Exercises: Ethics of data and artificial intelligence

Explain with an example what algorithmic bias is, and why an artificial intelligence trained on limited or unbalanced data can treat certain groups of people unfairly.

The figure summarizes how an artificial intelligence system learns: from examples.

The training data in the example contain many cases of one type and very few of another. Explain what problem will arise in the decision marked with ?, where this problem actually comes from, and how it could be corrected.

Explain what a chatbot is, why it can state made-up facts with complete confidence (so-called "hallucinations"), and what you should always do when it gives you an important piece of information: a date, a figure, or a source.

If a self-driving car hits a pedestrian, who is responsible? Lay out the options (owner, manufacturer, programmer, shared liability) and defend an initial position with an argument.

Explain what data a video platform's recommendation algorithm uses to select the next video (what you watch, for how long, what you skip) and why it tends to lead you toward increasingly intense or extreme content.

Explain what it means for an artificial intelligence to function as a "black box", and why it is concerning that even its creators cannot explain a specific decision, such as denying a grant or a loan.

Artificial intelligence systems learn using "training data". Explain why how these data were obtained matters, and evaluate the case of millions of photos of people being used without their consent to train a facial recognition system.

A company uses artificial intelligence to screen résumés, and the system rejects more applications from women because in the past the company mainly hired men. Explain where this bias comes from, why "the machine made the decision" does not relieve the company of responsibility, and what measure would fix the system.

Explain what is meant by technological sovereignty, and why there is concern about a small number of countries or companies controlling the microchips, cloud infrastructure, and large artificial intelligence models on which millions of people depend.

A school uses artificial intelligence to grade essays. Analyze its advantages (speed, consistent criteria, does not get tired) and its risks (does not understand irony, penalizes unconventional styles, errors with no appeal process), and define what human oversight you would demand before accepting its grades.

A fake video of a female classmate "singing" something ridiculous is created, and the whole school laughs. Analyze which rights it infringes (image rights, reputation), what responsibility lies with the creator and those who forward it, and what she and those receiving it should do.

AI image generators learn from artists' artwork, often without paying or asking for permission. Analyze the conflict between innovation and copyright, and evaluate proposed solutions: training licenses, opt-out lists, and labeling generated artwork.

Training a large artificial intelligence model consumes as much electricity as hundreds of homes in a year, and every query also consumes energy. Explain what sustainability questions companies should answer and what responsible use criteria you could apply.

A city council proposes installing facial recognition cameras "to improve security." Construct a balanced argument: two potential benefits, two risks to privacy, and one condition you consider essential before approving the measure.

Artificial intelligence assistants write full texts in seconds. Analyze which uses are legitimate in your studies and which undermine learning, and propose a classroom rule on the use of AI that you consider fair and enforceable.

Evaluate the statement "technology is neutral: it all depends on how it is used." Use the examples of algorithmic bias and planned obsolescence to support or qualify it with arguments.

Design an "ethical traffic light" system for student use of artificial intelligence: which uses would be green (learning, accessibility, practicing languages), yellow (translating entire homework assignments, generating images of real people), and red (impersonation, fabricating fake news, presenting AI-generated text as your own). Justify each category.

The classic trolley problem applied to self-driving cars: if an accident is unavoidable, how does the car decide whom to put at risk? Analyze who should make that decision (the engineer, the law, a citizen panel) and defend an argued position.

Much of the work of labeling data to train artificial intelligence is outsourced to countries with very low wages, including the review of traumatic content. Analyze what injustices you see in this global supply chain, and what conditions (fair pay, psychological support, transparency) you would demand from companies outsourcing this work.

An artificial intelligence composes a song imitating an artist's style. Who owns the work? Analyze the perspectives (the programmer, the user, no one, the imitated artist), what value humans bring to art, and argue for a reasonable solution.

Some experts called for a pause in the development of the most powerful artificial intelligence models until clear rules are in place. Argue in favor (precautionary principle) and against (international competition, delaying beneficial uses) and propose a set of minimum regulations that you consider reasonable.