Subjects and lessons to explore.

Law

Philosophy & Ethics · Foundations

AI Ethics

9 min read 5 practice questions
8 key termsSourcesReport an error
Want it in plain words first? Jump to Eli explains — the same idea, no jargon.
On this page 9 sections
  1. In 30 seconds
  2. Why this matters
  3. The college version
  4. Eli explains
  5. Worked example
  6. Key takeaway
  7. Quick check
  8. Study tools
  9. Sources & references

In 30 seconds

is the branch of applied ethics that studies moral questions raised by artificial intelligence: systems that decide, recommend, or predict. Its main issue clusters include fairness and bias, and , accountability for harms, privacy and surveillance, safety and , and effects on work. Since about 2015, governments, companies, and professional bodies have published many AI ethics principles, but most are voluntary. Whether principles can change practice, and who should enforce them, is itself an open debate.

Why this matters

Automated systems already influence who gets a loan, a job interview, a medical recommendation, or a housing application. These are morally significant decisions, and the systems behind them can reproduce bias, resist explanation, and spread responsibility across many actors. Studying AI ethics builds the habit of asking who is affected, what a system's limits are, and who answers for outcomes. It also clarifies why written principles alone do not settle the matter: implementation, enforcement, and power decide whether values become practice.

The college version

What AI ethics is

AI ethics is the branch of applied ethics that examines moral questions raised by the design, deployment, and use of artificial intelligence. It is useful to treat AI systems as decision systems: they take in data and produce outputs that recommend, rank, or decide, and those outputs can affect real people. The Stanford Encyclopedia of Philosophy observes that the field is difficult because it requires normative ethics, an understanding of the technologies, and expertise in the many areas of social impact. The field is commonly divided into short-term questions about current systems, such as bias, privacy, and accountability, and long-term questions about future systems, such as safety and superintelligence. Scholars also caution that discussion should not be dominated by imagined technology rather than the systems that actually exist.

Fairness, bias, and the quality of data

Automated systems can reproduce or amplify bias. NIST, the U.S. standards agency, reports that biases are endemic across technology processes and can lead to harmful impacts regardless of intent, and that AI can newly introduce or amplify biases. Bias can enter at several points: through the training data, when a dataset under-represents a group or encodes past discrimination; through design choices, such as which features a system uses or how success is defined; and through feedback, when a system's own outputs shape the future data it receives. Documented cases reported in the research literature include gender bias in hiring tools, racial disparities in credit and parole risk assessments, and facial recognition systems performing worse on darker skin tones. Remedies are not purely technical: NIST identifies systemic, statistical, and human categories of bias, and philosophers note that competing mathematical definitions of fairness are hard to reconcile.

Opacity, explainability, and the trade-off debate

Many AI systems are opaque, or black boxes: the person affected by a decision, such as a denied loan or a rejected application, often cannot know how the output was produced, and with some machine-learning systems even the programmers do not fully know which patterns in the data drove the result. The ACM identifies three sources of opacity: technical, when an algorithm does not lend itself to easy explanation; economic, when transparency is costly or would compromise trade secrets; and social, when revealing inputs would violate privacy expectations. Some ethicists argue that explainability is a basic requirement, because decisions about important parts of people's lives should be justifiable to them. Others argue that what matters is whether machine decisions are well calibrated, and that a human-style explanation may not be what is actually needed. Explainable AI is now a significant technical field, but whether explanation, justification, or calibration should be the goal remains an open debate.

Accountability, privacy, and surveillance

When an automated system harms someone, who answers? Responsibility is distributed across many actors, including data collectors, developers, deployers, users, and regulators. Philosophers have discussed a : in some cases neither the machine nor any individual human appears clearly responsible for the harm. Proposed remedies include keeping humans in the loop or in meaningful control, though whether that genuinely solves the allocation problem is debated, and some argue the difficulty is only temporary confusion about how to allocate responsibility among humans. Privacy is a related cluster. AI depends on large amounts of personal data, and surveillance can be conducted by states, companies, and individuals; the encyclopedia literature describes regulation as slow to respond to these changes.

