Digital ethics
The ethical questions raised by a digital product and the tools to answer them: reasoning frames (consequences, duties, virtues), deceptive interfaces, the attention economy, algorithmic bias, privacy by design, accessibility and responsibility.
- 25 cards
- 5 topics
- Free
The outline, topic by topic
Each topic is weighted the way the exam weighs it.
- Concepts
- Questions
- Sort and link
- Scenarios
- Understand
A few cards from the deck
Digital ethics: what is it?
Show the answer
Reflection on what is right to design and do with a digital product, beyond what the law requires.
The law sets a minimum and often comes after the technology; ethics asks what is right when no text settles it. A legal practice can still be unethical.
A facial recognition model trained mostly on light faces makes more mistakes on dark faces. What is the most likely origin?
- Training data that are not representative
- Too low an image resolution, the same for every user
- A model too small to recognise a face properly
- The team's wish to disadvantage certain users
Show the answer
Training data that are not representative
Few examples of a group produce a model that is less reliable for that group. A low resolution or too small a model would degrade every face, not one group more than another. No intent is needed: it is a data bias, detected by measuring errors per group.
… judges an act by its results, … by whether it respects duties and rules, and … ethics by the character it shows.
Show the answer
Consequentialism judges an act by its results, deontology by whether it respects duties and rules, and virtue ethics by the character it shows.
The three frames ask different questions: what does it produce? am I allowed to? what would an honest person do? They sometimes give opposite answers to the same case.
Dark pattern: what is it?
Show the answer
A screen designed to lead the user to do what they would not have chosen knowingly, for the publisher's benefit.
It exploits visual hierarchy, pre-ticked boxes, urgency or guilt. The test is informed consent, not how ugly the interface is.
A tool that ranks job applications, trained on ten years of hiring, screens out women's CVs. What should be done?
- Remove the sex field from the form, which is enough to remove the bias
- Train the model for longer on the same data
- Change nothing, since a model only calculates and stays neutral
- Audit the data and the results by group, and keep a human decision
Show the answer
Audit the data and the results by group, and keep a human decision
The model reproduces a biased past. Removing the field is not enough: other information (career, hobbies, first name) stands in for it. Gaps must be measured per group and the tool must not decide alone.