Week 11 of 12 · Beginner+

AI And Society: Work, Education, Law, Regulation

You examine how AI is changing jobs, schools, legal questions, and public rules, with a high-level look at the EU AI Act. You leave with a practical framework for using AI responsibly.

3 hours18 slides6 min read

Open slides Read study notes

Why Society Matters In A Technical Course Work: Tasks, Not Whole Jobs Education Law And Liability Copyright And Training Data Regulation And The EU AI Act Other Approaches And Responsible Use

You Will Be Able To

Describe how AI changes tasks inside jobs and which skills gain value.
Discuss benefits and risks of AI in education, and set personal rules for study use.
Summarize the main open questions about copyright and AI training data.
Explain the risk-based structure of the EU AI Act at a high level.
Apply a responsible-use checklist to a real AI project.

Why Society Matters In A Technical Course

For ten weeks you learned how AI systems work and how to build them. In week 9 you met the risks of a single system: errors, bias, and privacy loss. This week widens the view. Tools that write, summarize, and decide at scale affect workplaces, classrooms, courts, and governments.

People who build these systems make choices that shape those effects, so the discussion belongs in a builder's education. The aim here is to give you facts, a few clear frameworks, and honest uncertainty. Many questions are still open, and thoughtful people disagree.

Keep one habit in mind through the whole week: separate what is known from what is predicted. Court rulings and enacted laws are known. Forecasts about jobs are predictions, and good ones come with ranges and caveats.

Work: Tasks, Not Whole Jobs

A useful way to think about work is to break a job into tasks. Most jobs mix many tasks: writing, analyzing, talking, planning, and physical work. AI tends to take over or speed up some tasks inside a job and leave others to people. A paralegal may spend less time on first-draft summaries and more time checking sources and advising clients.

Studies from economists and from technology companies have tried to estimate which occupations have the most tasks that language models could assist. Such estimates carry wide uncertainty, since they measure what is technically possible and not what employers will actually do. History also shows that new technology both removes some roles and creates others, and that the transition can be hard for the people in the middle of it.

What gains value in this setting? Skills that combine with AI: asking clear questions, checking outputs, knowing a field deeply enough to catch errors, communicating with people, and taking responsibility for decisions. The checking skill matters most. Someone who can judge whether an AI answer is right is more useful than someone who can only produce answers.

List the tasks in your own job or study. Mark which ones AI could assist.
Decide which tasks you want to keep doing yourself because they build your skill.
Plan time each month to learn one new tool carefully.

Education

A model can explain a concept ten different ways, quiz you, and give feedback at any hour. Used well, it works like a patient tutor. Used badly, it does the thinking for you, and you learn little. Researchers describe a risk of offloading, where the effort that builds memory and understanding is skipped.

Schools and universities respond in different ways, from bans to guided use. Detection software for AI-written text is unreliable and can wrongly accuse students, so many institutions move toward assessments that show process: drafts, oral explanations, in-class work, and projects with reflection.

A personal rule that works well is to try the problem yourself first, then use AI to check, explain, or challenge your work. Always follow your institution's policy, and say openly when you used AI. Educators gain too, with help on lesson plans, examples, and feedback, as long as they check the output.

Law And Liability

When an AI system makes a harmful decision, who is responsible? The general principle in most legal systems is that people and organizations stay responsible, not the software. A company that deploys a hiring tool answers for its effects. A lawyer who files a brief with invented case citations from a chatbot answers to the court, and such incidents have led to sanctions in real cases.

Existing laws already apply to AI uses: consumer protection, anti-discrimination rules, product safety, and data protection such as the GDPR (week 9). New rules fill the gaps, and they differ by country. This is a general introduction, not legal advice. For a real product, consult a qualified lawyer in your jurisdiction.

Copyright And Training Data

Language and image models learn from huge collections of text and images, much of it protected by copyright. Two main questions are debated. First, is training a model on copyrighted works permitted, for example under doctrines such as fair use in the United States or text and data mining exceptions in the European Union? Courts in several countries are hearing cases, and the answers are not settled.

Second, who owns what the model produces? In the United States, the Copyright Office has stated that works need human authorship to receive protection, so material generated entirely by AI without meaningful human creative control is not protected, while human contributions such as selection and arrangement can be. Rules differ elsewhere.