Safety, work, and the limits of principles

Technical AI safety, sometimes called alignment, is an active research area closely related to AI ethics but not identical to it; its definitions and goals are contested, so this lesson reports the debate rather than endorsing any position. On labor, responses to range from alarmed to neutral to optimistic. Historical experience shows that productivity gains have not necessarily meant net job loss, though job polarization, growth at the high- and low-skill ends of the labor market with pressure on mid-skill jobs, is a documented pattern. Since roughly 2015 to 2020, governments, the European Union, companies, and professional associations have published many sets of AI ethics principles. A comparative review of 84 such sets found strong convergence on transparency, justice and fairness, non-maleficence, responsibility, and privacy. Yet most of these documents are voluntary, and critics describe some as or toothless, while defenders treat them as useful work in progress. Implementation, enforcement, and power remain the central limits of principle-based approaches, which is why some jurisdictions, such as the EU with its AI Act, have begun to move toward binding regulation.

Eli, the EliExplains learning guide

Eli explains

The same idea, in plain words

Explain it like I’m 10

AI ethics is the study of right and wrong in machines that decide things for us. A computer program that approves loans, ranks job applicants, or suggests medical treatments can be helpful, and it can also carry forward old unfairness without meaning to. The problems fall into a few big questions. Is the system treating everyone fairly, or just copying the bias in its data? Can anyone explain why it made its decision? If something goes wrong, who is responsible? What happens to privacy and jobs as these systems spread? Around the world, organizations have written down principles for ethical AI: be transparent, be fair, be accountable. But a principle is a promise, not a rule with teeth. Most of these documents are voluntary, so a central question in AI ethics is how to turn good intentions into practice that is actually checked and enforced.

Picture it like this

Think of an AI system as a very fast copy machine. It does not create the document from nothing; it reproduces whatever is in the original. If the original pages, the training data, contain smudges and missing lines, the copies will contain them too, only faster and in bulk. What looks like a neutral machine is really reproducing the quality of what it was given.

Where the picture stops working

A copy machine copies exactly, while an AI system learns patterns. That means it can do more than copy: it can amplify a small pattern into a large one, or find patterns the data did not obviously contain. So the machine is not passive, and fixing the input is only part of the solution.

Worked example

Consider a hypothetical city that replaces its paper rental-application screening with an automated system trained on a decade of past approvals. The goal is speed and consistency, but the old records reflect years of informal practices, including agents steering applicants of some backgrounds toward certain buildings. The new system learns patterns from that biased past. An applicant denied by the system cannot get a clear reason, because the model's outputs are opaque even to the vendor's engineers. The city council then asks who is responsible if the system discriminates: the data provider, the vendor, the agency that deployed it, or the official who approved it. The case shows three clusters working together: bias in data, opacity in decisions, and a responsibility gap among many hands.

Key takeaway

AI ethics examines fairness, bias, opacity, accountability, privacy, safety, and the labor effects of automated systems; most current AI ethics principles are voluntary, so implementation, enforcement, and power remain open, contested questions.

Quick check

3 questions here, of 5 in this lesson’s practice set. Answers stay hidden until you check.

Question 1 of 3foundational

In AI ethics, the 'black box problem' refers to which situation?

Choose an answer, then check it.
Question 2 of 3intermediate

According to a comparative review of 84 sets of AI ethics guidelines, which statement best describes the findings?

Choose an answer, then check it.
Question 3 of 3intermediate

A company trains its hiring screener on decades of its own past hiring records, which reflect old patterns of favoring certain applicant backgrounds, and the screener now ranks applicants accordingly. Which ethical concern does this best illustrate?

Choose an answer, then check it.
Practice all 5

Keep learning

Ready to build on this? Continue to the next lesson.

