Artificial Intelligence Ethics

Introduction

Artificial Intelligence Ethics examines how artificial intelligence should be designed, developed, deployed, governed, and used when its effects reach people, institutions, society, and the environment. For learners in Grades 11–13, the topic connects Computer science, Ethics, Politics, Economics, Media literacy, Data protection, and Human rights.

AI ethics is not only about asking whether an algorithm is technically accurate. You also need to ask who chose the goal, which data were used, who benefits, who may be harmed, which rights are affected, how errors can be challenged, and who remains responsible when an AI-supported decision goes wrong. An AI system is therefore best understood as part of a socio-technical system: models, data, interfaces, institutions, rules, incentives, and human decisions interact.

The diagram above illustrates a neural-network structure. Ethical questions begin before such a model is trained and continue long after it is deployed. A technically sophisticated model can still be unethical if the surrounding system creates unfair outcomes, invades privacy, misleads users, or gives people no meaningful way to appeal.

Learning Goals

By the end of this aiMOOC, you should be able to:

Distinguish ethical, legal, and technical questions when evaluating an AI system.
Explain how bias can enter through data, objectives, labels, model design, deployment, and feedback loops.
Evaluate privacy and surveillance risks using ideas such as data minimization, purpose limitation, consent, and security.
Compare transparency and explainability with other goals such as accuracy, privacy, safety, and intellectual property protection.
Assign responsibility and oversight across developers, deployers, institutions, regulators, and users.
Design safeguards that reduce harm while preserving legitimate benefits.
Defend a reasoned position when important values conflict.

Why AI Ethics Matters

AI systems increasingly help produce text and images, recommend content, detect fraud, rank applicants, support medical decisions, predict equipment failures, personalize learning, recognize faces, and automate parts of public and private services. These uses can create real benefits, but the same technologies may also amplify discrimination, expose personal data, encourage overreliance, concentrate power, spread synthetic misinformation, or create environmental costs.

Ethics becomes especially important when an AI output affects a person's opportunities, safety, reputation, access to services, or ability to participate in society. High performance on a test set does not automatically prove that a system is safe or fair in the real world.

A useful starting question is: What would responsible use require in this specific context? The answer depends on the purpose of the system, the seriousness of possible harm, the people affected, available alternatives, and the safeguards that can realistically be enforced.

AI as a Socio-Technical System

AI ethics often fails when people look only at the model. Consider an automated hiring tool. Its outcomes may depend on how a job description is written, who applies, what historical examples are treated as successful, which variables are available, what the optimization target rewards, how recruiters interpret scores, and whether rejected applicants can appeal. Bias can therefore enter before, during, or after model training.

You can analyze an AI system across a lifecycle:

Lifecycle stage	Typical ethical questions
Problem definition	Is AI needed at all? Is the goal legitimate, proportionate, and clearly defined?
Data collection	Are the data representative, lawful to use, relevant, secure, and collected with appropriate respect for privacy?
Model development	Which errors matter most? Are proxy variables introducing unfairness? Can performance be tested across relevant groups?
Deployment	Does the system work in the real setting? Are users informed about its limits? Is human oversight meaningful?
Monitoring	Are errors, incidents, distribution shifts, and unequal impacts measured after release?
Retirement	Can the system be withdrawn safely? What happens to stored data, logs, and people affected by earlier decisions?

This lifecycle view prevents the misleading idea that ethics is a single checklist completed at the end.

Core Ethical Principles

Different ethical frameworks use different terms, but several themes appear repeatedly in international guidance and professional practice. These themes do not automatically produce one correct answer. Instead, they help you identify values, evidence, conflicts, and responsibilities.

Human Rights and Human Dignity

A human-rights approach asks whether an AI system respects dignity, equality, privacy, freedom of expression, access to remedy, and other protected interests. UNESCO's 2021 Recommendation on the Ethics of Artificial Intelligence places human rights and human dignity at the center of its framework.

Human dignity also means that a person should not be reduced to a score or profile when the decision has serious consequences. Statistical predictions can support a decision, but they do not capture the full moral worth or individual circumstances of a human being.

UNESCO: Recommendation on the Ethics of Artificial Intelligence

Fairness and Non-Discrimination

Fairness is not a single mathematical property. In many real situations, different fairness criteria can conflict. For example, equal error rates between groups, equal acceptance rates, and equally calibrated risk scores are different goals. Which one matters most depends on the context, the decision, historical inequalities, and the harm caused by different kinds of error.

