Artificially Intelligent Language

How can a labeling system be applied for programming an Artificial General Intelligence (AGI) agent?

Since language seems so significant and beneficial for human intelligence (as I described in my previous post; Intelligence Evolution by Language), it would likely be an effective application in the concept of programming an AGI. Throughout any person's life, learning to be generally intelligent of their surroundings, and their own existence would likely be very difficult without language. A labelling system allows shortcut references for communication to quickly transfer information from 1 individual to another, but also aids in individual thought and understanding. As an individual consciously considers any situation they’re in, they have those shortcut references to use in their thought processes, to distinctly and quickly think about various factors within the situation, by referring to their saved memory link, of the label applied to those factors. Without the labels, consciously analyzing any situation would be very difficult, as the individual would have to think of many memories of each factor involved, and think of many instances of those factors and the interaction of that factor.

To apply this concept of a labelling system into the programming of AGI, it would have to be made with the ability to save shortcut references in its memory. This could be done with common human language, which might also be effective for humans teaching it, by communicating through normal language, but could potentially also be done with the AGI’s own unique language.

Using English for eg, the AGI would somehow have to be programmed to relate sight (and other senses) of objects and actions, with the verbal sound and sight of the spelling, of each label applied to that object. This could likely potentially be done through the process of machine learning, which is used today, by feeding it enough examples of a factor, and giving it a label to go with the virtual boundaries which are included for the definition of that label. For it to be intelligent to a significant degree of generality, it should be able to learn new labels, through saved memory of its new experiences, as it encounters new situations and new factors. If it could continue to learn new labels, as it encounters new factors, it could gain more and more shortcut references in memory, saved as a link with those factors to which the label applies.

If it additionally, has the ability to access memory of 1 labelled factor, and that factors interaction with another labelled factor, it could then use the ability of conscious comprehension. It seems that accessing memory of an interaction between 2 factors, requires the memory of action of the factors. Memory of action seems to require a saved memory of a time period, where 1 factor had enough time to change. Without the minimum time period for a factor to change, there is no interaction of any factor with another. For eg, to comprehend the interaction of 1+1=2, physically, 1 new object has to be added, which requires a time period for the additional “1” object to be moved or created. To comprehend how a piece of paper interacts with fire, it requires memory of the minimum time period for fire to burn the paper.

But if this ability is used, to access memory of a time period where 1 labelled factor interacts with another, this unlocks a whole new exponential potential for comprehension, by means of saving new labels, applied to those interactions. With this ability, an AGI could access memory of a labelled factor, then access memory of another labelled factor, and memory of how those 2 factors interact. This allows the ability of conscious comprehension of how any 2 factors relate, through cause and effect, and potentially the comprehension of how anything and everything in this world interacts and functions.

For eg, if a newly learning AGI came across a bunch of dogs, and saved the label of “dog” for the boundaries of that which is included in the definition of “dog”, then did the same for “cats”, then saved the label of “chasing”, as it applies to our definition, it could comprehend that dogs chase cats. The memory of the time period required for “chasing”, would be the interaction. After simultaneously accessing memories of these 2 factors (cats & dogs) + their interaction (chasing), the AGI could save a new memory of that interaction. This new memory, could be used as a reference in the future, for new situations which involve any of those 3 factors (including the interaction). This would help it more quickly comprehend new factors interactions (for eg, wolves chasing deer).

As significant as a labelling language seems to be, for the development and active use of human general intelligence, it could be almost as significant in creating AGI. Besides the advantage of transferring information quickly, via communication of labels, labels also seem to benefit the allowance of comprehension involved in general intelligence and learning about functions of this world.