Practice this lesson
Study tools & related lessonsYou’ll learn to · Common mistakes · Easily confused · Key vocabulary · Related

You’ll learn to

  • Define AI ethics as the field of applied ethics examining moral questions raised by artificial intelligence systems.
  • Explain how automated systems can reproduce or amplify bias in training data and design.
  • Distinguish transparency from explainability and describe the debate over whether opaque decisions are ever acceptable.
  • Analyze the many-hands problem of allocating responsibility when AI systems cause harm.
  • Describe the landscape of AI ethics principles and evaluate the limits of principle-based approaches.
  • Summarize the main positions in debates about AI safety and labor effects without adjudicating between them.

Common mistakes

  • Assuming machines are neutral and objective because they are computers.

    Bias can enter through training data and design choices, and harmful outcomes can occur regardless of intent; a system can reproduce or amplify patterns in what it was given.

  • Assuming transparency always fixes everything, or that opacity is always wrong.

    The trade-off is debated: transparency has costs such as trade secrets and privacy, and some ethicists argue well-calibrated decisions matter more than explanations.

  • Blaming the AI itself as if it were a moral agent.

    In the standard view, machines are not moral agents; responsibility is distributed among the humans and institutions that collect data, build, deploy, and regulate the system.

  • Treating published AI ethics principles as enforceable law.

    Most principles are voluntary and non-binding; critics call them ethics washing or toothless, and enforcement and implementation are open problems.

  • Treating contested long-term scenarios as settled facts.

    Safety and alignment are active research areas with contested definitions; the honest move is to report the debate and attribute positions.

Easily confused

AI ethics vs. Technology ethics generally

AI ethics focuses on systems that decide, recommend, or predict, and their fairness, opacity, and accountability; technology ethics covers the broader moral questions of technology as a whole, including environmental and professional concerns.

Transparency vs. Explainability

Transparency is about access to how a system works, while explainability is about whether its specific decisions can be understood and justified to those affected.

Voluntary AI principles vs. Binding regulation

Principles are aspirational commitments with weak enforcement, while regulation such as the EU AI Act carries legal obligations, though its effectiveness is debated.

Key vocabulary

AI ethics
The branch of applied ethics that examines moral questions raised by the design, deployment, and use of artificial intelligence systems.
Algorithmic bias
Systematic unfairness in an automated system's outputs, which can enter through training data, design choices, or feedback effects.
Opacity
The condition in which the reasons for an AI system's output cannot be understood by the people affected, and sometimes not even by its developers; also called the black box problem.
Explainability
The quality of an AI system's decisions being describable in terms humans can understand, proposed by some ethicists as a requirement for acceptable automated decision-making.
Responsibility gap
The situation, discussed by philosophers, in which neither the machine nor any individual human appears clearly responsible for an AI-related harm.
Alignment
A contested term for the research goal of making AI systems' behavior match human values and intentions; an active area of technical AI safety.
Automation
The use of machines and software to perform tasks previously done by humans, with debated consequences for employment and the distribution of wealth.
Ethics washing
A critical term for using ethical language or guidelines to create an appearance of responsibility without meaningful change in practice.

Sources & references

  1. Ethics of Artificial Intelligence and Robotics — Stanford Encyclopedia of Philosophy
  2. Ethics of Artificial Intelligence — Internet Encyclopedia of Philosophy
  3. Towards a Standard for Identifying and Managing Bias in Artificial Intelligence (NIST Special Publication 1270) — National Institute of Standards and Technology, U.S. Department of Commerce
  4. Statement on Algorithmic Transparency and Accountability — ACM U.S. Public Policy Council and ACM Europe Policy Committee

EliExplains lessons are original prose written from the open, credible references above. See Copyright & Licensing.

Researched 2026-08-21

Educational content only. It is not medical, legal or professional advice. Found an error? Tell us.