Bias can enter through:

Sampling: some groups are underrepresented or measured differently.
Labels: historical decisions are treated as ground truth even when the past was unfair.
Proxies: apparently neutral variables may correlate with protected characteristics.
Objectives: the system may optimize efficiency while ignoring unequal harms.
Deployment context: a model trained for one population may be used in another.
Feedback loops: past outputs influence future data and reinforce earlier patterns.

Facial recognition is a useful case for discussing measurement, representation, privacy, consent, and unequal error rates. You should avoid jumping from a single accuracy number to a moral conclusion. Ask what task is being performed, who is affected by false matches or missed matches, what alternatives exist, and whether the use is proportionate.

The Crash Course video above introduces several ways algorithmic bias can arise. While watching, distinguish between bias in the data, bias in the objective, and bias created by the social setting in which an AI system is used.

Privacy and Data Governance

AI systems can infer patterns from large amounts of data, including information people did not expect to reveal. Privacy therefore involves more than hiding a name. Seemingly ordinary data points can sometimes be combined to infer location, health status, preferences, relationships, or identity.

Useful privacy questions include: What data are collected? Why are they needed? How long are they kept? Who can access them? Can people refuse or withdraw? Could less data achieve the same goal? How are data protected against misuse or breach?

Data minimization means collecting and retaining only what is reasonably needed for the stated purpose. Purpose limitation means that data collected for one reason should not silently be repurposed for unrelated uses without an appropriate basis. Exact legal requirements differ by jurisdiction, but these ideas are widely important in responsible data governance.

AI-enabled surveillance creates a tension between possible benefits such as security or traffic management and risks such as pervasive tracking, chilling effects, discriminatory targeting, and loss of anonymity in public space. Ethical analysis requires both evidence about effectiveness and scrutiny of power.

Transparency, Explainability, and Traceability

Transparency can mean that people are told an AI system is being used, what its purpose is, what data it relies on, what its limits are, and who is responsible for it. Explainability concerns whether relevant aspects of an output or decision can be made understandable to humans. Traceability concerns records that allow important data, decisions, versions, and actions to be reconstructed later.

Transparency does not always mean publishing every line of source code. A useful explanation should be appropriate to the audience. A developer may need technical documentation; a person denied a service may need understandable reasons and a route to challenge the decision.

Trade-offs are real. Releasing too much technical detail can create security or privacy risks. A highly interpretable model may sometimes perform differently from a more complex model. Ethical design therefore asks what level and type of explanation is necessary for the decision at stake.

Accountability, Human Oversight, and Redress

Accountability means that identifiable people or organizations remain responsible for decisions, monitoring, and corrective action. Saying "the algorithm decided" is not an adequate transfer of moral responsibility.

Meaningful human oversight requires more than placing a person at the end of an automated process. The reviewer needs time, authority, competence, information, and a realistic ability to disagree with the system. Otherwise the human may become a rubber stamp.

Automation bias occurs when people trust automated outputs too readily. The opposite error is also possible: humans may ignore useful automated evidence without good reason. Responsible oversight therefore needs training, clear escalation rules, and regular evaluation.

Redress means that people affected by a harmful or incorrect decision have a way to ask for review, correction, explanation, or remedy. Redress is central when an AI system influences education, employment, credit, healthcare, public benefits, or other important opportunities.

Safety, Security, and Robustness

An AI system may work well during normal testing and still fail under unusual conditions, malicious inputs, new populations, or changing environments. Robustness concerns whether the system continues to behave acceptably when conditions vary. Security concerns protection against attacks, manipulation, unauthorized access, and misuse.

Generative AI adds further risks. A system may produce confident but false statements, generate unsafe instructions, reveal sensitive information, imitate people, or be deliberately used to create deceptive content. Evaluating these systems requires tests for both accidental failure and intentional misuse.

The Computerphile video introduces AI safety as a design problem. After viewing, ask which concerns are immediate and measurable today, which concern future systems, and what kinds of evidence would justify different safeguards.

Autonomy and Human Agency

AI can support human choice, but it can also shape choice. Recommendation systems influence what people see; predictive systems can narrow which options are offered; conversational systems can sound authoritative even when uncertain.

