Task 3 Flashcards
(30 cards)
What is cognitive architecture?
A theoretical framework that describes the structure of the human mind and provides a computational model for simulating cognitive processes in AI.
What is the central hypothesis of cognitive science?
Thinking is best understood as representational structures in the mind and computational processes that manipulate those structures.
What is CRUM (Computational-Representational Understanding of Mind)?
The dominant approach in cognitive science, proposing that mental representations are like data structures and thought processes are like algorithms.
What are the five criteria for evaluating mental representation theories?
Representational Power – How much information can be expressed.
Computational Power – How efficiently it supports problem-solving, learning, and language.
Psychological Plausibility – Consistency with experimental psychology data.
Neurological Plausibility – Compatibility with neuroscience findings.
Practical Applicability – Usefulness in education, AI, and cognitive modeling.
What are production rules in AI?
If-Then structures (e.g., If X is a student, then X is overworked), used for decision-making and problem-solving in cognitive models.
What is GPS (General Problem Solver)?
One of the first AI systems to use rule-based reasoning for solving human-like problems.
How do rule-based systems handle exceptions?
Unlike formal logic, rule-based systems allow for default rules that can be overridden by more specific conditions.
How do rule-based systems solve problems?
By performing a search in a conceptual space to find a solution, similar to heuristic-based human reasoning.
What is bidirectional search in planning?
A method that combines forward reasoning (from the start) and backward reasoning (from the goal) to find solutions more efficiently.
Why do rule-based systems need heuristics?
Because exhaustive search is computationally impossible for complex problems, heuristics guide problem-solving by prioritizing promising paths.
What are the different ways rules can be learned?
Inductive Generalization – Learning from examples (e.g., If a course is popular, it fills up quickly).
Chunking (SOAR) / Composition (ACT-R) – Combining multiple rules into a single efficient rule.
Specialization – Refining rules to handle specific cases (e.g., If it is Friday and traffic is heavy, don’t drive home).
What is abductive reasoning in rule-based learning?
Backward reasoning to infer possible causes (e.g., If a student is angry and depressed, they might have received a bad grade).
What are the two types of memory in ACT-R?
Declarative Memory – Stores facts (e.g., “Paris is the capital of France”).
Procedural Memory – Stores how-to knowledge (e.g., how to ride a bike).
What is production compilation in ACT-R?
A process where declarative knowledge is transformed into procedural knowledge, making problem-solving more efficient over time.
What is spreading activation in ACT-R?
A mechanism where activation flows between related concepts, strengthening connections and improving recall.
How does ACT-R explain primacy and recency effects in memory?
Primacy Effect – Items at the beginning of a list are rehearsed more, increasing their base-level activation.
Recency Effect – Recently encountered items have high activation due to recent rehearsal.
What is the fan effect in ACT-R?
More associations to a concept lead to slower recall because activation is spread thinly across multiple connections.
What is the difference between propositional logic and first-order logic?
Propositional Logic – Uses true/false statements, but lacks generalization ability.
First-Order Logic – Uses objects, properties, and relationships, allowing general rules (e.g., “All birds can fly”).
What are Bayesian networks?
A probabilistic reasoning model that represents relationships between variables using conditional probabilities.
How do Bayesian networks improve AI reasoning?
They allow efficient computation of probabilities in large, uncertain environments (e.g., diagnosing diseases based on symptoms).
How does deep learning differ from ACT-R?
Deep Learning – Learns patterns from massive datasets using layered neural networks.
ACT-R – Models symbolic reasoning and memory, focusing on human-like cognitive processes.
What is explanation-based learning?
A form of learning where an agent derives general rules by explaining individual experiences (similar to human learning through reasoning).
What is SNIF-ACT?
A model that combines ACT-R with Information Foraging Theory to predict how users browse the web.
What is information scent in SNIF-ACT?
The degree to which text and images on a webpage signal relevant content, guiding users in their search decisions.
