Observations of Society

Knowledge Tends to Increase over Time

Note: in order to run the simulation referred to in this slide, go here, to the Java Applet version. You will be directed to download the latest version of the Java plug-in.

How do we measure improvement? We can measure the degree to which the code corresponds to the environment as well as the average degree to which agent beliefs correspond to the environment.(1) We call these measures the code accuracy and the average agent accuracy.

The society to the left has been created with an initial set of agents, a code, and an environment. Press "Go" to run the society forward through time and see that the code accuracy and average agent accuracy improve in the graph at the bottom of the screen. Code accuracy appears in red. Average agent accuracy in blue.

Note also that the agent accuracy is consistently lower than the code accuracy. In fact, this remains true under a wide variety of conditions.

Each of the learning steps acts to bring agent beliefs and the code beliefs closer. Thus, after a time, an equilibrium is reached in which agents and the code share the same belief vectors. At this point, no more change can occur.

1. More specifically, we define the accuracy of a belief vector as its dot product with the environment vector, divided by the number of components, M. Thus, accuracy can range from -1 to +1.

Previous Slide Next Slide