AI Dictionary

AI refers to computer systems designed to perform tasks that typically require human intelligence, such as recognizing patterns, making decisions, or translating languages.

It is not a single magical machine, but a family of technologies that learn from data.

The system that recommends your next movie or helps your phone understand your voice.

Machine learning is a method where computers learn to recognize patterns from examples instead of being explicitly programmed with rules.

Rather than telling a computer exactly how to identify a cat, you show it thousands of pictures of cats until it learns the pattern.

Your email spam filter learning which messages are junk based on what you delete.

A model is the underlying software engine that has been trained on vast amounts of data to recognize patterns and make predictions or generate text.

Think of it as a highly complex mathematical recipe that takes an input (like a question) and calculates the most likely useful output.

GPT-4 is a model that powers various chatbot applications.

An LLM is a specific type of AI model trained on massive amounts of text to understand and generate human-like language.

It predicts the next most likely word in a sequence based on patterns it learned during training, but it does not actually understand the meaning the way a human does [1].

The engine behind tools that can draft emails or summarize articles for you.

Generative AI refers to systems that can create new content—such as text, images, audio, or video—based on patterns they have learned.

Instead of just analyzing existing data, these systems produce original outputs that mimic human creation.

A tool that creates a custom illustration from a written description.

An algorithm is a step-by-step set of instructions that a computer follows to solve a problem or complete a task.

In AI, algorithms determine how the system learns from data and makes decisions.

The recipe your navigation app uses to find the fastest route home.

A neural network is a computing system inspired by the structure of the human brain, designed to recognize complex patterns.

It consists of interconnected layers of "nodes" that process information, allowing the system to handle messy, real-world data like images or speech.

The technology that allows your phone to recognize your face to unlock the screen.

A chatbot is a conversational interface that allows you to interact with an AI model using everyday language.

The chatbot is the steering wheel and dashboard; the model is the engine underneath.

The chat window where you type a question and receive an AI-generated answer.

A prompt is the instruction, question, or context you give to an AI tool to get the result you want.

Writing a good prompt involves providing clear goals, context, and constraints, rather than relying on secret tricks.

"Draft a polite, three-sentence email declining a meeting invitation for next Tuesday."

Context is the background information and specific details you provide in a prompt to help the AI understand exactly what you need.

The more relevant context you provide (without sharing private information), the more accurate and useful the output will be.

Telling the AI that you are writing to a formal business client rather than a close friend.

Iteration is the process of refining your prompt or the AI's output through back-and-forth conversation to improve the final result.

You rarely get a perfect answer on the first try; the best results come from treating the AI like a drafting partner and guiding it toward what you need.

Asking the AI to make its previous answer shorter and less formal.

The output is the text, image, or other result generated by the AI system in response to your prompt.

The quality of the output depends heavily on the clarity of your prompt and the capabilities of the underlying model.

The completed travel itinerary the AI generates after you ask for vacation ideas.

A persona is a specific role or perspective you ask the AI to adopt when generating its response.

Assigning a persona helps the AI match the tone and vocabulary appropriate for your task.

"Act as a patient, experienced math tutor explaining fractions to a ten-year-old."

Using AI as a drafting partner means relying on it to create a starting point, while you retain responsibility for reviewing, editing, and finalizing the work.

This approach keeps human judgment in charge and treats the AI's output as a rough draft rather than a finished product [2].

Having the AI outline a presentation that you then fill in with your own expertise.

A hallucination occurs when an AI model generates information that sounds confident and plausible but is entirely fabricated or factually incorrect.

Because language models predict the next likely word rather than checking a database of facts, they can smoothly invent details when they lack the right information [2].

An AI citing a scientific study that never actually happened.

Verification is the essential human habit of checking AI-generated claims against trusted, independent sources before relying on them.

You should never use an AI tool as the final authority for important facts, especially regarding health, law, or finance [2].

Looking up a statistic provided by an AI on an official government website to ensure it is accurate.

A deepfake is a highly realistic but manipulated video, audio recording, or image created by AI to make it appear that someone said or did something they did not.

The technology has advanced to the point where you can no longer rely solely on your eyes or ears to verify authenticity [3].

A fake video of a politician making a controversial statement they never actually made.

A voice clone is an AI-generated audio replica of a specific person's voice, often created from just a short sample of them speaking.

Scammers use voice clones to impersonate family members in distress; always verify suspicious calls through a separate, trusted channel [3].

Receiving a frantic phone call that sounds exactly like your grandchild asking for money.

Private information includes any sensitive data—such as passwords, health records, financial details, or confidential work documents—that should never be entered into a public AI tool.

Many AI services use the information you provide to train future models, meaning your private data could potentially be exposed [2].

Your social security number or a client's unreleased financial report.