Practical habits follow from this. Do not present AI output as your own research without checking it. Check the license of any text, code, or image you feed in or publish. Keep a record of how you used AI in creative work. Some creators and publishers now license their work for training, and others object, so expect the rules to keep changing.

Regulation And The EU AI Act

The European Union adopted the AI Act, Regulation (EU) 2024/1689, which entered into force in August 2024. It is the first broad law of its kind, and its obligations phase in over several years. Its central idea is to regulate by risk: the more a system can harm people, the stricter the rules.

At a high level the Act sorts systems into tiers. The details are long, so treat this as a map and read the official text or summaries for a real project.

Unacceptable risk: practices that are banned, such as social scoring by public authorities and certain manipulative or exploitative systems.
High risk: systems used in sensitive areas such as hiring, education, credit, essential services, and law enforcement. They must meet requirements for risk management, data quality, documentation, human oversight, accuracy, and security.
Limited risk: transparency duties, for example telling people they are talking to an AI and labeling AI-generated or manipulated content.
Minimal risk: most other systems, with no extra obligations.
General-purpose AI models: providers face duties such as technical documentation, information for downstream developers, and copyright policies, with extra duties for the most capable models.

EU AI Act Risk Tiers

Tier	Meaning	Example
Unacceptable	Banned	Social scoring by governments
High	Strict duties	CV screening, credit scoring
Limited	Transparency	Chatbots must say they are AI
Minimal	No new duties	Spam filters

Simplified overview, not legal advice. Source: Regulation (EU) 2024/1689

EU AI Act Key Dates

Date	What starts
1 Aug 2024	Act enters into force
2 Feb 2025	Bans on unacceptable practices
2 Aug 2025	Rules for general-purpose models
2 Aug 2026	Most remaining rules apply
2 Aug 2027	Some high risk product rules

Dates as published in the Regulation. Source: Regulation (EU) 2024/1689

Other Approaches And Responsible Use

Other regions take different paths. The United States relies on a mix of existing laws, agency guidance, and state laws. China has issued specific rules for areas such as recommendation algorithms and generative AI. Many countries publish voluntary frameworks, such as the NIST AI Risk Management Framework you saw in week 9, and the OECD AI Principles. If you serve users in several places, you often follow the strictest rule that applies.

Beyond the law, individuals and teams need a responsible-use habit. A short checklist helps, and you will use it in the capstone.

Purpose: is the benefit clear, and is AI the right tool?
People: who could be harmed or left out? Did you test with them?
Transparency: do users know they are using AI? Are sources shown?
Oversight: can a human review or reverse important decisions?
Data: do you have the right to use it, and have you minimized it?
Accountability: who owns the system and who handles complaints?
Environment: choose the smallest adequate model, since larger models use more energy.

Responsible Use Checklist

Name the purpose
Check data and rights
Rate the risk
Keep a human in charge
Record and review

Run before launch.

Key Terms

Task-based view of work: Analyzing a job as a bundle of tasks, some of which AI can assist.
Offloading: Handing mental effort to a tool so that you build less skill yourself.
Liability: Legal responsibility for harm caused by a decision or product.
Fair use: A US legal doctrine that allows some uses of copyrighted work without permission, judged case by case.
Human authorship: The requirement in US copyright practice that a person made the creative contribution to a work.
EU AI Act: A European law that regulates AI systems by risk level.
High-risk system: Under the AI Act, a system in a sensitive area that must meet strict requirements.
Transparency duty: A rule to tell people when they interact with AI or see AI-generated content.
Human oversight: A design in which a person can review, correct, or override an AI system.

Exercise: Responsible-Use Review Of A Real AI Use

Pick one AI use that affects people, such as a hiring screener, a homework helper, a medical chatbot, or an image generator in marketing.
Write two sentences on its purpose and who uses it.
Using the EU AI Act tiers on the slides, decide which tier fits and justify your choice in three sentences.
Go through the seven-item responsible-use checklist and give a score of strong, partial, or weak for each item with one line of evidence.
List the two biggest risks and one design change for each (for example, human review, a disclosure label, or less data).
Interview one person from a different field about how AI changes their work, and write down two tasks they expect to change.
Share a half-page summary with a partner and discuss where you disagree.