An Attempt At Defining Intelligence

This article tries to forge a definition for intelligence. The background of the starting point for this attempt is computer science. The most important requirement for the definition is the maximization of its applicability.

 The Turing test can be seen as part of an attempt at defining intelligent behavior of humans. It does not try to define intelligent human behavior explicitly, nor does it state that human behavior is not dull. Instead, it defines a way to determine, if something is a human behavior or not. If human behavior is seen as inherently intelligent, then this test classifies a subset of intelligent behaviour. It firstly isolates all important factors of the examined object and secondly strictly states, which properties can be used in order to classify the examined object in question. A downside of this approach is the fact, that not all intelligent human behavior can be classified as such, because biological processes are not considered. The test classifies a certain attribute of behavior based on an implicit definition. It is implied that human in general are able to distinguish between human behavior and other behavior that differs from them in at least one relevant property. The Turing test itself does not predict any concrete properties of the examined thing.

The (algorithmic) complexity of intelligence is an important property in this context, as it seems to be seen by many people. The size and the number of optimized variables alone are not assumed an important concern for the purpose of defining the properties of an intelligent process.

Additionally, in the real world global solutions doesn't seem to be possible as the set of all variables is not know.

Emotional intelligence is not considered to be different to a classical intelligence for the intelligence definition. In this context it is assumed, that emotional intelligence, is intelligence applied to a subset of things.

Simply speaking, the aim of this definition is to determine, if a given process is intelligent or not. In this context, it is assumed that all aspects of a reviewed process are visible, in order to simplify the definition. It is also assumed, that an instance of intelligence is equivalent to an intelligent process. This in turns allows one to describe an data structure for intelligence.

The formal type signature is:

  1. Formally a process w is defined as a string of Σ, where Σ is an non empty alphabet:w ∊ Σ.
  2. An intelligence definition is a classifier of any process w:intelligent(w) = 1 if w is intelligent else 0
    d:toDo
    1. Compare intelligence as event based history vs. intelligence as decision maker in a decision tree vs. intelligence as a string.

The most important requirement to such an implementation is, that it has to be useful. It should not contradict other theories except for other intelligence definition.

Such an implementation shall be useful at least for a nonempty subset of all actors, who can benefit from intelligent process identification. In the ideal case this set should contain all such actors. An actor finds it useful to identify intelligent processes, if a cost free usage of such a definition has benefits for that actor. This implies that actors are optimization systems.

    d:todo
    1. Formalize the maximization of the usefulness of the intelligence classification.

The implementation shall have such a structure, that third parties can be convinced, that the given construct in fact defines intelligence.

The aim of defining intelligence is not entirely objective. Also, there is a formal and objective signature of an intelligence definition, the additional implementation requirements require an already present intelligence definition.

The set of actors who can benefit from intelligent process identification, is also not completely documented in detail as it also implicitly depends on an already present intelligence definition. This in turn could lead to a circular definition.

The fact that such a definition has only to be useful to a subset of actors, who can benefit from intelligent process identification, indicates that there are possibly a multitude of intelligence definitions. This in turn indicates, that an intelligence definition may be partially depended on the observer in question.

The possible relativity of intelligence may be related to the algorithmic complexity and the processing capabilities of the observer and its objects in question. This seems to be compatible to the observation, that it is hard to describe or understand intelligent behavior in detail. It may even explain, why a part of artificial intelligence development seems to be driven by gains in information processing efficiency as access to more capable hardware allows one to execute more complex algorithms.

An intelligence is defined via a classifier as this in accordance to the intelligence definition signature:intelligent(w) = 1 if w is a random optimization else 0.

Randomness is defined in this context, as information/thing that is not accessible from the perspective of the actor. Note that this does not state, if there is indeterminism or not. This also does not state, that everything is deterministic. Neither does it state, why the information is not accessible.

A random optimization is an optimization, where the objective/constraint is random. This means, that an actor is not able to determine the objective of an intelligence completely by definition. This also means, that intelligence is subjective by this definition.

    d:toDo
    1. Subjectivity as an parameter of the definition.

    d:toDo
    1. Check requirments.

    d:toDo
    1. Some things need to be done via intelligent beings. These requirements also define intelligence.

    d:toDo
    1. Define effective intelligence.

    d:toDo
    1. Define competitive intelligence.

    d:toDo
    1. The question of intelligence is irrelevant (i.e. autonomy, free will, politics already provide similar definitions) for most.