AI Reflection, Project Cycle and Ethics

A machine is intelligent when it can learn from data, understand what is going on and make decisions on its own, a little like a human does. Artificial Intelligence (AI) is the science of building such machines. A normal machine only follows fixed instructions.

AI works with three kinds of data. Data (statistics) works with numbers and tables, Computer Vision works with images and video, and Natural Language Processing (NLP) works with text and speech. Beat the data-based AI below, then try the vision and language games linked at the end of the unit.

Building an AI project follows six stages, called the AI Project Cycle. It is a cycle because what we learn at the end often sends us back to improve an earlier stage.

Problem scoping means understanding a problem before trying to solve it. The 4Ws canvas asks Who is affected (the stakeholders), What the problem is, Where it happens and Why it is worth solving. The answers become a problem statement. Scoping is iterative: as you learn more, you go back and sharpen it.

Data acquisition means collecting the right data from reliable sources. First list the data features you need, then find where to get them: surveys, sensors, cameras, observations, websites or open datasets. A system map shows how the parts of a problem affect each other, which helps you decide what data to collect.

Raw data is just a long list of numbers. Data exploration arranges and visualises it so patterns jump out. Each kind of graph suits a different question: bar graphs compare groups, line graphs show change over time, pie charts show parts of a whole, scatter plots show a link between two numbers and histograms show how data is spread.

A model is the part of an AI system that makes decisions. In a rule-based model, people write the rules. In a learning-based model, the machine works out the rules by itself from examples, and can learn again when the data changes. Sort apples from oranges both ways.

To evaluate a model we compare its predictions with what really happened. Each prediction is one of four kinds: True Positive (predicted yes, really yes), True Negative (predicted no, really no), False Positive (predicted yes, really no) and False Negative (predicted no, really yes). Together they form a confusion matrix.

Deployment means putting the tested model to work in the real world, then watching how it performs and improving it. Follow a real kind of project: AI that helps prevent blindness caused by diabetes.

Ethics is about knowing right from wrong. AI systems make decisions that affect people, so the people who build them must think about fairness, safety, privacy and who is responsible when something goes wrong. Often there is no single right answer.

AI bias happens when an AI system treats some people unfairly, usually because its training data did not represent everyone. AI access is about who can use AI at all: people without devices, internet, money or skills can be left behind, which widens the gap between them and everyone else.

AI can save lives, time and effort, and do jobs that are dangerous or boring for people. It can also take away some jobs, make mistakes, copy unfair bias and invade privacy. In a balloon debate, each side argues its case. Weigh the arguments yourself.