The 36 terms worth knowing.
AI keeps coming up. These are the terms behind it, explained the way a patient friend would. No jargon. No assumed knowledge.
AI agent
An AI system set up to carry out a task for you, not only answer a question. Ask it to compare flights and it might open websites, fill in forms and report back. Handing over actions raises the stakes, so check what it plans to do before you let it spend money or send anything on your behalf.
Learn more in the Plainly app →AI ethics
Who is harmed when an AI system gets something wrong? Who is responsible? Should a face be scanned without consent? AI ethics covers questions like these about fairness, safety and accountability. Reasonable people disagree on many of them, which is why the debate continues and why ordinary users' views count.
Read the lesson: The companies behind AI, and what they want →Algorithm
A set of steps a computer follows to get something done. Your email app runs one when it filters junk mail. So does the route your map app picks. Think of it as a recipe written for a machine: precise, repeatable and only as good as its instructions and the information it is given.
Read the lesson: It's not a robot, and it's not magic →Artificial intelligence (AI)
Software that handles tasks we once thought needed a person, such as answering a question, recognizing a face or translating a menu. Underneath, most of it finds patterns in enormous numbers of examples and uses them to make predictions. That one idea explains much of what it does well, and where it slips.
Read the lesson: It's not a robot, and it's not magic →Automation
Having a machine do a repeated task without someone guiding it each time. Factories have automated work for well over a century. AI widens the range of tasks that qualify, especially work built on reading, writing and sorting information. It tends to take over tasks within a job rather than whole jobs.
Learn more in the Plainly app →Bias (in AI)
AI learns from data that people produced, so it absorbs their blind spots along with their knowledge. If some groups were underrepresented in that data, the system may serve them worse: a face it misreads, a loan application it scores unfairly. Fixing it starts with asking who was in the data and who was missing.
Read the lesson: How it learned everything it knows →Chatbot
Any program you talk to in ordinary sentences. Some follow a fixed script, like the help box on a utility company's website. Others, such as ChatGPT or Gemini, generate a fresh answer each time. Same label, very different abilities, so it helps to know which kind you're dealing with.
Read the lesson: ChatGPT and the moment AI became impossible to ignore →ChatGPT
An AI assistant made by the company OpenAI. Its public release in late 2022 is what turned AI chat tools into a household topic. You ask in plain language and it replies the same way. Its features change often, so the quickest way to learn what it can do is to ask it or check its settings.
Read the lesson: ChatGPT and the moment AI became impossible to ignore →Claude
An AI assistant made by the company Anthropic. Ask it to explain a confusing letter, summarize a long document or help word a difficult email, and it works through the request with you in conversation. Like its peers, it can be wrong with confidence, so check anything that matters.
Read the lesson: The others: Gemini, Copilot, Meta AI and Claude →Cloud
Computers in someone else's building, doing work your device doesn't do itself. When you ask a voice assistant a question or back up your photos, your data usually travels over the internet to a company's servers and back. That's why "where is my data stored?" is always a fair question to ask.
Read the lesson: AI and your privacy: what to know before you type →Computer vision
AI that interprets images and video. It's how a photo app finds every picture of a beach without labels, how a phone recognizes your face, and how a doorbell camera tells a person from a passing car. It can identify what's in a picture. It can't know what that person or place means to you.
Read the lesson: AI you're already using without realizing it →Copilot
Microsoft's AI assistant. Besides working as a chat app of its own, it appears inside Microsoft programs that many people already use for documents, email and spreadsheets. If you see a Copilot button in one of them, that's Microsoft's assistant offering help where you're working.
Read the lesson: The others: Gemini, Copilot, Meta AI and Claude →Data broker
A company that gathers personal information from many sources, such as apps, store loyalty programs and public records, then sells or shares profiles of people. You rarely deal with brokers directly, so most people don't know these companies hold a file on them. Some let you request a copy or opt out, and searching your own name is a good first step.
Learn more in the Plainly app →Data center
A large building packed with computers that store data and run online services, including AI. When you chat with an assistant, the heavy work usually happens in one of these rather than on your phone. They use a great deal of electricity, and water for cooling, which is why their environmental cost comes up in discussions of AI.
Learn more in the Plainly app →Deepfake
Video, images or audio made with AI to show a real person saying or doing something they never did. Judging them by eye gets harder as the technology improves, so rely on habits instead. Check where it came from, look for the original, and confirm through a source you already trust before you believe or share it.
Read the lesson: Scams and fakes: four habits that beat most of them →Facial recognition
Software that measures the features of a face and compares them with faces it already has on file. Your phone uses a version of it to unlock. Stores, airports and police use it too, and that's where the debate about consent, accuracy and surveillance lives.
Read the lesson: AI you're already using without realizing it →Gemini
Google's AI assistant. You can chat with it on its own, and Google has built it into many of its other products. What it can see of your Google account depends on settings you control, so look there if you want to know what it has access to.
Read the lesson: The others: Gemini, Copilot, Meta AI and Claude →Generative AI
AI that produces something: text, images, audio, video or computer code. Older AI mostly sorted and classified what already existed, deciding whether an email was spam, for instance. Generative tools write the email. Chat assistants such as ChatGPT, Gemini and Claude are the most familiar examples.
