AI and ethics

Why a machine needs ethics

It is tempting to think of a computer as neutral. It has no opinions, no grudges, so surely its judgements are fair. But modern AI does not follow rules a person wrote by hand. It learns from mountains of data, and that data is made by people. Whatever is in it, including our mistakes and our unfairness, can be learned along with everything else.

That matters because AI now helps make decisions that touch real lives. It can shape who sees a job advert, whose loan is flagged for a second look, which face a camera singles out of a crowd. When a system like that is wrong, it is not wrong in the abstract, since a person feels it. So the question is not only "does it work?" but "is it fair, and who is answerable when it is not?"

When bias hides in the data

Most AI learns from examples that humans have labelled: this photo is a cat, this email is spam, this applicant was a good hire. The trouble is that human choices carry human bias, and the machine cannot tell a fair pattern from an unfair one. It just copies whatever pattern is there.

Suppose a company trains a hiring model on its past decisions, and in the past it mostly hired one kind of person for a role. The model learns that pattern and quietly keeps favouring that same kind of person, not because anyone told it to, but because the history it was fed was already skewed. The bias was in the data before the machine ever touched it. This is one of the hardest problems in the field, because the unfairness can be invisible. The system looks objective, produces a confident answer, and passes on an old prejudice dressed up as maths.

Faces in the crowd

One use of AI raises these questions especially sharply: recognising people by their faces. The same computer vision that unlocks your phone can, at a larger scale, pick individuals out of a crowd and track where they go.

Imagine a government using face recognition to watch public spaces, matching passers-by against a database and logging who was where, and when. Used narrowly and with limits, such tools can help find a missing child. Turned on a whole population, they become something else: a way to monitor ordinary people who have done nothing wrong, to see who attended a protest, who met whom. The technology does not decide how it is used, people do. And once a system of constant watching is built, it is very hard to take back. That is why many argue this is one of the places where AI needs the firmest rules, not the loosest.

Fakes that look real

The same generative AI that can paint a picture or write a paragraph can also manufacture things that never happened. It can produce a photograph of an event that did not occur, or a video of a real person saying words they never said, convincing enough to fool a quick glance. Media faked this way is often called a deepfake.

Put that power beside the speed of social media and you have a serious problem. A believable fake can travel to millions before anyone checks it, and a correction rarely catches up. It can be used to smear a person, to swing opinion, or to make you doubt genuine footage by insisting it might be fake. On the hopeful side, AI can help here too, flagging likely fakes and offering people a fuller range of views instead of the narrow slice a feed tends to serve. But the core defence is not a clever tool. It is a habit of mind.

Learning to think critically

None of this means AI is the enemy. It means you meet its output with the same healthy caution you would give any powerful source. A few habits go a long way.

• Ask where it came from. Who made this, who trained it, and what might they want you to believe or do?
• Be slow with the startling. The content most designed to shock or delight you is the content most worth checking before you pass it on.
• Look for a second source. A real event usually leaves more than one trace, so a single dramatic clip with no corroboration deserves doubt.
• Remember the machine can be confidently wrong. A smooth, certain answer is not proof, only a well-formed guess.

AI will keep getting more capable, and the people who build it carry a real duty to weigh the consequences. But you are not helpless in front of it. A clear, questioning mind is the one tool that never goes out of date.

Put the module to work

Four habits are easy to nod at and hard to use. Here are two claims of the kind you actually meet. One of them holds up.

Module practice · 20 to 25 min

Judge two AI claims, and watch your own confidence move

The last lesson gave you four habits for meeting AI in the wild. This is where you use them on something, twice. Your work stays in this browser: nothing is uploaded, graded or kept.

Both claims below are written the way you would actually meet them. One of them holds up. Deciding which is the whole exercise.

Claim 1 · A notice from a school

"From next term our online exams use an AI proctor. It watches each student through the webcam and flags anyone whose eye movements suggest they are cheating. The system is 94% accurate."

Before you think about it too hard: would you believe this?

I would not believe itI could not say either wayI would broadly believe it

Record my first reaction

If you do not know where to start, ask what it saw

Every one of these systems learned from examples. That is the one thing you always know about it, even when you know nothing else, and it gives you a question that always works: what did it see, and does my situation look like that?

You do not need to know how the model works to ask it. You are not asking about the maths, you are asking about the photographs, the recordings, the past decisions it was shown. Those came from somewhere, and somewhere is usually not here.

If the honest answer is "it learned from people who do not look much like the people it is being used on", you have found the problem without understanding a single line of the system.

If that was straightforward

Now the harder version: suppose the proctor worked. Suppose someone solved the training data problem, and it really did detect cheating accurately across every kind of student. Would the school be right to install it?

This is where technical criticism runs out and something else has to take over. A system that watches every student through a camera in their own bedroom, and treats them as a suspect by default, is doing something to the relationship between school and student that has nothing to do with its accuracy. The ethics lesson made this point about face recognition in public squares. It is the same point, and accuracy does not answer it.

So the question that broke the proctor was a good one, and it is not the last question. Being able to say "this does not work" is easier than saying "this works and should not be built", and only the second one holds when the technology improves. It usually does improve.

Your turn: name one AI system you would object to even if it were highly accurate, and say what your objection rests on instead. Then notice whether that reason would survive someone showing you better accuracy figures.

Reset this practiceThis device remembers only which step you reached and which options you picked.

Found something unclear, outdated or improvable? Suggest an improvement

Check yourself

• AI often learns from data that people have labelled. Explain in your own words how human bias can end up inside a system that looks perfectly objective.
• Face recognition can help find a missing child or, at a larger scale, let a government track ordinary people. What makes the same technology helpful in one case and worrying in the other?
• You come across a shocking video clip of a public figure saying something outrageous. Using the habits in this lesson, what would you do before believing it or sharing it?