Respect for autonomy requires attention to manipulation, dependency, deceptive interface design, and the possibility that users may not understand when a system is generating or ranking content. Human agency is stronger when users know what the system can and cannot do, can access alternatives, and can contest important outcomes.

Sustainability and Environmental Impact

AI requires physical infrastructure. Training and operating large models can consume electricity, water for cooling, computing hardware, and materials. The environmental impact depends on many factors, including model size, hardware efficiency, data-center location, energy source, utilization, and how often the system is run.

Ethical sustainability asks whether the social benefit of an AI application justifies its resource use, whether more efficient approaches are available, and whether environmental costs are shifted onto communities with little influence over the decision.

You should avoid simplistic claims such as "all AI is environmentally harmful" or "digital systems have no material footprint." Instead, compare realistic alternatives and look for measured evidence.

Ethical Problems in Generative AI

Generative AI can create text, images, audio, video, software code, and other media. Its ethical challenges overlap with older AI systems but also introduce new questions about authorship, verification, attribution, synthetic media, and the scale at which content can be produced.

Misinformation, Deepfakes, and Information Integrity

A deepfake or synthetic text is not automatically harmful. Satire, accessibility tools, film production, language learning, and privacy-preserving avatars can use synthetic media legitimately. Harm becomes more likely when realistic media are used to deceive, impersonate, defraud, harass, manipulate public debate, or fabricate evidence.

Responsible practice includes labeling relevant synthetic content, preserving provenance where possible, verifying important claims against independent sources, and resisting the temptation to treat visual realism as proof.

Media literacy now requires a stronger question than "Does this look real?" You should ask: Who published it? What evidence supports it? Can the claim be verified elsewhere? Is the original source available? Has the material been edited, generated, or taken out of context?

Copyright, Authorship, and Creative Work

AI systems can be trained on or generate material connected to copyrighted works, performers, artists, writers, programmers, and photographers. Ethical questions include consent, licensing, attribution, compensation, imitation of living creators, and whether a generated work misleads people about its origin.

Copyright law differs across jurisdictions and continues to develop. Ethical judgment should therefore distinguish between what is legally permitted, what is contractually permitted, and what is fair to creators or communities.

In school, transparent use is especially important. If an AI tool substantially contributes to an assignment, your teacher should be able to understand how it was used. Passing generated work off as entirely your own can undermine assessment even when no law has been broken.

Education and Academic Integrity

Generative AI can support brainstorming, language feedback, explanation, coding practice, simulation, and accessibility. It can also encourage shallow learning, fabricated citations, overreliance, hidden outsourcing of work, and privacy risks when students upload personal or confidential data.

A responsible classroom policy should define permitted uses, required disclosure, privacy limits, expectations for verification, and which parts of an assessment must demonstrate unaided understanding. UNESCO's guidance for generative AI in education emphasizes a human-centered and age-appropriate approach.

UNESCO: Guidance for Generative AI in Education and Research

Moral Reasoning and AI

AI ethics becomes difficult when important values conflict. A system may be more accurate but less explainable. Stronger privacy may reduce the data available for certain forms of testing. More automation may improve speed while weakening human control. Ethical reasoning helps you make these trade-offs explicit rather than hiding them inside technical choices.

Consequences, Duties, and Virtues

Three broad ethical traditions offer different questions:

Ethical approach	Main question	AI example
Consequentialism	Which option is likely to produce the best overall consequences and reduce harm?	Compare expected benefits and harms of an AI triage tool.
Deontological ethics	Which duties, rights, or rules must be respected even if breaking them seems efficient?	Protect privacy or equal treatment even when surveillance could increase convenience.
Virtue ethics	What would a responsible, honest, just, and prudent actor do?	Ask whether a developer is communicating uncertainty honestly instead of overselling a system.

These approaches can support different conclusions. Strong ethical analysis explains which values are being prioritized and why.

The Trolley Problem and Its Limits

The Trolley problem is often used when discussing autonomous vehicles because it presents a dramatic conflict between harms. It is useful for practicing moral reasoning, but real engineering decisions are usually less clean. Designers must consider perception errors, braking distance, traffic law, uncertainty, system testing, accountability, and prevention long before a split-second crash choice occurs.

A strong AI ethics discussion therefore moves beyond dramatic thought experiments to the full system lifecycle.

A Practical AI Ethics Impact Assessment

When you evaluate an AI system, use evidence rather than slogans. The following sequence can guide a classroom audit.

