Lessons from the Edge

“Cellular Automata, Collective Intelligence, and Phase Transition”
In which Bill Sulis discussed brain functioning from a complexity perspective
William Sulis, MD, PhD, McMaster University, Departments of Psychiatry and Psychology

• Cellular Automata help us examine how the mind operates. “The bulk of my problems are in treating demented elderly,” Sulis explained. “Those problems are not just for my patients but for the systems they impact. They create a new social milieu. The hallmark of dementia is an erosion of conscious processes. People suffering dementia operate mostly unconsciously. I look for dynamics that would facilitate more controlled behavior.”
  
  
• Classical Cellular Automata (CA) are stupid. They look at their state and the state of their neighbors, then follow directions. CA generate patterns. They can generate many patterns. Some are boringly linear. Some are periodic, even complex, or apparently random.
  
  
• Sulis wanted to know what we can say about the dynamics behind the pattern. “On the surface, behavior seems regular. Yet we don’t know what is driving that behavior. What’s important is understanding the patterns as a whole. You need to understand the right set of patterns to solve your problems.”

• How do our brains solve problems? We don’t have a single cell that recognizes “grandma.” Are there assemblies of cells that enable us to recognize things? There’s been some evidence of this.
  
  
• Present a monkey with the same stimulus a couple of hundred times, and we find that the experience records differently from time to time. Only the average firing rate offers clear information.
  
  
• The system of the brain is extremely context dependent. A rat’s hyppocampus fires similarly every time. If you keep the rat in the same environment his responses remain alike. But after being in a new environment, the pattern of firing has changed.
  
  
• How, Sulis asked, does the brain understand change has occurred so it can interpret these differences in firing patterns?
  
  
• Living things must respond stably in their environments. A deer in the woods must be able to respond to a predator. What’s important is not what the neurons are doing, but what the body as a whole does. CA reflect this ability to produce different patterns from different initial states.

• Collective intelligence is adaptive behavior generated by many quasi-independent agents, interacting locally.
  
  
• Social insects colonies, for example, fall under causal influences: stochastic determinism; interactive determinism. Their decisions are very democratic. Movement to a new location appears random, but it’s not.
  
  
• Without hierarchical structures, you should begin to see collective behaviors emerge. The big problem is that agents can only access local information. That enables them to be trapped in blind alleys. The environment must be shaped to avoid dead ends and ineffective behavior. Given novel environments, collective intelligence enables people to build remarkable structures.

• Consider the phenomenon of emergence. Low level systems, whether ants or CA, examined locally don’t demonstrate much order. Only when viewed as a system do significant patterns emerge. We can witness cells forming patterns, and those patterns forming meta-patterns depending on their environments.
  
  
• To understand a system, you must understand the patterns it values. Each system decides which patterns are important to it.
  
  
• For organizations, the implications are striking. On top of the dynamics we acknowledge, a set of dynamics occurs because of the nature of the interaction of human parts. Those interactions cause a group dynamic that interacts with the environment. Imposing demands on that system can be self-destructive if they work counter to the inherent evolved dynamics.
  
  
• This is reflected in CA’s, where you can’t understand the behavior of the cell except in the context of its game; and you can’t understand the behavior of the game except in the larger context of which it is a part.
  
  
• Without recognizing these dynamics, it’s easy to ask a system to do things it simply is not capable of doing.
  
  
• With too much control, an organization reduces the opportunities for adaptive action; with too little, it will make foolish mistakes.

Q:    Would it be correct to say that systems with built-in randomness are more stable
than systems without?
Ary: Systems with the ability to respond to random signals have an intrinsic ability to
remain more stable. To maintain stability the system has innate variability. Equilibrium is the brink of extinction.
Bill: There are situations in which low levels of noise facilitate better performance. In CA, the presence of noise allows for a high level of recognition. Without noise, systems can lock into a response. Noise keeps that from happening.

Q: Can you sketch out guidelines to healthy level of noise? Or is it context dependent?
Ary: That’s difficult to say. Until recently, no one recognized the value of noise. How to use noise is experimental. Modeling it is extremely difficult. You can experiment to find out. Epilepsy control might include the introduction of some noise before the brain gets locked into this response. This is an example of complex issues about questions that would not have been asked before.

Copyright © 1999, VHA Inc. Permission
to copy for educational purposes only.