Models and Architecuture

3 Agent Architectures

Agent architectures illustrate the different components that make up an agent and

how those components are organized

What is an agent?

• An agent is a system that perceives its environment through sensory systems of some type and acts upon that environment through effector systems.

Three main types of agents

1. Simple reflex agent

2. Goal-based agent

3. Learning agent

Simple reflex agent “if this, then that”

Goal-based agent

Learning agent

Fodor on the Modularity of Mind

Modular and Nonmodular Processes

• Nonmodular (or central) processes: High-level and open-ended cognitive processes that involve integrating a wide range of information to address general, complex problems.

  • Central processes: The basic representations in central processing are personal level states – propositional attitudes and perceptions.

  • Quinean (sensitivity to global properties of system)

    • Organism's belief system likened to a scientific theory

    • Evaluation of the belief system as a whole for consistency and coherence

    • Interdependence of individual beliefs within the system

  • Isotropic (informational unencapsulation)

    • Lack of informational encapsulation

    • Relevance of any part of the belief system to confirm or disconfirm any other part

• Modular processes: Lower-level cognitive processes that operate quickly to provide rapid solutions to specific, well-defined problems.

  • Characteristics of Modular Processes

    • Domain specificity: Modules are highly specialized mechanisms designed to carry out specific and circumscribed information-processing tasks. They operate on a limited range of inputs (those relevant to their particular domain).

    • Informational encapsulation: Modular processing is not affected by what is going on elsewhere in the mind.

    • Mandatory application: Cognitive modules respond automatically to stimuli. They are not under any executive control and cannot be "switched off."

    • Speed: Modular processing transforms input (e.g., patterns of intensity values picked up by photoreceptors in the retina) into output (e.g., representations of three-dimensional objects) quickly and efficiently.

      • Additional Features that Sometimes are Characteristics of Modular Processes

        • Fixed neural architecture: It is sometimes possible to identify determinate regions of the brain associated with particular types of modular processing (e.g., an area in the fusiform gyrus is specialized for face recognition).

        • Specific breakdown patterns: Modular processing can fail in highly determinate ways. These breakdowns can provide clues as to the form and structure of that processing (e.g., prosopagnosia is a highly specific neuropsychological disorder that affects face-recognition abilities but not object recognition more generally).

Fodor’s Modular

Dedicated processing systems that are

• Domain-specific

• cognitively impenetrable

• Mandatory

• fast

Possibly have

• fixed neural architecture

• specific breakdown patterns

Examples

• Color perception

• Shape analysis

• Analysis of three-dimensional spatial relations

• Visual guidance of bodily motions

• Grammatical analysis of heard utterances

• Detection of the melodic or rhythmic structure of acoustic arrays

• Recognition of the voices of conspecifics (same species)

The Massive Modularity Hypothesis

The human mind is a collection of specialized modules, each of which have evolved to solve a very specific set of problems that were confronted by our early ancestors.

Examples:

Called Darwininian something. We are social animals

• Cheater detection

  • When the selection task is framed in terms of permissions and entitlements it engages the cheater detection module

• Folk psychology

• Kin selection

• Intuitive physics

• Number sense

How Did Cooperative Behavior Emerge?

Prisoner’s dilemma

• Many real-life interactions resemble indefinitely iterated prisoner's dilemmas.

Heuristic Strategy: TIT FOR TAT

• Involves two rules:

1. always cooperate first,

2. mimic opponent's previous move.

• It's effective due to simplicity and stability in evolutionary game theory.

Role of Cheater Detection Module:

• TIT FOR TAT requires identifying cooperation vs. defection.

• Evolution favored a specialized module for detecting cheaters and free riders.

The Cheater-Detection module helps in navigating social interactions by recognizing

breaches of conditional obligations.

• It explains our proficiency in reasoning about rules and obligations compared to abstract conditional reasoning.

Hybrid Architectures

Two approaches to mental architectures

1. Physical symbol systems

2. Artificial neural networks

John Anderson’s ACT-R cognitive architecture attempts to integrate both approaches.

ACT-R is modular

• Cognitive modules can only access sensory information through buffers

ACT-R (Adaptive Control of Thought—Rational)

ACT-R is modular

• The perceptual-motor layer

• Cognitive modules can only access sensory information through buffers

Two types of cognitive modules: declarative and

procedural

• Declarative (knowledge that): e.g., knowing that Paris is the capital of France

  • achieved through chunking

• Procedural (knowledge how): e.g., knowing how to speak French

  • achieved through production rules

What makes it a hybrid architecture?

There is no central processor

Decisions are made subsymbolically, prior to cognitive modular processing

• Serial processing: only one production rule can be active at a time

• Pattern-matching module chooses which production rule to be active

• Pattern-matching module chooses according to utility – which production rule will achieve the system’s goals most efficiently

Subsymbolic processing

Each declarative chunk and each production rule is symbolic, but the information that determines their utility is subsymbolic.

• Subsymbolic processes determine buffer placement and activation level, which determine whether a chunk or rule is used.

Take Home from ACT-R

First, debates about the organization of the mind are closely connected to debates

about the nature of information processing. Thinking properly about the modular

organization of the mind requires thinking about how the different modules might

execute their information-processing tasks.

Second, the different parts of a mental architecture might exploit different models of

information processing. Some tasks lend themselves to a symbolic approach, others to

a subsymbolic approach. The debate between models of information processing is not

an all-or-nothing case.