cog sci (mod9-12)

markov chain

a system describing a sequence of distinct states, where the probability of each transition depends only on the state that came before

long-distance dependency

scenarios where the current state remains correlated with states far in the past

phrase structure grammar

system of rules that builds sentences by recursively expanding symbols into hierarchical structures (like trees)

recursive rules

rule that embeds one instance of a symbol inside another instance of the same symbol

hierarchical structure

symbols are chunked into larger and larger nested constituents

SHRDLU

a classical AI system that could make plans and carry on conversations about simple world of blocks (3-in-1: syntax, semantics, deduction)

connectionism

movement to model the mind using artificial neural networks

neural network

composed of interconnected “units” which serve as model neurons

biological neurons

brain cells that communicate through electrical and chemical signals, forming networks where knowledge is distributed across connections

artificial neurons

simplified computational models of neurons that take weighted inputs, sum them, and produce output based on an activation function

hebbian learning

a learning rule stating that connections between neurons strengthen when they are activated together (“fire together, wire together”)

perception

the first neural network used for tasks like pattern recognition (limited in what it can learn)

backpropagation

a learning algorithm that allows multi-layer neural networks to adjust weights by propagating error backward (enable more complex learning)

supervised learning

method where a neural network is trained using labeled input-output pairs, adjusting its weights to produce correct outputs

unsupervised learning

where network is given input data without labels and must discover patterns/structure on its own

competition learning

where hidden units compete to respond to an input & only the “winning” unit updates its weights, strengthening its response to similar inputs (example of supervised learning)

parallel processing

ability of neural networks to perform many computations simultaneously across multiple units, enabling fast processing

100-step problem

idea that human cognition must occur in fewer than about 100 sequential steps, motivating parallel processing

distributed representation

information is represented across many units and connections rather than in a single symbol or location

graceful degradation

when part of network is damaged/nasty, performance declines gradually rather than failing completely

one-off learning

ability to learn something from single example (strength of humans, but a weakness of most neural networks)

systematicity

property that the ability to think one thought implies the ability to think related thoughts

radical connectionism

reject classical computing and seek to eliminate the idea of symbolic processing entirely

implementational connectionism

connectionism might be the implementational way to realize classical computational algorithms

concept

building blocks of thought; abstract - concepts stand for an idea/object, but don’t have a genuine resemblance; computational - simple concepts can be combined into complex thoughts

classical theory

concepts are complex mental representations that encode a set of necessary and sufficient conditions; strengths: analytic inferences; weaknesses: fuzziness problem & typicality problem

prototype theory

concepts are complex mental representations that encode conditions that items in their extension tend to have; strengths: fuzziness & typicality effects; weaknesses: missing prototype problem & pet fish problem

conceptual atomism

lexical concepts don’t have any structure; strengths: ignorance and error & conceptual stability; weaknesses: coextension problem & explanatory impotence

fuzziness problem

many concepts appear to be “fuzzy” or inexact

typicality problem

people consider some subcategories more typical or representative of a given concept than others

missing prototype problem

many concepts lack prototypes

pet fish problem

prototypes of complex concepts aren’t generally a function of the prototpes of their constituents

problem of ignorance and error

we can have concepts even if our beliefs are sparse/totally wrong

coextension problem

concepts with same referents can’t be distinguished

jennifer aniston neuron

a neuron that responds selectively to a specific person/concept; found in the medial temporal lobe (MTL); highly specific/abstract representation in brain)

cascading activation

process where activation flows through layers of neurons in a hierarchy; lower-level inputs triggering higher-level (more abstract representation)

theory theory

idea that concepts are embedded in mental structures/theories; meaning comes from how they relate to other concepts within that structured system

schemas

mental structures that help us organize, interpret, and store information about the world and provide a framework for future understanding

assimilation

when new information is modified to fit into existing schemas

accomodation

when an existing schema is modified, or a new schema is created, to fit new information

social cognition

how people think about, understand, and interpret others in social situations

eusociality

a social system where individuals live in large, cooperative groups with organized roles and strong interdependence

Wason selection task

a logical reasoning task used to study how people evaluate “if-then” rules. People often perform poorly on abstract versions but do much better when the task involves social rules or contracts, revealing how reasoning depends on context.

social contract theory

humans evolved domain-specific cognitive processes (a module) for reasoning about social exchanges, specifically rules involving benefits and obligations

cheater-detection algorithm

a set of procedures for figuring out when someone has violated a social contract - when they’ve cheated