What Is Artificial Intelligence (AI)?

How AI, machine learning, and generative AI fit together

It helps to picture AI as a set of nested circles, each one roughly a narrower slice of the one around it.

• Artificial intelligence is the outer circle: any technique that gets a computer to do something we would call intelligent. Older AI systems, often called rule-based or expert systems, ran on rules written by people. A simple example of the idea is a spam rule like “if an email contains these phrases, mark it as spam.”
• Machine learning sits inside AI. Instead of following only hand-written rules, the system learns patterns from examples, like thousands of emails already labeled spam or not spam.
• Deep learning is a type of machine learning that uses neural networks (systems of simple, connected math units loosely inspired by brain cells) stacked in many layers. It is behind much of the recent progress in speech, image, and language tools.
• Generative AI is a subset, mostly built on deep learning, that creates new content such as text, images, audio, or code. Many chat assistants run on large language models, one type of generative AI.

When a machine learning system finishes training, the result is an AI model: the learned patterns that a product runs to make predictions or produce output. AI is the field; a model is the specific thing doing the work.

Narrow AI vs. general AI

Most AI in everyday business use is narrow AI (sometimes called weak AI): it is built and trained for a specific task or a defined range of tasks. A tool that pulls totals off invoices cannot schedule technicians, and a demand forecast cannot write a proposal.

General-purpose assistants such as ChatGPT, Claude, and Gemini blur the line a little, because they handle many kinds of requests, from drafting emails to explaining a spreadsheet. Experts disagree about how to classify them. Either way, they work from patterns learned in training plus whatever information they are given or can look up, such as your prompt, uploaded files, connected apps, or web search results, depending on the tool and how it is set up. How reliable they are varies from task to task.

Artificial general intelligence (AGI) refers to a system that could learn and perform almost any intellectual task a person can, at a human level or beyond. There is no agreed definition or test for it, so whether it exists yet depends partly on whose definition you use. Some argue today’s most capable systems already qualify under looser definitions; many experts disagree about whether or when it will arrive. For business decisions, judge AI tools by what they do reliably today, not by predictions about where the field is headed.

Where AI already shows up at work

Much of the AI in business is built into tools you already use, so it often goes unnoticed. Common examples include:

• Spam and fraud filtering: classifying emails, logins, or payments as normal or suspicious.
• Recommendations: suggesting products, videos, or articles based on your past behavior and what similar people chose.
• Forecasting: predicting demand, call volume, staffing needs, or inventory from historical data.
• Language and documents: transcribing calls, translating text, and pulling key fields out of invoices or forms.
• Images and inspection: reading photos or video to spot damage, count stock, or check quality.
• Chat and writing assistants: drafting emails, summarizing documents, and answering questions, which are typical jobs for generative AI.

Some newer systems go a step further and take actions, not just produce answers. An AI agent might look up a customer record, draft a reply, and update a support ticket, usually within limits a person sets. The more an AI system can do on its own, the more its permissions and review steps matter.

What AI is not

AI can produce fluent, useful output, but it does not understand the world the way a person does. Most modern AI works by finding statistical patterns in data. It has no lived experience, no stake in the outcome, and no built-in sense of whether an answer is true. A few practical consequences follow:

• It can be confidently wrong. Generative tools sometimes produce plausible statements that are false, known as hallucinations.
• It reflects its data. If training data is incomplete, outdated, or skewed, the output can be too, including unfair patterns about groups of people.
• It does not reliably know its limits. A system may answer a question it has no good basis for instead of saying it does not know.
• “AI” on a label is not specific. It can describe anything from a simple prediction formula to a large language model, so ask what a tool actually does and how it was tested.

For decisions that affect customers, money, safety, or people’s jobs, treat AI output as a draft or a recommendation, and keep a person accountable for the final call.

Further reading

• AI Tool LibraryBrowse AI tools sorted by the job they do to see how AI shows up in everyday work.
• AI Fluency AssessmentTake a free assessment to see where your business stands on AI adoption.