Glossary

This is a glossary of terms that I use in the context of my own work on artificial cognition.

Definitions are acyclic, i.e. a definition may contain only links to previously defined terms. This forms a DAG (directed acyclic graph) of terms, with the most foundational terms at the bottom.

system: An instance of the machine we’re building and studying.

environment: Anything that is not part of the system, with which the system exchanges information via input and output ports. Parts of the environment that are inaccessible to other systems constitute the system’s private environment.

world: The union of a system and its environment.

cloning: Copying a full system. Unlike biological cloning, this includes copying all the acquired cognitive skills and therefore the identity of the system.

node: Any kind of persistent element in the system, that transmits information. More generally, it has the usual meaning of vertex in graph theory.

activation: Setting a single node to its active state. A node remains active only for one cycle, after which it is systematically deactivated i.e. switched to its default state. In order to be considered continuously active, a node needs to be activated at each cycle. An active node transmits a unary signal to other nodes or components, while an inactive node doesn’t.

concept: A node that represents a set of states of the world. An active instance of such node.

action: (1) The activation of an output port, which may modify the environment. An action is said internal if other systems cannot sense these changes in the environment. (2) An output port.

abstract: Given two concepts A and B, A is said more abstract than B if the activation of B triggers the activation of A, but the activation of A can take place without B.

fitness: The single numeric variable that changes over time and that the system tries to maximize. At least some of the rules that the system uses to compute the fitness are built in. For example, an external operator would press a green button to add 1 to the fitness at the current cycle or a red button to add -1. The intended effect is to reward or punish the system and encourage certain behaviors.

anticipation: A mechanism by which the system rewards itself based on the expected reward for actions that it just decided to take. A reward refers to the value of the fitness considered to be resulting from a set of actions.

concrete: The opposite of abstract.

context: (1) A concept. (2) The set of all the states of the world which are sufficient to activate a node.

emotion: (1) A built-in parameter of the system which can take a range of values and affects the behavior of the whole system in a predetermined manner. Its value changes over time and can be sensed by the system itself. (2) A specific value of such parameter.

input: Shorthand term for "sensing input", which designates the fixed-length bit vector refreshed at regular intervals that’s available to the system for sensing the environment. This excludes the external reward passed to the system via a special-purpose port and used in the computation of the fitness.

sensing: The activation of input ports.

mind: A system that works as expected.

model: A representation of the world.

optimization: Numeric optimization, usually referring to the maximization of the fitness variable by the system.

output: Fixed-length bit vector updated by the system periodically. A cell in the output vector is called an action. The activation of an output cell is called an action as well. Such activation typically results in a modification of the environment.

pattern: A set of input values identifiable by the system.

perception: The activation of a concept that’s considered similar to a concept of reference in another system or more generally in an external model of the world.

imagination: The ability to activate a concept in either of two contexts, one being more abstract than the other.

intuition: The ability for a system to make decisions without resorting to an emulation of Boolean logic. For design and implementation purposes, imagination, intuition, and perception are considered equivalent.

intelligence: The ability for a system to adapt increasingly faster to new environments.

Avoided terms

We don’t have a good use for these common terms, so we’ll avoid them until we do:

This acyclic glossary and its graph representation were formatted with vanity and Graphviz (tred and dot).