Lecture 4 - Data & Misinformation

External Storage

The function of a visual display to store information outside the human mind, freeing up working memory resources for other cognitive tasks.

Organization of Information

The grouping and spatial indexing of related data points (e.g., via proximity or Gestalt principles) to facilitate search, comparison, and mental integration.

Offloading Cognition onto Perception

Mapping non-visual data onto visual variables so that complex mathematical computations are replaced by automatic visual pattern recognition (e.g., detecting emergent features).

Offloading Cognition onto Action

Using physical interactions with a display (e.g., dragging, rotating, filtering, zooming) to perform work in the external world that would otherwise require effortful internal mental calculations

Iconic Display

A visual-spatial display that represents objects that are themselves visual-spatial entities (e.g., road maps, machinery diagrams).

Relational Display

A visual-spatial display that represents abstract relationships or non-spatial entities using visual variables like location, color, or shape as metaphors (e.g., scatterplots, line graphs).

Display Schema

A mental framework derived from prior experience that structures how a viewer expects a visual display to look and behave (e.g., expecting a bar chart’s Y-axis to begin at zero).

Internal Representation

The mental reconstruction—not an exact copy—of a visual display constructed in memory, shaped heavily by attention, perceptual encoding, and the viewer’s display schema.

Data-Ink Ratio

Edward Tufte’s core principle stating that graphics should maximize the proportion of ink used to convey actual data relative to the total ink used; any ink that does not convey relevant information is wasted

Chart Junk

Decorative or non-essential visual elements added to a graph that distract the viewer or grab attention without conveying data relevance.

Relevance Principle

The rule that a graphic should present no more and no less information than is required for the specific task at hand.

Apprehension Principle

The rule that a display can only be effective if its visual features can be accurately and easily perceived by the visual system.

Salience Principle

The design requirement to make the most important thematic information visually prominent so that it immediately draws attentional focus.

Capacity Limitation Principle

The mandate to design displays that respect human constraints in working memory and visual attention.

Naïve Realism

The misguided belief or preference that visual displays should be as realistic, detailed, or 3D as possible, despite evidence that realistic displays often degrade user performance due to visual clutter or depth misperception.

Lie Factor

A value far from $1.0$ indicates that the graph visually exaggerates or understates the true change in data.

Y-Axis Truncation

Starting the vertical axis above zero, which visually exaggerates minor relative differences between data points.

Area-as-Quantity Distortion

Representing a one-dimensional linear variable using a two-dimensional shape or area, causing ratios to be visually exaggerated because area grows exponentially relative to linear scale.

Aspect Ratio (X-Axis) Distortion

Stretching or compressing the horizontal axis to artificially steepen or flatten a trendline.

Inverted Axis Distortion

Flipping an axis (e.g., placing $0$ at the top and $1000$ at the bottom) so that upward trends visually appear to go downward, often used for message reversal.

Graphical Liar Tactics

Sender-driven manipulation strategies, including: (1) hiding relevant data, (2) displaying excessive data to obscure the true signal, or (3) using truncated/distorted axes.

Inoculation Effect (Graph Literacy)

The psychological protection achieved when warning viewers in advance about potential graph distortions (e.g., Y-axis truncation), which significantly reduces the distortion effect—though its protection can fade over time

Deepfake

Media (image, video, or audio) that has undergone algorithmic manipulation using deep learning techniques

Cheapfake

A non-AI media manipulation created using simple editing tools, such as splicing, changing playback speeds, or mislabeling contexts.

CGI (Computer-Generated Imagery)

Manually rendered digital imagery that does not rely on deep-learning neural network algorithms.

CNN (Convolutional Neural Network)

A specialized form of artificial neural network used in computer vision to serve as either a detector (recognizing features) or a generator (creating media).

Autoencoder

A deep-learning architecture consisting of an encoder (compressing facial features) and a decoder (reconstructing them); linking an encoder trained on Person A with a decoder trained on Person B allows face-swapping.

GAN (Generative Adversarial Network)

A dual-network deep-learning framework (generator vs. discriminator) capable of generating hyper-realistic synthetic human faces, voices, or images from scratch using limited initial data.

Model Collapse

A degenerative process wherein AI models trained recursively on synthetic AI-generated data gradually lose detail and variance over generations, causing outputs to deteriorate into noise.