Cognitive Science: AI Methods, Neuroscience, Linguistics, and Thought Experiments

Comprehensive vocabulary flashcards covering AI methods, neuroscience techniques (EEG and fMRI), linguistic relativity and spatial cognition, and philosophy of mind thought experiments.

Supervised Learning

A machine-learning approach in which an algorithm is trained on a dataset containing both predictors (features) and the correct target attribute (classification output) so it can generalize and classify new, unfamiliar cases.

Unsupervised Learning

A machine-learning method in which an algorithm is provided with training data without correct target answers, tasked with discovering internal regularities and patterns on its own.

Information Gain

A statistical metric used by decision-tree algorithms to choose splits, defined as the reduction in uncertainty (entropy) produced when comparing a dataset before and after a split.

Holdout Method

A model-testing technique in which a labeled dataset is partitioned into separate training and testing subsets (frequently near a $60\%$ to $40\%$ split) to evaluate whether a learned model generalizes to unseen data.

Confusion Matrix

A cross-tabulation of actual versus predicted classification labels that specifies the frequency of true positives, true negatives, false positives, and false negatives based on the designated class under consideration.

Thought Experiment

An imagined scenario with defined parameters and an introduced constraint or tension, utilized in philosophy of mind to elicit intuitions or draw conclusions about a conceptual problem.

Experience Machine

Robert Nozick's 1970s thought experiment in which an individual can enter a brain-stimulating machine offering an ideal simulated life from which they cannot leave and cannot tell is simulated, testing whether well-being consists exclusively of subjective pleasure.

Occam's Razor (Simplicity)

A non-empirical virtue of scientific theories stating that when two competing explanations account for a phenomenon equally well, the comparatively simpler explanation should be preferred.

N400

A negative event-related potential component that peaks approximately $400\,\text{ms}$ post-stimulus, elicited by semantically incongruent, unexpected, or low-frequency words, and associated with semantic context integration or cognitive load.

P600

A positive event-related potential component peaking around $600\,\text{ms}$ after stimulus onset, elicited by syntactic violations and linked to structural processing in both language and music.

Functional Connectivity

A statistical relationship reflecting correlations in activity levels between different brain regions in response to a stimulus or task, indicating functional linkage without establishing direct causation.

Effective Connectivity

The directional causal relationships and information flow between brain areas, which cannot be measured directly by imaging tools and must be inferred through statistical modeling.

Voxel

A three-dimensional cubic unit of measurement in fMRI imaging, typically measuring between roughly $0.5\,\text{mm}$ and $3\,\text{mm}$ (up to $4\,\text{mm}$) per side, containing hundreds of thousands of neurons.

Forward Inference

A neuroimaging research strategy based on controlled experimental design, where researchers manipulate a known cognitive process while holding other factors constant to identify the brain regions activated by that process.

Reverse Inference

A flawed logical strategy in neuroimaging that reasons backward from observed activation in a brain area to infer that a specific cognitive process occurred, compromised because individual brain regions support multiple cognitive functions.

Dual-Track Theory

Joshua Greene's moral reasoning framework proposing that personal moral dilemmas (such as the footbridge dilemma) engage an emotional reasoning track, whereas impersonal moral dilemmas (such as the switch trolley dilemma) engage a cognitive reasoning track.

Default Mode Network

A set of interconnected brain regions that becomes active during resting states, daydreaming, and internally directed thought, rather than during focused, externally directed cognitive tasks.

Relative Frame of Reference

A viewer-centered spatial coordinate system that defines the location of an object in terms of the viewer's own body axes and viewpoint, using terms such as 'left,' 'right,' 'front,' or 'behind.'

Intrinsic Frame of Reference

An object-centered spatial coordinate system that defines the location of an item in relation to conventional, inherent parts of the reference object, such as its top, bottom, front, or back.

Absolute Frame of Reference

An environment-centered spatial coordinate system that describes relationships among objects using fixed environmental axes or landmarks, such as 'north,' 'south,' 'uphill,' or 'downstream.'

Acceptability Judgment Task

An experimental linguistics method in which participants view a spatial arrangement and rate how natural or acceptable a descriptive sentence sounds on a Likert-type scale.

Rotation Task

A nonverbal spatial task where participants view an object arrangement on a table, are physically rotated to a new perspective, and must select or reconstruct the arrangement to reveal whether they rely on relative or absolute mental representations.

Sapir–Whorf Hypothesis

The proposition, also known as linguistic relativity, that differences in the syntax and semantics of languages shape the way their speakers perceive, categorize, and conceptualize the world.

Linguistic Determinism

The strong and largely unsupported version of the Sapir–Whorf hypothesis claiming that the specific vocabulary and grammar of a language strictly determine and limit what its speakers are able to think.

10–20 System

A standardized international scalp electrode placement protocol for EEG where letter codes identify anatomical lobes and numbers denote proportional distances of $10\%$ or $20\%$ across cranial landmarks.

Electrocorticography (ECoG)

An electrophysiological recording technique in clinical patients (such as those with epilepsy) in which an electrode grid is placed directly on the cortical surface to record neural activity and deliver electrical stimulation.

Restricted Environmental Stimulation Therapy (REST)

A modern therapeutic approach derived from sensory deprivation research, typically using flotation tanks to isolate sensory input, which produces increased alpha and theta oscillatory brain activity associated with deep relaxation.

Event-Related Potential (ERP)

A scalp-recorded electrical voltage change time-locked to a specific sensory, cognitive, or motor event, generated by the summation of postsynaptic potential dipoles across many similarly oriented neurons.

Oddball Paradigm

A classic ERP experimental design in which participants view a sequence of stimuli consisting of frequent standard events (e.g., $80\%$) intermixed with infrequent target events (e.g., $20\%$) to observe differential component amplitudes and latencies.

P3 Wave (P300)

A positive ERP component with enhanced amplitude to infrequent or unexpected stimuli in an oddball task, variously interpreted as reflecting probability processing, prediction violation, context updating, or attention allocation.

N170

A negative event-related potential component that peaks at approximately $170\,\text{ms}$ following visual stimulus onset, which shows larger amplitude to human faces and is debated as reflecting either a face-specific mechanism or visual expertise.

Symbolic AI (GOFAI)

Good old-fashioned artificial intelligence; a paradigm within cognitive science that models thought using explicit, rule-based representations, including propositions, equations, and logic statements.

Production Rule

A modular knowledge representation in symbolic cognitive simulations composed of an 'if' condition specifying triggering criteria and a 'then' action that executes when the rule fires to pursue a specific goal.

Protocol Analysis

An empirical method used to design cognitive simulations, often taking the form of a 'think-aloud' protocol where participants verbalize their reasoning steps as they solve a problem.

Subsymbolic Methods

Artificial intelligence approaches, including connectionist artificial neural networks, where information is encoded not in explicit symbolic rules, but in distributed numerical patterns across weighted connections between units.

Backpropagation of Error

A supervised training algorithm for artificial neural networks where the discrepancy between the network's generated output and the correct target label is transmitted backward through the layers to adjust connection weights.

Embodied AI Methods

A robotics and artificial intelligence framework that prioritizes bodily morphology, physical sensors, and direct environmental interaction to reduce the computational demands placed on a central processor.

Morphological Computation

The principle developed by Rolf Pfeifer stating that the physical structure, materials, and biomechanics of a body perform operations and stabilize actions that would otherwise have to be computed by the brain or central processor.

Passive Dynamic Walker

A simple legged mechanical device that maintains a stable walking pattern down a gentle incline using only an initial push, gravity, and the physical morphology of its joints, without any active central processor calculating steps.