Searle defines thinking as a biological process that can only be simulated but not duplicated in silicon, just like other biological processes such as digestion. Taking this definition to be true, it follows that thinking cannot be duplicated. But in analogy, if running is defined as a process that includes contraction and relaxation of muscles controlled by signals from central nervous system, we cannot create machines that can run. However, we can have machines that can traverse along a given path. Similarly, if we think of definition of thinking as manipulation and grounding of symbols then that it can be duplicated on silica.
In an simulation experiment by Harnard using neural networks, a decrease in within-category inter-stimulus distances and an increase in between-category distances was observed on training neural networks to sort lines based on their length. The networks exhibited successful categorization, as well as natural side-effect similar to human categorization, i.e., CP (category perception). In other words, within category compression and between-categories expansion can be observed both in humans and networks.
We agree with Searle that understanding natural languages requires semantics but the kind of semantics provided by an internal semantic interpretation though syntactic in nature is sufficient and hence computable.
We did additional reading from the book "Syntactic Semantics: Foundations of Computational Natural-Language Understanding" by William J. Rapaport. Rapaport claims that computers are capable of manipulating formal symbols in a way which is sufficient to understand natural language and hence, it is possible for computers to understand natural language. To support his claim, he has presented a prototype AI Natural-Language-Understanding system which uses knowledge representation and reasoning to "build" a mind CASSIE. Following is a brief conversation with CASSIE(taken from [6]) -
- Young Lucy petted a yellow dog I understand that young Lucy petted a yellow dog
- What is yellow a dog is yellow
- Dogs are animals I understand that dogs are animals
- Who petted an animal young Lucy petted a yellow dog
- Lucy is clever I understand that young Lucy is clever
- Lucy is a girl I understand that clever young Lucy is a girl
- Girls are people I understand that girls are people
- What was petted by a person a yellow dog was petted by clever young Lucy
- The dog is named Rover I understand that Rover is a yellow dog
- What was petted by a person yellow Rover was petted by clever young Lucy
- For every d if d is a dog then there is a b such that b is a boy and b owns d I understand that for every d, if d is a dog then there exists a b such that b is a boy and b owns d
- The boy is named John I understand that John is a boy
- Who owns Rover John owns yellow Rover
- John believes that Lucy is rich I understand that John believes that Lucy is rich
- Who is rich I don't know.
- Who is clever young Lucy is clever
- John believes that Lucy is old I understand that John believes that rich Lucy is old
- Boys are people I understand that boys are people
- Dogs are pets I understand that dogs are pets
- For every p and d if p is a person and d is a pet then p loves d I understand that for every d and p, if p is a person and d is a pet then p loves d
- Who loves a pet clever young Lucy loves yellow Rover and John loves yellow Rover
Harnad, Cangelosi and Greco in their papers talk about the "Symbol Grounding Problem" which is the problem of how words(symbols) get their meanings, and hence to the problem of what meaning itself really is?
A word(symbol) has two aspects - its referent and its meaning. A referent is something that the word is referring to and is not the same as its meaning. Consider the example taken in presentation today, the word "out" was referring to the processes of moving, escaping and spilling but the meaning of "out" is to be absent from a pre-defined space (not necessarily physical).
There is a view that the meaning of a word is the rule or features that one must use in order to successfully pick out its referent but its not clearly mentioned how it is done i.e. how is "being absent from a pre defined space" picking up "moving"?
There is another notion that has been talked about and that is - consciousness. It is a highly debated topic that what consciousness actually means but to give a broad sense we would like to adopt the following line from Wikipedia - "Consciousness is the quality or state of being aware of an external object or something within oneself". I do not know wether it implies to a particular level of activity in brain or wether it can emerge from a non biological system like computer.
Now, if some one can make a conscious connection between a word and its referent, then we can say that the word is grounded. Put in another way if one can consciously pick a referent for a symbol then we say that the symbol is grounded. So the symbol grounding problem boils down to the question that how one makes a conscious connection and if we try to ground symbols in non biological systems like brain what would the procedure be? The second question can be answered if we know the answer to the first. If we know how actually we are making connections then the procedure can be expressed in a language and hence can be encoded in a computer.
Harnad in his paper "The symbol grounding problem " proposes a way that might be the one through which we connect the symbols to its suitable referent. He proposes that a bottom up approach is followed and a representations of inputs that we receive is formed. Three kinds of representations are described(definitions taken from [3]) -
- Iconic representation : Internal analog transforms of projections of distal objects on our sensory surface.
- Categorical representations : Iconic representations that have been selectively filtered to preserve only some of the features of the shape of sensory projections that reliably distinguish members from non members of a particular category of the objects.
- Symbolic representations : These are the representations derived from combining one or more kinds of iconic and categorical representations that have been grounded previously. Zebra from horse and stripes.
He further proposed connectionism as a possible mechanism to explain how the all important categorical representations could be formed and how the hybrid system find the invariant features of the sensory projection that make it possible to categorize and identify objects correctly.