The Modelling Agent Model

20 Mar 2022

Agent

An Agent exists inside an Environment. The Agent can not read or manipulate this Environment directly.
However, the Agent can sense the Environment through their Senses, giving them an Image of it, and interact with the Environment through Actions, which in turn have an Effect, transforming the Environment.
Through a feedback loop of sensing and acting, the Agent pursues their Goals. What mechanism stands between an Agent’s Senses, Goals and Actions, is the Agent’s Mind.

Hard-Wired Agent

Perhaps the simplest possible non-trivial Agent, is one that, for any given Image of the world, executes a fixed predetermined Action. The Agent’s success or failure in achieving their goals is determined by the Mind they come into existence with and the Environment they find themselves in.

Soft-Wired Agent

While still having a fixed Action for a given Image of the Environment, an Agent could have a mechanism for evaluating how it’s Actions have served towards achieving it’s Goals, which instruct a direct adaption of it’s Mind.

Modelling Agent

With an Agent’s inability to directly read the Environment, instead of using the Image their Senses take of it directly in determining their Actions, an Agent could construct an immediate representation, a Model on which to base their decisions, which is constructed and adapted through their Image of the Environment.

Model

In having a Model of the Environment, the Agent can consider Actions as if they had the ability to directly read and manipulate the Environment, with only the accuracy of the Model their decisions are based upon being restricted by the representativeness of the Image their Senses produce and their Mind’s capacity.
In having a representation of the Environment that the Agent can directly read and manipulate, the Agent is capable of taking Actions as if they were capable of reading the Environment itself.
An inaccurate direct Model of the Environment is stronger than an accurate indirect Image of the Environment in deciding Goal-achieving Actions.
This is due to the Model not representing the Senses’ Image itself directly, but merely having to be consistent with it. Additionally, through any structure inherent in the Modelling-System of the Agent’s Mind, the accuracy of the model to the Environment can be greater than permitted by the total complexity of the Senses of the Agent, by restraining the possible interpretations of their Image.

Natural Intelligence

While the human mind is not cleanly approachable as a Modelling Agent Mind, having partially retained both Hard-Wired and Soft-Wired processes, applying the Modelling Agent Model to Human behaviour still is of value in understanding the capabilities and limitations of humans.

Tools

Causing through an Action an interaction between parts of the Environment to achieve an Effect that the Agent could not achieve directly through an Action, while commonly considered a signifier of “intelligence”, does not necessarily require for an Agent to base their decisions upon a Model of the Environment.
However, a Modelling Agent, being capable of considering an Action’s Effect upon the Environment by considering the Action in the Model of the Environment without having to do the Action in the real Environment, can use tools much more spontaneously and flexibly than a Soft-Wired Agent of equivalent capacity, who would have to learn the use of tools through trial and error in the real Environment.

Social Behaviour

If the Environment an Agent finds themselves in is generally consistent in variation and complexity over space and time, through an evolutionary process, a Soft-Wired Agent, or even a Hard-Wired Agent, can be converged towards possessing a Mind well adapted to interacting with the Environment in a manner that furthers their Goals. However, if a part of the Environment is highly unpredictable, neither type of Agent will perform well.
In particular in the case of other complex Agents being part of the Environment, a Modelling Agent, capable of constructing a Model of the Minds of other Agents, can reason about other Agent’s behaviour, given the Image these other Agents would have of the Environment, which in turn is informed by the Actions of the Agent themselves.

Self

While the Actor themselves as part of their Environment has to be considered in some form in every Mind, a Modelling Agent in particular, can develop a model of their Self, their own physical presence in their Environment, and any internal influences upon their decisions.

Communication

Artificial Intelligence

With the presently commonly applied machine learning architectures falling largely under the category of a Soft-Wired Agent, it would be of interest to consider whether a Modelling Agent could be similarly implemented and used to further the capabilities of artificial intelligent systems.