Read the lesson: ChatGPT and the moment AI became impossible to ignore →Hallucination
When an AI tool states something false in the same confident tone it uses for facts. A made-up book title, a court case that never happened, a wrong dosage. It isn't lying. It produced something shaped like a correct answer. For anything that matters, check it against a source you trust.
Read the lesson: When it's wrong and doesn't know it →Large language model (LLM)
The kind of system behind chat assistants. It was trained on a vast amount of text and learned to predict which words should come next, one piece at a time. Do that well enough, with enough context, and the result reads like conversation, explanation or advice.
Read the lesson: How it learned everything it knows →Machine learning
Teaching software with examples instead of writing a rule for every case. Show a system thousands of spam emails and thousands of good ones, and it works out the differences on its own. Nearly every AI tool you meet, from spam filters to chat assistants, was built this way.
Read the lesson: It's not a robot, and it's not magic →Memory (in AI assistants)
Some AI assistants can carry details from one conversation to the next, such as your name, your preferences or a project you mentioned. That saves repeating yourself. You can usually view, edit or delete what's remembered, or turn the feature off, in the assistant's settings.
Read the lesson: AI and your privacy: what to know before you type →Model
The trained system that does the work, as distinct from the app you open. ChatGPT is a product. The model underneath generates the words, and companies swap in improved models from time to time, which is one reason a tool's answers can change.
Read the lesson: How it learned everything it knows →Natural language processing (NLP)
The branch of AI that deals with ordinary human language, written or spoken. Translation apps, voice assistants, spam filters and chat assistants all rely on it. It's why you can type a question the way you'd ask a friend instead of hunting for the right keywords.
Read the lesson: ChatGPT and the moment AI became impossible to ignore →Neural network
Layers of simple calculations wired together, loosely inspired by how brain cells connect. Each layer passes signals to the next, and training adjusts the strength of the connections until the output comes out right. The brain comparison is a handy image, but the resemblance is shallow.
Read the lesson: How it learned everything it knows →Open source AI
AI models whose underlying files are released for anyone to download, inspect, run and modify. Supporters say openness lets outsiders find flaws and keeps power from concentrating in a few companies. Critics worry it lets anyone strip out safety limits. Both concerns are real, and the label covers different degrees of openness.
Learn more in the Plainly app →Phishing
A message designed to trick you into clicking a link, sharing a password or sending money, usually by posing as your bank, a delivery company or someone you know. AI helps scammers write polished, personal messages, so spelling mistakes are no longer a reliable clue. Go to the company's website or app yourself instead of using the link.
Read the lesson: Scams and fakes: four habits that beat most of them →Privacy (in AI)
The question of what happens to what you type, say or upload. Assume it may be stored and, in some cases, reviewed by people or used to train future systems. Check the privacy controls in any assistant you use, and don't type passwords, account numbers or anything else you wouldn't want saved.
Read the lesson: AI and your privacy: what to know before you type →Prompt
Whatever you type or say to an AI tool. A short one gets a generic answer. Explain your situation, what you want, the tone and anything to avoid, and the answer improves noticeably. If the first reply misses, say what's wrong and ask again. That back-and-forth is most of the skill.
Read the lesson: Context: the one habit that changes every answer →Recommendation algorithm
The system choosing what a streaming service suggests, what appears in your social feed and which products a shop puts in front of you. It predicts what will hold your attention or lead to a purchase. That isn't always what you'd pick on reflection, so it's worth noticing when a feed is steering you.
Learn more in the Plainly app →Robo-advisor
An automated investing service that builds and rebalances a portfolio using set rules instead of a human advisor. Fees tend to be lower, but it knows only what you tell it. Check that the firm is registered, for example through FINRA BrokerCheck or the SEC's investment adviser search, and remember that registration is not an endorsement.
Learn more in the Plainly app →Training cutoff
The point where a model's training data ends. Events after that date aren't part of what it learned, though some assistants can look up current information separately. For anything time-sensitive, ask the tool where its answer came from and confirm it with a current source.
Read the lesson: How it learned everything it knows →Training data
Everything a system learned from while it was being built: books, websites, articles, photos, recordings. Its abilities come from that pile, and so do its gaps and slants. Whatever was missing or skewed in the data tends to show up in the answers.
Read the lesson: How it learned everything it knows →Transformer
A design for AI systems that researchers introduced in 2017. It lets a model weigh how each word in a passage relates to every other word, which made language systems far more capable. It's the T in ChatGPT, and the approach behind the chat assistants that made AI a household topic.
Read the lesson: ChatGPT and the moment AI became impossible to ignore →Voice assistant
A program that listens for a wake word, such as "Siri" or "Alexa," then acts on what you say next. Your request is often sent to the company's servers to be processed. Most keep a history of recordings or transcripts, and you can usually review and delete it in the app's privacy settings.
Read the lesson: AI you're already using without realizing it →Voice cloning
Using AI to copy a person's voice from a short recording, sometimes only seconds long. Scammers use it to sound like a relative in trouble. Agree on a family code word in person, and if a caller claims to be family in an emergency, hang up and call back on the number you already have.
Read the lesson: Scams and fakes: four habits that beat most of them →