Add Definition for Agent

[macterra]

My conceptualization of Agent has evolved somewhat, influenced by our initial discussions. I'm now coming around to the idea that an Agent that acts according to beliefs is just one kind of Agent, a subclass, and that the more general Agent is any System that causes the state of the world to change, distinguished from passive Systems. (System to be defined.)

I think this subclass is what Sever was proposing to distinguish as a "rational agent", but I suggest that is even further down the class hierarchy. I propose to call the kind of Agent that acts according to beliefs (including preferences) an "intentional agent" after Dennett's intentional stance. This would include both rational and irrational agents, depending on the coherence of their internal belief systems.

[SeverTopan]

I like the idea of a distinction between passive and active systems, the key trait of an agent being the presence of a decision that needs to be made.

Just two ideas on how we could structure the distinction:

1. A passive system is an active system where the decision function always evaluates to the same action (e.g. apply gravity). In other words, every system has a set of actions and a decision function, passive systems simply enact the same action every time.
2. An active system is a sub-type of a passive system. where every system has a function that they perform (a 'step function'). An active system enacts a decision function and subsequent actions as its step function. E.g. gravity's step function is to apply gravity, an agent's step function is to choose an action, then perform the chosen action.

[macterra]

"A passive system is an active system"... I thought these were exclusive categories, all systems are either active xor passive.

[SeverTopan]

The above comment served to show how one might be expressed in terms of the other.

Its predicated on the distinction between active and passive agents being the presence of a choice between ways in which world state can be changed. (e.g. gravity has no choice -> passive, humans do -> active).

[SeverTopan]

Just wanted to reference our off-site discussion to rekindle this ticket, @macterra suggested Agency to be a gradient, the higher the complexity of a system's step function, the higher the system's agency. Complexity would be measured using Kolmogorov Complexity.

[macterra]

System: set of related objects

Object: set of related properties

Property: the result of a measurement/sample/observation

[macterra]

I'd like to define agent as a system that fits the definition of philosophical agency meaning something that acts which I think we can equate with changing the state of the world in an intentional way. I'm still struggling with whether a force like gravity can be considered as an agent. I understand that it is convenient to do so in a simulation, but I'm not sure that carries over to reality. My current intuition excludes gravity as an agent.

[macterra]

From the link above..

In social cybernetics:

Autonomous agency is able to embrace the concepts of both the economic agency and the emergent interactive agency. An autonomous system is self-directed, operating in, and being influenced by, interactive environments. It usually has its own immanent dynamics that impact on the way it interacts. It is also adaptable and (hence viable thus having a durable existence), proactive, self-organizing, self-regulating and so forth, participates in creating its own behaviour, and contributes to its life circumstances through cognitive and cultural functionality. Autonomous agency may also be concerned with the relationship between two or more agencies in a mutual relationship with each other and their environments, with imperatives for an agency's behaviour within an interactive context due to immanent emergent attributes.

[SeverTopan]

Reopening given our conversation on 15-10-18; Before we dive into this I'd like to lay the groundwork in #60.

[SeverTopan]

Strawman for the Stages of Agency based on our discussion on 22-10-18. Disclaimer - this is just my interpretation of the ideas that everyone contributed so please correct me if I got anything wrong.

• World Model/Belief System - e.g. the presence of a feedback loop in a control system.
• Valuation - the ability to prefer different World States.
• A Self-Model - in accordance with an awareness loop.
• Counterfactual Depth - the ability to 'turn the knobs' on a World Model and simulate different outcomes.
• Self-Control - the ability to change one's own preferences.

[macterra]

A belief system is the component of a model of an agent used to explain the agent's past behavior and predict the agent's future behavior. It is a model of the agent's own world model. A belief is a related set of rules that assign (intentional) actions to (interpreted) conditions.

A value/preference/desire is a kind of belief, one that (if put into words) asserts that some condition is more valuable/preferable/desirable/better than another condition. The fact that conditions can be compared means they can be measured (quantified).

Maybe we should schedule a special Vulcan meeting to watch Spirituality Explained at SF Mensa with a discussion after.

[macterra]

On self-control I think the ability to change one's own preferences is sufficient but not necessary. Or put another way, that ability is necessary to exceed a particular level of self-control.

New straw man: self-control is a property of agent models, not agents. It is a measure of how predictable a system is given that all information that crosses the system-boundary is known. If you can predict exactly how a system will behave by observing all events then it has zero self-control. I'm trying to capture the intuition that a rock, thermostat, sunflower, spider, mouse, NPC and Vulcan have increasing levels of self-control (freedom of choice, self-determination, free will).