Chapter 18: Artificial Intelligence and Generative AI

Comprehensive vocabulary definitions covering modern AI concepts, development processes, risks, and industry applications as discussed in Chapter 18.

Artificial Intelligence (AI)

Computer software that can mimic or improve upon functions that would otherwise require human intelligence.

Generative AI (GenAI)

A type of AI where a user provides a request—usually via text—and the AI generates creative output such as text, images, audio, or video.

AI Winter

A prolonged period of reduced interest in and funding for AI projects resulting from AI failing to deliver on much-hyped initial promises.

Prompt

A request made to a generative AI system, usually in written or spoken text, that guides the quality and content of the output.

AI Hallucination

A phenomenon where a generative AI system fabricates an answer, presenting incorrect information in a believable, articulate prose.

Large Language Model (LLM)

A type of AI used for general-purpose language understanding and generation that serves as the backbone for tools like ChatGPT and Gemini.

Foundation Model

The base model or technology used to train a large language model, such as OpenAI's GPT.

Parameters

Values used to determine text elements and relationships in a model; GPT-4 is estimated to have over a trillion ($10^{12}$) of these.

Corpus

The full dataset used to train an AI model, defining what the model knows and the endpoint of its knowledge.

DeepSeek

A powerful LLM developed in China that achieved performance on par with OpenAI models in 2025 at a fraction of the cost ($6M$ vs $100M$).

Machine Learning

A type of AI broadly defined as software with the ability to learn or improve without being explicitly programmed for each scenario.

Neural Networks

Statistical techniques modeled loosely after the human brain used to identify patterns in large datasets.

Deep Learning

A type of machine learning that uses multiple layers of interconnections among data to identify patterns; the engine behind most modern GenAI.

Artificial General Intelligence (AGI)

Software capable of learning and reasoning on any task or subject, including topics never encountered during training.

Turing Test

A 1950 proposal by Alan Turing to test a computer's ability to exhibit behavior indistinguishable from a human.

Expert Systems

An earlier generation of AI that leverages programmed decision rules or example outcomes to mimic applied human expertise.

Genetic Algorithms

An AI category inspired by biological evolution where computers iteratively modify and compare mathematical solutions to find the best function.

Supervised Learning

A technique where algorithms are trained by providing explicit, labeled examples of desired results.

Self-Supervised Learning

Machine learning where data is not explicitly labeled, allowing the model to discover its own structure; the approach used for most modern LLMs.

Reinforcement Learning from Human Feedback (RLHF)

A training technique using a reward model and human evaluators to tune results toward preferred outcomes.

Constitutional AI

A method for providing alignment and safety by incorporating a specific set of rules that the system must follow during machine learning.

Transformer

An AI architecture developed at Google that enables parallel processing of text and serves as the 'T' in ChatGPT.

Parallel Processing

An architectural feature where all words in a body of text are analyzed simultaneously rather than sequentially.

Agentic AI

Systems that can independently plan, make decisions, and take actions to achieve goals with minimal human oversight.

Vibe Coding

AI-assisted iterative software development where code is built 'by feel and conversation' rather than formal design specifications.

Adobe Firefly

Adobe's underlying GenAI technology used to create images, vector images, and video effects within professional creative suites.

Harvey AI

A GenAI model built on GPT-4 specifically trained and extended to meet the unique needs of the legal industry.

Data Poisoning

The act of deliberately feeding incorrect data to an AI to sabotage its results or corrupt its output.

Prompt Injection

A cybersecurity threat where malicious prompts are entered into an AI to make it behave in unintended or dangerous ways.

Deepfakes

Sophisticated AI-generated audio, images, or video created to look or sound like a real person or event.

The Black Box Problem

The challenge that deep learning systems work through statistical weights that are extraordinarily difficult for even their creators to interpret or explain.

Change Management

Techniques for preparing individuals for organizational change through training and support, often underestimated in AI deployment.

Red Teaming

The practice of engaging an independent group to deliberately try to undermine safety procedures or hack technology to surface concerns before deployment.

Audit Trail

A record that exposes how and when information systems are used so that the path to any particular outcome can be traced and defended.

The Echo Nest

The company acquired by Spotify in 2014 that enabled the analysis of audio characteristics and cultural context to power personalization.