A high-stakes decision involves medical, legal, financial, employment, security, or account-access matters where errors carry significant real-world consequences.

AI should never be the final decision-maker in these areas; they require qualified human authority and review [2].

Relying on a chatbot to diagnose a medical symptom or draft a binding legal contract.

Bias in AI occurs when a model produces skewed, unfair, or prejudiced results because of imbalances or historical prejudices in its training data.

If a system is trained primarily on examples from one demographic, it may perform poorly or unfairly for others [2].

A hiring tool that unfairly downgrades resumes from female candidates because it was trained on historical data favoring men.

An independent channel is a separate, trusted method of communication used to verify a suspicious message or request.

If you receive an alarming text or call, do not reply directly; instead, hang up and contact the person using a phone number you already know belongs to them [3].

Calling your bank directly using the number on the back of your card rather than clicking a link in a text message.

An AI agent is a system designed not just to answer questions, but to use tools and take actions to achieve a specific goal under defined permissions.

While a chatbot drafts an email, an agent might be given permission to read your calendar, draft the email, and send it—though consequential actions still require human approval [2] [4].

A system that monitors your inbox for invoices and automatically drafts payment approvals for you to review.

The permission ladder is a framework for safely delegating tasks to AI, starting with reading and drafting, and only moving to autonomous action after proving reliability.

It ensures you do not hand over control of important workflows before establishing proper approval gates and safeguards [2] [4].

Requiring you to click "approve" before an AI agent is allowed to send an email to a client.

Task transformation refers to how AI changes specific parts of a job—such as drafting reports or organizing data—rather than replacing the entire occupation.

Most jobs consist of many tasks; AI is more likely to assist with some of those tasks than to completely eliminate the need for the human worker [5].

A lawyer using AI to quickly summarize hundreds of pages of documents, freeing up time for complex legal strategy.

Augmentation is the use of AI to enhance human capabilities, improve efficiency, or elevate the quality of work, keeping the human firmly in the loop.

The goal is to make people better at what they do, not to replace their judgment or expertise [5].

A graphic designer using AI to generate background textures while focusing their creative energy on the main subject.

Human oversight is the practice of maintaining active human responsibility for reviewing, approving, and correcting the actions or outputs of an AI system.

Even as AI becomes more capable, human judgment remains essential for context, ethics, and final accountability [2].

A doctor reviewing an AI-generated summary of a patient's medical history before making a diagnosis.

Automation is the use of technology to perform repetitive tasks with minimal human intervention.

While traditional automation follows rigid rules, AI automation can handle more complex, variable tasks—but it requires careful supervision [4].

Automatically categorizing incoming customer support tickets based on their content.

Workflow redesign involves rethinking how a complete process operates to safely and effectively incorporate AI, rather than just inserting a chatbot into an old system.

It requires establishing new rules for data permissions, human review, and error recovery to achieve dependable results [4].

Changing a customer service process so AI handles initial information gathering, but a human agent always handles the final resolution.

Training data is the massive collection of text, images, or other information used to teach an AI model how to recognize patterns.

The quality, diversity, and accuracy of this data fundamentally determine how well the resulting model will perform and whether it will exhibit bias [2].

Millions of digitized books, articles, and websites used to teach a language model how humans write.

A data center is the physical facility housing the thousands of specialized computers (servers) required to train and run large AI models.

AI is not weightless software; it relies on massive physical infrastructure that requires significant space, cooling, and electricity [6].

A warehouse-sized building filled with rows of servers processing AI requests.

Compute refers to the processing power and specialized hardware—particularly computer chips—needed to train and operate AI systems.

Advanced AI requires massive amounts of compute, which is why the chips that power these systems are in such high demand globally [6].

The specialized graphics processing units (GPUs) that calculate the billions of mathematical operations needed for a single AI response.

Energy intensity describes the substantial amount of electricity required to power and cool the data centers that run AI models.

An AI search request can require significantly more electricity than a traditional web search, making energy efficiency a major challenge for the industry [6].

The power needed to keep thousands of servers from overheating while they process complex AI calculations.

An open source model is an AI system whose underlying code and architecture are made publicly available for others to inspect, modify, and build upon.

This approach encourages innovation and transparency, though it also raises debates about how to prevent misuse of powerful systems.

A language model released by a research group that allows developers to run it on their own servers for free.

A proprietary model is an AI system owned and controlled by a specific company, which keeps the underlying code and training details secret.

Users typically access these models through a paid subscription or API, and the company maintains control over updates and safety guardrails.

The commercial models offered by major technology companies that you access via their websites.

An API is a software bridge that allows different computer programs to communicate with each other, enabling developers to build AI capabilities into their own apps.

It is how a small business can add a powerful AI chatbot to their website without having to build or train the model themselves.

A travel website using an API to send your destination request to an AI model and receive the generated itinerary to display on your screen.