AI Ethics in Practice: Fairness, Bias & the EU AI Act
Published 10/2026
Created by Armaan Sidana
MP4 | Video: h264, 1920×1080 | Audio: AAC, 44.1 KHz, 2 Ch
Level: All Levels | Genre: eLearning | Language: English | Duration: 25 Lectures ( 2h 14m ) | Size: 979.1 MB
Responsible AI in practice: harm triage, fairness metrics, model cards, oversight. Mapped to NIST AI RMF and ISO 42001
What you’ll learn
⚡ Name any AI harm precisely: which kind, to whom, how badly, and whether it can be undone, in words that survive a room full of lawyers
⚡ Tell allocative harm from representational harm in thirty seconds, and spot the four kinds of harm that never produce an error message
⚡ Choose a fairness criterion and defend it, knowing why demographic parity, equal opportunity and calibration cannot all hold at once
⚡ Run disaggregated evaluation by subgroup and turn fairness into a test that fails the build, instead of a paragraph in a report
⚡ Audit where bias actually enters: the target variable, representation, conditions and absence, rather than only testing the model
⚡ Tell honest transparency from decorative explanation, and write disclosure that helps a person act on a decision made about them
⚡ Read a model card, system card or data statement like an auditor, and catch generic limitations that are worse than saying nothing
⚡ Apply purpose limitation and the secondary-use test, and answer a deletion request when the data is already inside a trained model
⚡ Measure whether human oversight is real, design review that resists automation bias, and name where persuasion becomes manipulation
⚡ Run a one-hour AI ethics review mapped to the EU AI Act, NIST AI RMF and ISO 42001 that ends in a decision, not a discussion
Requirements
❗ You have seen an AI system in production, or close to it.
❗ You can read a system diagram.
❗ No statistics, philosophy or law required. Where the course touches a regulation it explains the shape of the obligation, not the citation.
❗ Helpful if you have sat in a meeting where someone said "we should probably think about the ethics" and nobody knew the next sentence.
Description
Most AI ethics training stops at a values statement. This course is about the decisions that get made anyway, usually by whoever happens to be in the room.
If you are the person who gets asked "is this okay to ship?", this was written for you. Not as philosophy, and not as a compliance recital. As a set of moves you can make in a meeting, with words that survive a room containing lawyers, engineers and a product owner at the same time.
The uncomfortable premise is that ethics is decided at design time and discovered later. By the time anyone convenes a review, most of the outcomes are already fixed by choices nobody flagged as ethical: what the target variable is, who the system was aimed at, which data was already lying around. So this course works on those choices, in the order you actually meet them.
WHAT YOU WILL BE ABLE TO DO
Name the harm precisely. There are two families of harm and only one of them gets measured. You will learn to tell allocative from representational harm in thirty seconds, recognise the four kinds of harm that never produce an error message, and triage any system on three axes instead of a single meaningless score.
Choose a fairness criterion and defend it. Demographic parity, equal opportunity and calibration are all reasonable, and they cannot all hold at once. That is a theorem, not an oversight. You will see the conflict on one slide, learn which criterion fits which decision type, and write a fairness statement that outlives your tenure. Then you will turn it into disaggregated evaluation by subgroup, as a test that fails the build.
Tell honest transparency from decorative transparency. The second is the far more common failure. You will learn the five things a decision subject is owed, why an explanation may not describe the model at all, how to read a model card like an auditor, and why generic limitations sections are worse than none. Overclaiming gets its own lecture, because it is now enforcement territory.
Handle data with some dignity. Provenance questions for any dataset. The three ways a model leaks the people in its training data, and why rare records are remembered best. Purpose limitation and the secondary-use test. What happens when "delete my data" meets a model that has already been trained.
Keep humans actually in the loop. A human in the loop is not a control. People defer to machines, reliably and measurably. You will learn to design review that resists that deference, measure whether your oversight is real, name the line where persuasion becomes manipulation, and close the accountability gap with one named person before launch.
Run the review. The final section maps everything onto the EU AI Act, the NIST AI Risk Management Framework and ISO 42001, gives you a one-page responsible-use policy people follow rather than route around, and ends with an eight-question review you can run on one real system in an hour and leave with a decision.
WHAT THIS COURSE IS NOT
It is not a values statement. It is not an argument about whether AI is good. It assumes no statistics, no philosophy and no law. Where it touches a regulation it explains the shape of the obligation rather than reciting the citation. Where it touches a fairness metric it explains what the number means rather than how to derive it.
Twenty-five short lectures, none longer than about six minutes, built so you can watch one before a review and use it the same afternoon. By the end you will have a vocabulary that holds up under challenge, a method you can repeat, and an artefact to leave behind.
Who this course is for
⭐ Security, risk and GRC practitioners who already run reviews and need the one dimension they were never trained on.
⭐ Architects and engineering leads who make the design calls that decide whether a harm is possible at all.
⭐ Product owners and programme managers, because choosing what a system is for and who it is aimed at does more ethical work than any model card.
⭐ Anyone writing the responsible-use policy who wants one people follow instead of route around
⭐ Not for you if you want moral philosophy, legal citations, or code for fairness libraries.
https://rapidgator.net/file/778efcde47db411341a45cbf7e82924a/AI_Ethics_in_Practice_Fairness,_Bias_&_the_EU_AI_Act.rar.html