Step	Questions to ask
Define the purpose	What problem is the system supposed to solve, and is AI necessary?
Map stakeholders	Who builds, buys, operates, uses, experiences, or is indirectly affected by the system?
Identify benefits	What measurable improvements are expected, and for whom?
Identify harms	What could go wrong through error, bias, misuse, exclusion, surveillance, or overreliance?
Assess severity and likelihood	Which harms would be most serious, and what evidence exists about how likely they are?
Examine distribution	Are benefits and burdens distributed fairly across groups and communities?
Design safeguards	Which technical, organizational, legal, educational, or human controls reduce risk?
Plan oversight	Who monitors the system, which indicators are tracked, and when must use stop?
Provide redress	How can an affected person ask for explanation, correction, review, or compensation?
Reassess	What new evidence would require retraining, redesign, limitation, or withdrawal?

A risk assessment is not complete merely because a form was filled in. Evidence must be updated as the system and its context change.

Mini Case Study: AI Hiring Screen

Imagine that a company uses an AI tool to rank job applicants. The tool appears to save time and predicts which applicants resemble previous successful employees.

A superficial evaluation might ask only whether the predictions are accurate. A deeper ethical evaluation asks whether past employment decisions were biased, whether the training sample reflects the future workforce, whether proxies encode protected characteristics, how false rejections are distributed, whether candidates know AI is being used, whether a recruiter can override the score, and whether rejected applicants can seek review.

The key lesson is that accuracy is one property of a system, not a complete ethical verdict.

Governance: Turning Principles into Practice

Ethical principles become useful only when institutions translate them into procedures, documentation, testing, monitoring, and accountability.

UNESCO, OECD, and NIST

The UNESCO Recommendation on the Ethics of Artificial Intelligence was adopted in 2021 and uses a human-rights-centered approach. Its themes include proportionality, safety, privacy, fairness, transparency, human oversight, accountability, sustainability, and AI literacy.

The OECD AI Principles, first adopted in 2019 and updated in 2024, promote trustworthy AI that respects human rights and democratic values. They emphasize inclusive well-being, fairness and privacy, transparency and explainability, robustness and safety, and accountability.

The NIST AI Risk Management Framework 1.0 is a voluntary framework for managing AI risks. It describes trustworthy AI using characteristics such as validity and reliability, safety, security and resilience, accountability and transparency, explainability and interpretability, privacy enhancement, and fairness with harmful bias managed.

These frameworks overlap but are not identical. Comparing them teaches you that responsible AI is not defined by one organization or one checklist.

The UNESCO video above discusses AI ethics, governance, inequalities, and the challenge of translating principles into practice.

The European Union AI Act as a Legal Example

Ethics and law overlap, but they are not the same. The European Union AI Act provides a useful legal example of risk-based AI regulation. It includes prohibited practices, obligations for certain high-risk systems, rules for general-purpose AI, and transparency duties for specific uses.

At the time this course was checked in August 2026, major enforcement powers and transparency requirements had begun to apply from 2 August 2026, while some obligations have later timelines. Because regulation changes, always consult the current official text rather than relying on an old summary.

European Commission: AI Act overview

Legal compliance is a minimum requirement, not a complete ethical judgment. A system can be lawful yet still deserve criticism for being wasteful, manipulative, inaccessible, or socially harmful.

Ethical Design Strategies

Responsible AI is easier to achieve when ethics is built into design rather than added after deployment.

Useful strategies include participatory design, diverse and documented datasets, pre-deployment testing, subgroup evaluation where appropriate, red-team exercises, privacy-preserving techniques, security reviews, understandable user interfaces, logging, independent audits, incident reporting, human override, appeal procedures, accessibility testing, and regular reassessment.

No single safeguard is sufficient. For example, a "human in the loop" does not solve bias if the reviewer lacks time or authority. A fairness metric does not solve an unjust goal. A privacy notice does not justify unnecessary data collection. Ethical design requires multiple layers of protection.

A Four-Question Test

Before approving an AI use, ask four connected questions:

Purpose: Is the goal legitimate, necessary, and proportionate?
Evidence: What do we know about performance, limitations, and unequal impacts?
Power: Who controls the system, who can challenge it, and who bears the risk?
Safeguards: What prevents harm, detects failure, and provides remedy?

If you cannot answer these questions with evidence, the system may not be ready for responsible deployment.

