Thursday, April 27, 2017

Performance-Based Cognitive Architecture

Performance-Based Cognitive Architecture
a visual take

Previously, we’ve used CLARION to talk about top-down and bottom-up types of learning. I’d like to expand on cognitive-architecture today, as it can be very helpful to visualize mental processes and can suggest interesting ideas.

Upon review, CLARION (to my understanding used chiefly in AI and simulation) is not especially well-suited for talking about performance. For this reason I have made some substantial edits to better suit our subject matter. It should be noted that this new model, like any cognitive architecture, is not remotely close to literally true. Cognitive science is far too underdeveloped. The goal is not so much reproduction of the complex system as it is to sufficiently mimic paths as an aid to understanding our mental experiences.


Sense Experience is input.
Perceptions and sense experiences elicit responses in the Drivers Subsystem as well as the Meta-Cognitive Subsystem, even if we aren’t consciously aware of them. These could be sounds, images, smells, touches, conditions, states, etc.


Drivers Subsystem
The Drivers Subsystem is designed to initiate or facilitate change or action. They suggest our current needs to the the rest of the cognitive system. It contains a natural hierarchy of

Emotional Drivers
Emotions are chemical flags that appear in an effort to provide us with unmistakable status updates. If left to accumulate they are specifically designed to shut down and bypass meta-cognitive function (“overwhelm logic and reason”) in order to control our actions. Emotional drives include motivation/inspiration, overconfidence, perfectionism and a number of other feelings that can dictate behavior in addition the more obvious fear, anger, boredom, uncertainty, shame, lust, etc. For the sake of relevancy I didn’t want to include it as a separate function, but animal instincts would function like a stronger emotional driver.

Habitual/Structural Drivers
Habitual or structural expectations suggest what should be done and massively reduce the amount of effort needed to commit to action. These are useful because they are independent of how we are feeling emotionally or located mentally at the moment. They are quick to form and difficult to break.

Value Drivers
Value drivers act like emotional drivers but are less fickle. Our values change, but not so rapidly or unpredictably. Orienting our behavior around what is important to us on the long-term is a principled process rather than a chemical one. For this reason it has less oomph as a driver but is normally of greater importance in a modern context.

Each of these driving functions feeds into the greater Meta-Cognitive Subsystem but can feed directly into the Action Subsystem if the Meta-Cognitive process is weak, preoccupied, or overwhelmed.

Meta-Cognitive Subsystem
The Meta-Cognitive Subsystem directs and negotiates cognitive processes involved with learning and behavior, which is to say that granted force, the meta-cognitive process directs and empowers cognition itself.  It includes

Reinforcement. When we do or perceive something, there is an internal mechanism that instantly (and largely unconsciously) judges the outcome. A quick yes or no. A yes reenforces neural pathways and thus feeds right into our habits (procedural knowledge) as well as experiential, intuitive knowledge. A no quickly checks and refines neural paths. Reinforcement is a powerful tool that functions best with instant and unambiguous feedback.

Goal-Setting translates our drives into specific wants and develops plans to fulfill them. These plans may be short or long-term but are formulated as priorities for our focus and values in the future.

Thoughts (Stream of Consciousness) are located on the outside edge of the MCS. While it has enormous influence over meta-cognition, it is important to recognize that the popular view of stream of consciousness thought as the seat of consciousness if not identity is wildly inaccurate. It is in fact just a small, non-central part of a subsystem within a system. Thoughts are an instrument used to better navigate and negotiate the complex intermingling of knowledge, drives, and sensations within meta-cognition. It is easy to let it overwhelm and stall the rest of the cognitive process in a similar way that unchecked emotional drives can be overwhelming. With practice (most accessibly via meditation), identifying thoughts as simple internal states not unlike sense experiences comes more easily.

Focus
is an additive function that allows for greater efficacy within. It does not exclude what lies outside. Focus may wander but can be deliberately held in place with effort. If trained, Focus, in tandem with Goal-Setting, may dictate behavior independently of a driver. Additionally, it is the only means to learning beyond rote repetition/reinforcement. Focus is uniquely widened and actualized during Flowstate.


Knowledge Subsystem
The Knowledge Subsystem operates as a two-level database containing vast quantities of learned information.

Declarative knowledge
is specific, formulaic, and can be easily communicated in verbal or representable form. It includes all named concepts or procedural rules. If you can explain it to another person without metaphor then it’s declarative.

Intuitive knowledge is holistic, fundamental, and outside of our verbal grasp. It is knowledge that has been experientially gathered, then pieced together and solved unconsciously. It comes in the form of hunches, notions, and insights just beyond the edge of our conscious understanding.


Behavioral Subsystem
The Behavioral Subsystem operates on two levels and initiates action.

Deliberate behavior is purposeful, measured, and demanded by our decision-making.

Automatic behavior is not quite involuntary but outside of our deliberate control. This may be because it is driven by declarative, procedural knowledge (knowledge that is practiced to the degree of unconscious competence) or because it is driven by intuitive knowledge that we aren’t ready to acknowledge verbally. Automatic behavior driven by intuitive knowledge is precious and a hallmark of growth.

Action is output.


Notable Implications:
• Each function is like a skill in that it can be weak through neglect or improved in its use with concentrated effort. Similarly, they can be “re-programmed.”
• Drivers can be helpful or harmful but are not enough on their own to determine behavior UNLESS we are not mentally present or able.
• Drivers are not actually necessary to commit to action. You can draw a line from experience to an action without passing through any drivers. In this way they are like stimulants. They lend some quick and easy strength to our decision-making but you can achieve the same effect or even overcome them with trained mental-muscle.
• Thought is not central. In this model focus is the most important function.
• Learning is ONLY possible through focus and reinforcement.
• Deliberate action is ONLY possible through focus.

No comments:

Post a Comment