Neural Computation and the Theory of Cognition
At its core, neural computation is the process of information processing carried out by networks of neurons. This concept is deeply rooted in the philosophical tradition of computationalism, which posits that the mechanisms of neural computation are what explain human cognition. The foundation for this field was laid in 1943 by Warren McCulloch and Walter Pitts in their landmark paper, "A Logical Calculus of the Ideas Immanent in Nervous Activity," where they first proposed that neural activity could be understood as a computational process.
Key Facts
- Neural computation is the information processing performed by neuronal networks.
- Computationalism is the thesis that cognition is explained by neural computation.
- The field is divided into three main branches: classicism, connectionism, and computational neuroscience.
- Computation is defined as processing variables according to a set of rules to produce a specific output.
- Neural activity can be modeled as either digital (discrete) or analog (continuous) computation.
The Three Branches of Computationalism
While all computationalists agree that cognition is a form of computation, they differ significantly on the nature of those computations and the evidence used to support their theories.
Classicism
The classicism tradition suggests that the brain operates via digital computation, functioning in a manner analogous to a digital computer.
Connectionism
Connectionism does not require cognition to be digital. Instead, this branch relies heavily on behavioral evidence to construct models that explain various cognitive phenomena.
Computational Neuroscience
Like connectionism, computational neuroscience does not insist on digital computation. However, it differs by leveraging neuroanatomical and neurophysiological information to create mathematical models of cognition.
Defining Computation in the Brain
To understand how the brain processes information, it is necessary to define computation generally. Computation is the processing of variables or entities according to a set of rules. A rule serves as an instruction to manipulate the current state of a variable to produce a specific output. A computing system is the mechanism organized to execute these rules.
In cognitive science, two primary types of computation are proposed regarding neural activity:
- Digital Computation: This involves operations on strings of digits. Some argue that neural spike trains (the sequence of action potentials) are digital, as a neuron either fires or it does not, effectively acting as a binary 0 or 1.
- Analog Computation: This involves manipulations of non-discrete, continuous variables that change over time. These operations are typically described using systems of differential equations.
[ไม่มีภาพประกอบ]
Comparing Computational Approaches
| Branch | Nature of Computation | Primary Evidence Source |
|---|---|---|
| Classicism | Digital | Digital computing analogies |
| Connectionism | Digital or Analog | Behavioral evidence |
| Computational Neuroscience | Digital or Analog | Neuroanatomical & Neurophysiological data |
The study of neural computation often involves building theoretical models to simulate how the brain works. These insights have served as a significant inspiration for the development of artificial neural networks.
Frequently Asked Questions
What is neural computation?
Neural computation is the information processing performed by networks of neurons in the brain.
Who first proposed that neural activity is computational?
Warren McCulloch and Walter Pitts first proposed this account in their 1943 paper, "A Logical Calculus of the Ideas Immanent in Nervous Activity."
What is the difference between connectionism and computational neuroscience?
While both may accept non-digital computation, connectionists use behavioral evidence to build their models, whereas computational neuroscientists use neuroanatomical and neurophysiological data.
How is digital computation represented in the brain?
It is argued that neural spike trains implement digital computation because the firing of an action potential is a discrete event (either the neuron fires or it does not), similar to binary digits.
What characterizes analog computation in a neural context?
Analog computation involves continuous variables that vary over time, and these processes are typically characterized by systems of differential equations.