Interactive Tasks

Quiz: Test Your Knowledge

Which idea best describes algorithmic fairness? (Fairness depends on context and may require comparing different impacts and error patterns) (!Fairness means every AI system must give identical outputs to everyone) (!Fairness is guaranteed whenever a model has high overall accuracy) (!Fairness is only a question about computer speed)

Which example shows bias entering before model training? (A biased historical labeling process is treated as ground truth) (!A monitor cable becomes disconnected) (!A server fan runs faster than expected) (!A user changes the screen brightness)

What is the main idea of data minimization? (Collect and retain only data reasonably needed for the stated purpose) (!Collect every available data point in case it becomes useful) (!Keep all data permanently) (!Publish personal data so everyone can inspect it)

What does explainability aim to provide? (Understandable information about relevant reasons or processes behind an AI output) (!A guarantee that every AI prediction is correct) (!A requirement that all software must be open source) (!A promise that no data will ever be collected)

When is human oversight most meaningful? (When a trained reviewer has information authority and time to challenge the AI output) (!When a person automatically approves every AI recommendation) (!When a human name appears on a report but no review occurs) (!When the AI hides uncertainty from the reviewer)

What is the purpose of the NIST AI Risk Management Framework? (To help organizations manage AI risks and trustworthiness considerations) (!To replace every national law with one global law) (!To certify that any model is perfectly safe) (!To ban all uses of machine learning)

Which idea is central to UNESCO's AI ethics recommendation? (Human rights and human dignity should guide responsible AI) (!AI systems should always replace human judgment) (!Privacy should be ignored when automation is efficient) (!Only technical experts should influence AI governance)

Which statement best describes the European Union AI Act? (It uses a risk-based legal approach with different obligations for different AI uses) (!It is a voluntary classroom ethics code with no legal effect) (!It requires the same rule for every software program) (!It prohibits all generative AI systems)

Why can deepfakes create an ethical problem? (They can be used to deceive impersonate or fabricate evidence) (!Every edited image is automatically illegal) (!Synthetic media can never have a legitimate use) (!Deepfakes only affect computer hardware)

How are law and ethics related in AI? (Legal compliance can be necessary while ethical responsibility may require more) (!Anything legal is automatically ethical) (!Ethics matters only when no laws exist) (!Ethical analysis can ignore rights and consequences)

Memory Game

Algorithmic bias	Systematic skew that can create unfair outcomes
Data minimization	Limiting collection to information reasonably needed for a purpose
Explainability	Making relevant reasons or model behavior understandable to people
Human oversight	Meaningful human capacity to review challenge or stop an automated process
Redress	A route for affected people to seek review correction or remedy
Traceability	Records that allow important decisions data and system versions to be reconstructed
Robustness	Ability to maintain acceptable behavior under changing or difficult conditions
Proportionality	Matching the use and safeguards of AI to the seriousness of the goal and risk

Drag and Drop

Match the correct terms.	Topic
Can carry historical patterns or underrepresentation into learning	Training data
Can encode priorities through objectives features and error costs	Model design
Can change who is affected and whether earlier tests remain valid	Deployment context
Can catch errors only when reviewers have real authority and information	Human review
Tracks performance incidents and unequal impacts after release	Monitoring

...

Crossword Puzzle

Fairness	Which principle asks whether benefits errors and burdens are distributed justly?
Privacy	Which value concerns control protection and appropriate use of personal information?
Oversight	What is meaningful human review of an AI-supported process called?
Accountability	Which principle requires identifiable responsibility for decisions and harms?
Transparency	Which principle concerns openness about an AI system's use purpose and limits?
Sustainability	Which principle asks about long-term environmental and social resource impacts?

LearningApps

Cloze Text

Artificial intelligence ethics examines how AI affects people institutions society and the

. A responsible evaluation considers the whole

. Unfair outcomes can result from biased data objectives labels deployment or

. Privacy can be strengthened through measures such as data

. Meaningful human review requires authority competence and time for

. People affected by serious errors should have access to

. International frameworks repeatedly emphasize human rights fairness transparency safety and

. Ethical responsibility can extend beyond minimum legal

.

