Share Site Map Contact us Home Page
Home > Theory > The Child Machine > State of Mind


Research Plan

Applied Philosophy

The Child Machine
Education
State of Mind
First Words

White Papers

Eleanor

Introduction to Social Solipsism (off topic?)
State of Mind
  Printable version
What kind of "brain" does an intelligent machine require? It is likely to be very different from the human brain; a machine does not require motor control and the ability to eat, although it must be able to discuss them. Our machine's brain need not control and operate the complicated machinery of the human physical body. It may be much simpler, as it is only concerned with a single function: language manipulation.

No one knows exactly how the human brain works, or how infants acquire language. So far, all attempts to design machines with real language ability have failed. So, how will Ai's approach succeed where others have not?
 
First, it's important to understand that we are not trying to create a digital model of the human brain. We are trying to simulate human intelligence, as measured by the use of language. The structure of the child-machine's "brain" is likely to be entirely different than the structure of our own.
 
No prejudiced assumptions are made on what the child's "brain" will look like, and we feel that any such preconceptions may hinder our efforts. Instead of trying to build an adult mind from scratch, we are programming our child-machine by using general learning algorithms. This is a bottom-up approach to language acquisition, in contrast to the common top-down approach.
 
In the top-down approach, you tell the child-machine about the world, hard-wiring the knowledge into the system. Logic programmers enter facts like "dogs have paws" and "fathers are male parents" into a knowledge base. Top-down experiments, like the CYC project, have shown us that it is possible to spend millions of dollars, program millions of rules, and still end up with a machine that can't figure out the difference between an article of clothing and an article in a magazine.
 
With the bottom-up approach, we start with a blank slate or, mostly blank. The Ai child-machine began with no words, no knowledge, and very few guidelines - just a simulated desire to learn. It was programmed with some basic, useful rules of information theory, so that it can recognize patterns - and patterns of patterns.
 
Dabbling in Babbling