It Is Probably Not What You Think
Ask ten people what artificial intelligence is and you will get ten answers that cluster around the same few images: ChatGPT, robots that look like humans, self-driving cars, maybe HAL 9000 from a movie they saw years ago. Those answers are not wrong exactly — all of those things involve AI — but they are a little like answering "what is medicine?" by pointing at a hospital emergency room. The emergency room is real and important, but medicine is much larger than that, and most of it is quiet and invisible.
The same is true of AI. Most of it is quiet and invisible. It is in the filter that catches spam before it reaches your inbox. It is in the algorithm that decides which search results appear first. It is in the system that reads the numbers on checks at your bank, the software that flags unusual charges on your credit card, the navigation app that reroutes you around an accident. None of those systems look like a robot. None of them can hold a conversation. Most people who use them every day would not think to call them AI at all.
This is the first and most important thing to understand about artificial intelligence: it is not a single technology. It is a field of computer science — a broad collection of research questions, techniques, and systems united by a common goal. That goal, stated simply, is to build machines that can carry out complex tasks without requiring a human to direct every step.
Agents: The Central Idea
To talk about AI with any precision, we need one key concept: the agent. An agent is any system that perceives its environment and takes actions in response. That definition is intentionally broad. A thermostat is a simple agent — it perceives temperature and responds by turning heat on or off. A chess-playing program is an agent — it perceives the state of the board and responds by selecting a move. A spam filter is an agent — it perceives incoming email and responds by routing messages to your inbox or your spam folder.
Most agents have two basic components: sensors and actuators. Sensors are how an agent takes in information about its environment. Actuators are how an agent affects its environment — its outputs, its actions, its responses.
For a self-driving car: cameras, radar, and GPS are sensors; the steering system, accelerator, and brakes are actuators. For a spam filter: the incoming email is the sensor input; moving a message to the spam folder is the actuator output. For a voice assistant: the microphone is a sensor; the speaker is an actuator.
An agent does not have to be a physical robot. It does not have to look like anything at all. What makes something an agent is that it takes in information and acts on it — automatically, without a human making each individual decision.
This framing is useful because it helps us see AI systems clearly. When we ask "what kind of AI is this?" we can ask: what does it sense? What does it do in response? How does it decide what to do? Those questions cut through a lot of marketing language and get to what is actually happening.
AI Is Already Everywhere
Let us make the breadth of the field concrete. The following are all examples of AI systems. They are different in almost every way — in what they sense, what they do, and how they make decisions — but they all fit the definition of an agent carrying out complex tasks without human intervention at every step.
- Spam filters. Classify incoming email as legitimate or spam based on learned patterns. Sensors: the email content. Actuators: routing decisions.
- GPS navigation. Finds optimal routes through a road network, adjusting in real time based on traffic. Sensors: map data, traffic data, your location. Actuators: route instructions.
- Optical character recognition (OCR). Converts images of text into editable digital text. This is how your phone camera can read a document, how banks process paper checks, and how the postal service sorts mail automatically.
- Recommendation systems. Suggest what to watch, read, or buy based on patterns in your past behavior and the behavior of similar users. The algorithm deciding what appears in your social media feed is an AI system.
- Medical diagnosis assistance. Systems that analyze medical images — X-rays, MRIs, pathology slides — to flag potential diagnoses for a physician to review. In some contexts, these systems are more accurate than individual human readers.
- Game-playing systems. From chess programs to systems that play video games at superhuman levels. These were historically important research targets in AI because games provide a constrained, measurable environment.
- Large language models. Systems like ChatGPT and Claude that generate text in response to prompts. These are the most visible AI systems right now — and we will study them in depth in Week 7, after we have built up the context to understand them properly.
Notice how different these systems are. A spam filter and a medical imaging system use completely different techniques, run on completely different data, and solve completely different problems. What they share is the goal: perform a complex task autonomously, without a human making every decision.
A Running Example: The Spam Filter
Throughout the Week 6 readings, we will use the spam filter as a recurring example. It is not the most glamorous AI system, but it has a property that makes it ideal for learning: it is simple enough to reason about clearly, but rich enough to illustrate almost every concept we will cover.
Here is the basic setup. Your email inbox receives hundreds of messages. Some are legitimate. Some are spam — unwanted commercial messages, phishing attempts, outright scams. A spam filter is an agent whose job is to classify each incoming message and route it appropriately. It does this automatically, for every message, without you making the decision.
How does it decide? That turns out to be a rich question — one whose answer has changed dramatically over the decades as AI techniques have evolved. The simplest spam filters used fixed rules: if the message contains certain words ("FREE!!!", "Click here to claim"), route it to spam. More sophisticated filters learned from patterns in large collections of known spam and legitimate mail. The most modern filters use techniques that we will cover in Week 7.
As we work through the concepts in Week 6, we will return to the spam filter to see how each idea applies. By the time we reach Week 7, you will be able to trace the full evolution from a simple rule-based system to a modern learned classifier — and that arc will help you understand what has changed in AI, and why.
What Actually Makes Something "AI"?
Given how broad the field is, you might wonder: is there a principled line between AI and "just a program"? The honest answer is: not really. The boundary has always been fuzzy, and it tends to move. There is a saying in the field, sometimes called the AI effect, that goes something like this: once a problem is solved, it stops being called AI. When computers first beat humans at chess, it was a landmark AI achievement. Today, chess programs are just programs — nobody calls your phone's chess app "artificial intelligence" anymore.
What most people mean when they say AI is something closer to: a system that does something that, until recently, required human judgment. That is an intentionally vague definition, and it gets vaguer as technology advances. But it is a useful working definition for our purposes, and it explains why the field keeps growing: every time a capability is mastered, it gets absorbed into regular software, and the frontier moves to the next hard problem.
For our purposes as educators, the important thing is not to draw a precise line between AI and non-AI. It is to understand enough about how these systems work that we can reason about them clearly, explain them to students accurately, and engage thoughtfully with the questions they raise. That understanding is what the next two weeks are designed to build.
What Comes Next
In Reading 2, we will look more closely at the different ways AI agents make decisions — the spectrum from simple reflexes to goal-directed planning to genuine learning. That distinction matters a great deal, both for understanding how different AI systems work and for explaining them to students.
In Reading 3, we will examine the Turing Test: what it was, why it seemed like a good idea, and what a program called ELIZA revealed about both the test and our own tendency to see intelligence where there is none.