Open-Ended Tasks

Easy

AI Ethics Diary: For one week, record four situations in which you encounter an AI-supported system and write what benefit, risk, and stakeholder you notice in each case.
Bias Example Poster: Create an original poster or infographic that shows three different ways bias can enter an AI system, and add one practical safeguard for each.
Synthetic Media Check: Compare one authentic media item with one openly labeled synthetic example and produce a short checklist explaining how you would verify each before sharing it.
Classroom AI Interview: Interview a teacher, librarian, or classmate about responsible AI use in education and summarize the person's strongest benefit, concern, and proposed rule.

Standard

Mini Algorithm Audit: Use a small public or synthetic dataset to compare error patterns across two clearly defined groups, then explain why your chosen fairness measure is useful and what it does not prove.
Privacy Redesign: Choose a fictional AI app that collects excessive information and redesign its data flow so that it follows data minimization and clearer consent.
Ethics Debate Video: Produce a three-to-five-minute video in which two speakers argue opposite positions on AI-enabled facial recognition in a public setting, then add a final evidence-based synthesis.
Stakeholder Mapping Visit: Visit a school office, library, local company, public administration office, science museum, or other relevant place and map where algorithmic decision support could affect staff, users, or visitors.

Advanced

AI Impact Assessment Project: Select a realistic high-impact AI use and write a structured impact assessment covering purpose, stakeholders, benefits, harms, distributional effects, safeguards, monitoring, and redress.
Policy Comparison Research: Compare the UNESCO Recommendation, OECD AI Principles, NIST AI Risk Management Framework, and one current legal regime, then explain where their priorities overlap and differ.
Human Oversight Experiment: Design a classroom experiment using harmless synthetic decisions to compare unaided human judgment, AI advice, and human review with AI advice, then analyze signs of automation bias.
Public Hearing Simulation: Organize a simulated hearing on whether a school or municipality should deploy a specific AI system, assign stakeholder roles, collect evidence, question witnesses, and produce a reasoned final recommendation.

Learning Assessment

Case Analysis Assessment: Analyze a fictional AI system used for school admissions and explain how data quality, fairness, privacy, explainability, oversight, and redress interact rather than treating them as separate checkboxes.
Fairness Trade-Off Assessment: Given two models with different group error rates, choose which evidence you would need before recommending either model and justify why one fairness metric alone is insufficient.
Governance Design Assessment: Design a responsibility map for developers, deployers, teachers or managers, affected users, auditors, and regulators in a high-impact AI scenario.
Ethical Argument Assessment: Write a reasoned argument that uses at least two ethical approaches to evaluate an AI use where efficiency conflicts with privacy or autonomy.
Incident Response Assessment: Respond to a fictional case in which an AI system causes repeated harmful errors after deployment and propose immediate containment, investigation, communication, redress, and long-term prevention.
Transfer Assessment: Apply the four-question test of purpose, evidence, power, and safeguards to a new technology not discussed in the course and explain what transfers from AI ethics to the new case.

Evidence of Learning

Strong evidence of learning should show more than recall. It should demonstrate that you can connect technical properties with human consequences, examine competing values, use evidence, and propose realistic safeguards.

Evidence type	What strong evidence looks like
Knowledge	You accurately explain fairness, privacy, transparency, accountability, safety, human oversight, redress, and sustainability in context.
Analysis	You identify stakeholders, benefits, harms, uncertainty, power relationships, and trade-offs across the AI lifecycle.
Reasoning	You justify conclusions with evidence and ethical principles instead of relying on slogans such as AI is always good or AI is always dangerous.
Practical skill	You can perform a basic impact assessment, inspect data or error patterns, and design safeguards that fit the risk.
Communication	You can explain an AI ethics issue clearly to both technical and non-technical audiences.
Product	You create an audit, policy, poster, video, debate, research comparison, or impact assessment that documents your reasoning.
Transfer	You apply the same ethical reasoning to a new AI application or another emerging technology and explain which principles still apply.

OERs on the Topic

The English Wikipedia article below provides an open starting point for further study. Use it critically and follow its references to primary and authoritative sources when preparing assessed work.

Further reliable starting points include the UNESCO Recommendation on the Ethics of Artificial Intelligence, the OECD AI Principles, the NIST AI Risk Management Framework, and the European Commission AI Act overview.

Linked Learning Areas

Artificial Intelligence Ethics

Artificial Intelligence Ethics connects technical understanding with questions of justice, responsibility, law, communication, and social impact. You should be able to move between the code-and-data level and the institutional level: a model's behavior matters, but so do the rules, incentives, interfaces, people, and power structures around it.