Section 3.7 Glossary

Glossary Glossary

ai triad.

🔗

The three converging components—data, algorithms, and compute—whose simultaneous availability is credited with driving the modern boom in artificial intelligence.

🔗

algorithmic bias.

🔗

A systematic and repeatable error in a computer system that creates unfair outcomes, typically caused by human prejudices embedded in the historical training data.

🔗

artificial general intelligence (agi).

🔗

A theoretical form of AI that possesses human-level adaptability, reasoning, and the ability to apply intelligence to any intellectual task rather than a single specific problem.

🔗

artificial narrow intelligence (ani).

🔗

AI systems designed, trained, and optimized to handle a single, highly specific task, such as playing chess, translating text, or recommending videos.

🔗

autonomous system.

🔗

A physical or software system, such as a self-driving car or delivery drone, that continuously executes a closed loop of sensing, planning, and acting without direct human control.

🔗

black box problem.

🔗

The difficulty of explaining exactly why a deep learning model reached a particular decision, since it may adjust millions of internal numerical parameters rather than follow clear, human-readable logic.

🔗

classification.

🔗

A supervised learning task where the model is trained to sort data inputs into distinct, predefined categories or labels (e.g., identifying a fruit as either an apple or an orange).

🔗

cognitive offloading.

🔗

The practice of relying on an AI tool to perform the foundational mechanics of a task, which can lead to skill atrophy if it replaces the mental struggle needed to build genuine understanding.

🔗

collaborative filtering.

🔗

A recommendation technique that predicts a user’s preferences based on the shared tastes of other users with similar historical behavior.

🔗

computer vision.

🔗

A subfield of AI that gives machines the ability to interpret visual inputs by parsing images as grids of numerical pixel values, enabling tasks like object detection and facial recognition.

🔗

content-based filtering.

🔗

A recommendation technique that suggests items based on the measurable features of items a user has previously engaged with, rather than the behavior of other users.

🔗

data echo (stochastic parrot).

🔗

A failure mode in which a generative model reproduces misconceptions, biases, or outdated beliefs that were common in its training data.

🔗

deep learning.

🔗

A specialized branch of machine learning that uses multi-layer neural networks to learn complex patterns directly from large, raw datasets such as text, audio, or images.

🔗

diffusion model.

🔗

A generative model that learns to create images by reversing a noising process, starting from random noise and progressively denoising it into an output aligned with a prompt.

🔗

features.

🔗

The individual, measurable properties, inputs, or characteristics of a dataset used by a machine learning model to find patterns and make predictions.

🔗

generative ai.

🔗

A category of AI systems that creates novel content, such as text, code, or images, by learning probabilistic patterns from massive datasets of human-created examples.

🔗

hallucination (confabulation).

🔗

A factually incorrect statement, fabricated reference, or invented data point produced confidently by a generative AI model.

🔗

labels.

🔗

The target outcomes, answers, or ground-truth classifications attached to a dataset in supervised learning that the model attempts to predict.

🔗

large language model (llm).

🔗

A generative AI system, such as ChatGPT, that produces text by repeatedly calculating and selecting the most probable next token given a prompt.

🔗

machine learning (ml).

🔗

A subfield of artificial intelligence focused on building systems that learn rules and patterns directly from data, shifting the programming paradigm from "Data + Rules = Answers" to "Data + Answers = Rules."

🔗

natural language processing (nlp).

🔗

A subfield of AI focused on how machines read, decipher, synthesize, and understand human languages, enabling applications like translation and sentiment analysis.

🔗

neural network.

🔗

A layered computational structure, loosely inspired by the brain, in which early layers detect simple features and deeper layers combine them into higher-level concepts during training.

🔗

overfitting.

🔗

A modeling error that occurs when a machine learning algorithm memorizes the specific details and noise of its training data too perfectly. This results in high training accuracy but poor performance on new, unseen data.

🔗

recommendation engine.

🔗

A machine learning system, used by platforms like Netflix or Spotify, that ranks and predicts which items out of a large catalog a user is most likely to engage with.

🔗

reinforcement learning.

🔗

A type of machine learning in which an agent interacts with an environment and learns a strategy to maximize cumulative reward through trial and error, rather than from labeled data.

🔗

socratic dialogue.

🔗

A strategy for using AI as an interactive sparring partner—posing challenging questions or counterarguments—rather than as a source of immediate answers, in order to deepen human understanding.

🔗

specification-driven interaction.

🔗

An approach to using AI in which a human explicitly defines the architectural boundaries, constraints, and requirements of a problem before directing the AI to fill in implementation details.

🔗

supervised learning.

🔗

A type of machine learning where the algorithm is trained on a labeled dataset, meaning every input example is paired with its correct output answer.

🔗

testing set.

🔗

A separate partition of data withheld during the training phase, used exclusively to evaluate how well a machine learning model generalizes to new, unseen data.

🔗

training set.

🔗

The primary portion of a dataset provided to a machine learning model during the learning phase, allowing the algorithm to adjust its internal parameters and discover underlying patterns.

🔗

turing test.

🔗

A thought experiment proposed by Alan Turing in which a machine demonstrates intelligent behavior if a human judge cannot reliably distinguish it from a real human during a free-flowing conversation.

🔗

unsupervised learning.

🔗

A type of machine learning where the algorithm processes unlabeled data, attempting to discover hidden structures, groupings, or clusters entirely on its own.

🔗