Try it / four everyday decisions
Who makes the decision?
A pilot at the surface
A student watches a live camera feed and moves a joystick. The underwater vehicle turns when the student commands it.
Select an answer to reveal the explanation.
An original classification exercise, not a live AI model. These examples separate particular decisions; a real system can combine all three approaches.
A robot does not have to use AI
In Spare Parts, the engineering challenge involves an underwater robot. The real Carl Hayden vehicle, Stinky, was a remotely operated vehicle (ROV): people at the surface used feedback and controls to operate it. The team’s original report describes the hardware and control system. It does not establish the use of a modern trained AI model.
That distinction matters. Motors make movement possible; a camera provides images. Neither tells you who interprets those images or chooses the next movement. Ask what information enters the system, how a decision is made, and what happens next.
| Idea | What it describes | Example |
|---|---|---|
| Robotics | Designing and controlling a physical machine that interacts with its surroundings. | A vehicle with thrusters, a camera and a tether. |
| Automation | Carrying out a process with reduced human intervention. | A controller stops when a distance reading crosses a fixed threshold. |
| Artificial intelligence | Computational methods for tasks such as perception, reasoning and prediction. | A trained image model predicts which animal appears in a camera frame. |
| Machine learning | An approach within AI that learns patterns from training data. | Learning to distinguish labelled examples rather than hand-writing every visual rule. |
These are working distinctions, not sealed categories. AI includes approaches beyond machine learning. A robot can combine manual control, fixed rules and learned models; its behaviour may change between modes.
Autonomous does not automatically mean AI
An autonomous system performs a task without a person selecting every individual action. How much independence it has depends on the task and conditions. A simple programmed route and a learned navigation system are different designs, even if both run without continuous steering.
Likewise, AI does not need a robot body. An image classifier can run on a laptop and produce a label without moving anything. A prediction becomes part of robotics only when it contributes to a physical system’s operation.
From a human pilot to underwater AI
NOAA’s Deployable AI project provides a modern comparison. Its researchers developed systems to find and follow underwater animals using cameras, computers and algorithms on underwater vehicles. This combines perception with control; it is not how the 2004 Stinky design should be described.
For a classroom discussion, separate three questions: what does the camera measure, what does the model predict, and what is the controller allowed to do? Then consider a difficult image. A model can be wrong even when the motors work exactly as commanded.
Try the contrast: use the stopping-rule experiment to build automatic behaviour without machine learning. Then choose measurements in the sensor mission activity.
Sources & teaching notes
- Carl Hayden’s original 2004 technical report — historical vehicle and control details.
- Intel: robots and artificial intelligence — examples of sensing, processing and action.
- NOAA: Deployable AI — a documented contemporary research project.
The four scenarios and explanations are original teaching examples. No AI service runs in this activity, and answers are not submitted as student records.