..

Prolog: Computer Modeling Revisited

Additional Credits:
Funding
This module was supported by National Science Foundation Grants #9981217 and #0127561.

The cognitive and learning sciences include philosophy, psychology, computer science, neuroscience, linguistics, artificial intelligence, and robotics (among others). What binds researchers from these diverse fields is the aim to understand how intelligent systems work. An intelligent system is something that processes internal information in order to do something purposeful. Although humans are intelligent systems, other "natural" kinds of intelligent systems are also studied. So too are robots and other "artificial" intelligent systems.

Given the variety of intelligent systems studied, as well as the variety of sensory, perceptual, linguistic, motor and other information processing subsystems through which intelligent systems do what they do, a great deal of cognitive science research occurs through computer modeling. Within the context of trying to understand how an intelligent system works, a computer model is a computer-generated simulation (imitation) or emulation (duplication) of an intelligent system's behavior. And because 'behavior' can mean either what an intelligent system does (e.g., recognize objects) or how it does it, there are many uses of computer modeling in cognitive science. Hence, because computer modeling requires the use of computers, computers play a significant role in cognitive science research.

This is to be expected. After all, insofar as computers allow researchers to study complex systems in ways that would be impossible otherwise, computers are used as a research tool in every field of study, not just those within cognitive science.

But insofar as the computer metaphor (minds are to brains what algorithms are to computers) is the fundamental assumption upon which the study of intelligent systems rests, the status of computers within the cognitive and learning sciences is different than their status anywhere else. Consequently, computers are not just an object through which researchers study intelligent systems, computers are the subject of much of that research. The main reasons for this are identified in the following set of additional assumptions shared by almost everyone within the cognitive and learning sciences:

  1. An intelligent system is something that processes internal information in order to do something purposeful.
  2. Internal information processing occurs through computation.
  3. Understanding the nature of intelligent systems requires understanding the nature of the computational processes that make intelligent actions possible.
  4. The computational processes that make intelligent actions possible can be simulated or emulated via computer modeling.

While a great many computer models are offered to understand the nature of the computational processes through which intelligent systems do what they do, there is considerable disagreement among cognitive scientists concerning what qualifies as a computational process. Indeed! As you discovered in the lessons concerning What is a computer? and Classical (Digital) vs. Nonclassical Computers, there is considerable disagreement concerning what qualifies as a computer. It is impossible to offer a computer model of a computational process without presupposing answers to certain philosophical questions about the nature of computers and computation. Consequently, there is considerable disagreement among cognitive scientists concerning whose computational framework is the best framework within which to understand how human intelligent systems work. As there are two main computational frameworks, there are two main computer modeling (or information processing) approaches. One is classical. The other is nonclassical. While both approaches share the assumptions above, there the similarities between these two approaches ends.

According to most proponents of the classical information processing framework:

  1. Minds are to brains what programs (algorithmic functions) are to digital computers.
  2. All intelligent systems are digital computers.
  3. Something is a digital computer just in case its behavior can be described as implementing an algorithmic function.
  4. Implementing an algorithmic function requires the rule-following (serial) manipulation of discrete (digital) internal symbolic representations found in a particular parts of the system (localized).
  5. Any algorithmic function can be implemented on any digital computer (multiple realizability).
  6. The algorithmic function a digital computer implements CAN be understood without reference to the "hardware" that implements it (e.g., the brain).

Proponents of the nonclassical information processing framework view things differently. According to most proponents of this view:

  1. Minds are to brains what probabilistic functions are to analog computers.
  2. Some intelligent systems are analog computers.
  3. Something is an analog computer only if its behavior can be described as implementing a probabilistic function.
  4. Implementing an probabilistic function requires the rule-governed (often parallel) processing of nondiscrete (analog) information that is distributed throughout the system.
  5. Not every probabilistic function can be implemented on every analog computer.
  6. The probabilistic function an analog computer implements CANNOT be understood without reference to the "hardware" that implements it (e.g., the brain).

Which framework is the best framework within which to explain how our minds work? We cannot explore this question, much less settle it, without first exploring the nonclassical information processing framework in greater detail. Doing so is what we are now about to do. Connectionism and dynamical system theory both fall within the nonclassical framework. Because connectionism is widely considered the be the main competitor to the classical approach, what follows is your introduction to connectionism.


Copyright: 2006

You've reached the end of this component.
[Explore Complete Module]