Skip to content

Chapter 20

Fairness, work, ownership and the planet


The chapter in a minute

  • A system learns whatever pattern sits in its examples, fair or not. When one makes decisions about people, ask what it was built to predict and who checked it.
  • Most big job numbers are exposure estimates, not counts of jobs lost. Find the source behind the next jobs headline you see.
  • Who owns what AI makes is still being settled, and the answer differs by country. Keep a note of what you added yourself.
  • The electricity and water behind AI are real and growing. One everyday text question uses little electricity, but nobody can yet say how much water it takes. Pick the simple tool for the simple job, and go easy on video.
  • Whatever you send, post or share goes out under your name. Give it the Check it test first.

Try it now

Read one headline twice

0 of 8 steps done

Step 1 of 8

Read the story once, the way you normally would. Write its headline at the top of your page.

Progress is saved in this browser only.

Try this

Try this, from the book

Pick one AI headline you see today and give yourself five minutes to find the source behind it. Look for a study, an official statistic or a report from an international body. If the story never says where its claim came from, that tells you something too!

Tools and settings, as of a date

The book points to the ‘Debates and sources’ section of the companion page. Here it is in full, with the date each detail was checked. Open it on the companion page

Chapter 20 points here: the full jobs figures and ownership positions. Access date 24 October 2026 unless stated; the law, energy and open items were read again on 2 November 2026.

Jobs

  • ILO exposure shares. About one worker in four worldwide (24%) is in a job with some exposure. 3.3% are in the highest group (4.7% of women's jobs, 2.4% of men's). Exposure is 11% in low-income and 34% in high-income countries. The 2025 estimates are lower than the 2023 ones. The ILO says exposure does not imply immediate automation of an entire occupation. Gmyrek and colleagues, "Generative AI and Jobs: A Refined Global Index of Occupational Exposure", ILO Working Paper 140, May 2025, DOI 10.54394/HETP0387 (opens in a new tab).
  • Young workers in exposed US jobs. A gap of 13%, 16% or 19% depending on the version, with the authors' caveats. Brynjolfsson and colleagues, "Canaries in the Coal Mine? (opens in a new tab)", August 2026 version, data through June 2026. The book describes this as a payroll study that is not peer reviewed; keep that caveat.
  • US unemployment by exposure group, 2022 to early 2025. Stanford HAI, 2026 AI Index (opens in a new tab), economy chapter (accessed 24 October 2026).
  • Denmark. Effects above 2% in earnings and hours were ruled out. Humlum and Vestergaard, NBER Working Paper 33777, May 2025, revised March 2026 (accessed 23 October 2026, read again 24 October 2026).
  • A US exposure estimate. Eloundou and colleagues, "GPTs are GPTs", Science 2024. Abstract read; note the authors' conflict of interest.
  • IMF exposure figures. Cazzaniga and colleagues, "Gen-AI: Artificial Intelligence and the Future of Work". The IMF pages refused automated access and the widely reported 40% and 60% figures were not found in the pages that could be read, so no figure is given here.
  • The 2013 study's method. Frey and Osborne, "The future of employment" (702 occupations; high risk means a probability above 0.7).
  • The Obermeyer figures. Obermeyer and colleagues (2019), "Dissecting racial bias in an algorithm used to manage the health of populations": 17.7% to 46.5%, and an 84% reduction. These are easily confused with another figure in chapter 20, so label them clearly.

Who owns what an assistant makes

General information, not legal advice. Treat ownership of anything made mostly by AI as uncertain. Checked: 2 November 2026.

The EU AI Act

Dates were already amended in 2026, by Regulation (EU) 2026/1744 of 8 July 2026. In the consolidated text of 27 July 2026 (opens in a new tab), still the latest on EUR-Lex on 2 November 2026, Article 113 says the Act applies generally from 2 August 2026, with these changes: the new bans on intimate-image and child-abuse systems from 2 December 2026; most high-risk rules from 2 December 2027 or 2 August 2028; and Articles 102 to 110 from 27 July 2026. Systems already on the market before 2 August 2026 have until 2 December 2026 to mark their content (Article 111(4)). The deepfake labelling duty has exceptions. The Act does not place its duties on people using AI purely in their own private lives (Article 2(10)). Checked: 2 November 2026.

