Top-Down vs. Bottom-Up Processing
Most complex systems with hierarchical structure require a coordination of information or energy flow up and down the structure. When upper levels (or big parts) influence lower levels (or components) this is called top-down processing. When lower levels influence higher levels, this is called bottom-up processing.
For example, in an army, a general might benefit from a valuable piece of information passed on by a low-level intelligence officer. If that information alters the behavior of the general, then the upper level of the hierarchy has been influenced by a lower-level element. Obviously this provides an advantage, as long as the information is screened for relevance and importance.
A coordination of top-down and bottom-up processing is necessary for optimal outcomes in most systems. Joseph Tainter suggested in The Collapse of Complex Societies (1988) that civilizations fail when "the flow of information from the bottom to the top is...inaccurate and ineffective."
When people feel forgotten, unserved, or abused by their government, the society may disintegrate. When people have a sense of controlling their own destiny, they tend to be happier. A skillful mix of top-down and bottom-up processing is most effective.
In cognitive science, a mixture of top-down and bottom-up processing is necessary for cognitive productions. Reading, for example, involves a combination of influences from high-level concepts or pre-existing knowledge (top-down processing) with influences from individual letters and words (bottom-up processing).
Consider the following example from Neisser (1967, adapted from Selfridge, 1955).
Top-down or schema-driven processing leads us to interpret the same shape two different ways.
Bottom-up processing occurs as the reader combines letters into words. The data from smaller components (letters) guides perception of a larger whole (a word).
Top-down processing occurs when knowledge of words guides interpretation of individual letters. Knowledge of words (the and cat) leads us to see the same component first as an h, then as an a.
In cognitive science, a synonym for bottom-up is data-driven. A synonym for top-down in cognitive science is schema-driven. A schema is a piece of organized knowledge formed by repeated experiences.
Schemas (aka schemata) influence the interpretation of incoming data. Most people have repeatedly encountered the word cat many times, so they have that schema in their heads. This leads them to expect the letter A in between the C and the T in Selfridge's example.
In general, a delicate interplay of bottom-up and top-down processing is required for reading, object recognition, problem solving, memory, and all other cognitive activities.
The principle of top-down and bottom-up processing: Complex systems produce optimal outcomes through interaction of top-down and bottom-up processing.
What if there is no up or down in a control system, just a network of influences connecting all the parts? That descries a type of structure called a heterarchy.
Heterarchy is a made-up word from the artificial intelligence lab at MIT in the 1960s, combining hetero- (different) and -archy (rulers). A heterarchy can be described as a community of experts.
Instead of one boss, there are many bosses, and they all cooperate. In a pure heterarchy, there is no member above the others.
Real world organizations usually have a chief or chairperson for administrative purposes. Below that, power and influence are spread among a group of experts, approximating a heterarchy.
The Supreme Court is a heterarchy in the American political system. There is a chief (the Chief Justice) but all the justices have an equal vote, and each justice contributes to specific cases depending upon his or her specialty. When it comes to a vote, each justice including the Chief Justice has the same power: one vote.
Experts in artificial intelligence suggest the brain is organized as a heterarchy. Executive processes including the conscious self coordinate the activity of brain subcenters, but skills are performed by expert areas or modules throughout the brain.
AI expert Marvin Minsky of MIT wrote an influential book titled Society of Mind (1988) to convey the idea that cognitive networks are organized like societies. Many components interact to form the whole, and each component is semi-
Specialized areas of the brain enable skills like speaking, analyzing spatial relationships, and coordinating movement. When attention is distracted or a person is asleep, the semi-independent modules continue to process information and influence the system as a whole. The control scheme is more like a heterarchy than a strict hierarchy.
Waldrop (1984) described an optimal arrangement in human organizations: "Put an overall structure in place to impose long-term coherence, and then give each node (expert) a certain range of responsibility within which it can make its own decisions." That is the way academic departments work, for example, and perhaps a good description of how the brain works as well.
In a control hierarchy, top-down and bottom-up processes interact to produce optimal behavior. In a system with a heterarchy or no discernible hierarchy, interactions might be spread between any two components of the system. There is no top or bottom.
Network theoreticians calculate the importance of an element or node in such a network by counting the number of its connections. This is called a node's degree.
A node with more connections is said to have a higher degree. That has consequences in a variety of situations interesting to computer scientists and others working with complex systems.
In a social network, people with more connections have a higher degree of influence. What they say, or images they post, has influence on more people. Having a higher degree is analogous to being at the top of a hierarchy.
Discussion of nodes and connections takes us to the next topic. Together, nodes and connections define networks.
Minsky, M. (1988). The Society of Mind. New York: Simon & Schuster.
Tainter, J. A. (2003/1988). The Collapse of Complex Societies. Cambridge, UK: Cambridge University Press.
Waldrop, M. M. (1984) The Intelligence of Organizations. Science, 225, 1136-1137.
Write to Dr. Dewey at email@example.com.
Copyright © 2017 Russ Dewey