelligence (MI) in military projects