Energy and water

  • Data centre electricity. 415 TWh in 2024; 485 TWh in 2025; about 950 TWh in 2030 on the central forecast, about 3% of world demand. A 2035 range of 700 to 1,700 TWh across four cases. AI-focused centres roughly triple. Regional shares in 2024: US 45%, China 25%, Europe 15%. IEA, Energy and AI (opens in a new tab) (10 April 2025), and Key Questions on Energy and AI (opens in a new tab) (16 April 2026). These are forecasts, refreshed each year. On 2 November 2026 the IEA's page showed no newer release.
  • How fast it grew in 2025. The 2026 report gives two figures for the same thing: the executive summary (page 9) says demand "grew by 17% in 2025", and page 17 says use "grew by more than 15% between 2024 and 2025", to "almost 500 TWh, or slightly more than 1.5% of global electricity demand". Both are the IEA's own words.
  • Ireland. Data centres used 23% of metered electricity in 2025, against 5% in 2015, with homes at 28%. That is 7,663 GWh, up from 6,973 GWh in 2024. CSO release (opens in a new tab). The release line on the page still reads "07 July 2025", but the statistician's comment on the same page dates it 7 July 2026, which fits 2025 data; 7 July 2026 is the date to use.
  • Water. The IEA's global estimate is about 560 billion litres a year, possibly about 1,200 billion by 2030, and it splits the total by cause. An academic estimate of 500 ml per 10 to 50 answers is for an older system. The company's own measurement is about 0.26 ml per prompt. Sources: IEA, Li and colleagues, "Making AI Less Thirsty", and Elsworth and colleagues, "Measuring the environmental impact of delivering AI at Google Scale (opens in a new tab)" (a preprint by the company's own staff, not peer reviewed).
  • Per-query energy. Fractions of a watt-hour for short text, about ten times more for a high-resolution image, and the companies' disclosed 0.24 to 0.34 Wh. These are company figures and falling fast.

Checked: 2 November 2026.

Further reading

The IEA reports, the ILO paper, the AI Index economy chapter, the CSO release and the 2019 Science paper by Obermeyer and colleagues. All are linked above.

From the book

A forecast is not a fact. An estimate of what could happen is not a report of what has happened.
Your First Rung, chapter 20

Sources

  1. Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence (opens in a new tab) · Brynjolfsson, Erik and Chandar, Bharat and Chen, Ruyu, 2026
  2. Data Centres Metered Electricity Consumption 2025 (opens in a new tab) · Central Statistics Office, 2026
  3. Report on Copyright and Artificial Intelligence (opens in a new tab) · Department for Science, Innovation and Technology and Department for Culture, Media and Sport and Intellectual Property Office, 2026
  4. Measuring the environmental impact of delivering AI at Google Scale (opens in a new tab) · Elsworth, Cooper and Huang, Keguo and Patterson, David and Schneider, Ian and Sedivy, Robert and Goodman, Savannah and Townsend, Ben and Ranganathan, Parthasarathy and Dean, Jeff and Vahdat, Amin and Gomes, Ben and Manyika, James, 2025
  5. Copyright and generative artificial intelligence: opportunities and challenges. European Parliament resolution of 10 March 2026 (2025/2058(INI)) (opens in a new tab) · European Parliament, 2026
  6. Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) (opens in a new tab) · European Parliament and Council of the European Union, 2024
  7. Regulation (EU) 2024/1689 (Artificial Intelligence Act), consolidated text of 27 July 2026, as amended by Regulation (EU) 2026/1744 of 8 July 2026 (opens in a new tab) · European Union, 2026
  8. The future of employment: How susceptible are jobs to computerisation? (opens in a new tab) · Frey, Carl Benedikt and Osborne, Michael A., 2017
  9. Generative AI and Jobs: A Refined Global Index of Occupational Exposure (opens in a new tab) · Gmyrek, Pawe\l and Berg, Janine and Kami\'nski, Karol and Konopczy\'nski, Filip and \Ladna, Agnieszka and Nafradi, Balint and Ros\laniec, Konrad and Troszy\'nski, Marek, 2025
  10. The 2026 AI Index Report (opens in a new tab) · Stanford Institute for Human-Centered Artificial Intelligence, 2026
  11. Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI (opens in a new tab) · Humlum, Anders and Vestergaard, Emilie, 2025
  12. Energy and AI (opens in a new tab) · International Energy Agency, 2025
  13. Key Questions on Energy and AI (opens in a new tab) · International Energy Agency, 2026
  14. Copyright and Related Rights Act 2000 (No. 28 of 2000), sections 2, 21 and 30 (opens in a new tab) · Ireland, 2000
  15. Making AI Less 'Thirsty' (opens in a new tab) · Li, Pengfei and Yang, Jianyi and Islam, Mohammad A. and Ren, Shaolei, 2025
  16. Power Hungry Processing: Watts Driving the Cost of AI Deployment? (opens in a new tab) · Luccioni, Sasha and Jernite, Yacine and Strubell, Emma, 2024
  17. Thinking carefully before adopting agentic AI (opens in a new tab) · Martin R and Kate S, 2026
  18. Dissecting racial bias in an algorithm used to manage the health of populations (opens in a new tab) · Obermeyer, Ziad and Powers, Brian and Vogeli, Christine and Mullainathan, Sendhil, 2019
  19. Thaler v. Perlmutter, No. 25-449: docket (opens in a new tab) · Supreme Court of the United States, 2026
  20. Thaler v. Perlmutter, No. 23-5233 (opens in a new tab) · United States Court of Appeals for the District of Columbia Circuit, 2025
  21. Copyright, Designs and Patents Act 1988 (c. 48), sections 9, 12 and 178 (opens in a new tab) · United Kingdom, 1988
  22. Copyright and Artificial Intelligence, Part 2: Copyrightability (opens in a new tab) · United States Copyright Office, 2025