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.
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 The book describes this as a payroll study that is not peer reviewed; keep that caveat.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.
- Ireland. Copyright and Related Rights Act 2000, section 21(f) (opens in a new tab) (the author of a computer-generated work is "the person by whom the arrangements necessary for the creation of the work are undertaken") and section 30: copyright in a computer-generated work lasts 70 years from first lawful publication. The enacted text was read on 2 November 2026; the revised version on the Law Reform Commission's site would not display that day, and was last read on 24 October 2026, updated to 9 September 2026.
- UK. Copyright, Designs and Patents Act 1988, section 9(3) (opens in a new tab) and section 12(7) (50 years). On 2 November 2026 legislation.gov.uk showed section 9 with "currently no known outstanding effects", so the rule still stands. The government's report on copyright and AI (opens in a new tab), published 18 March 2026, proposes "that this specific type of protection should be removed" while keeping protection for works made with AI assistance. As of 2 November 2026 that is still only a proposal: no bill to change it had been found.
- US. Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability (opens in a new tab), January 2025. The 2025 appeal court opinion in Thaler v. Perlmutter (opens in a new tab), No. 23-5233 (decided 18 March 2025, rehearing denied 12 May 2025). On 2 March 2026 the Supreme Court denied certiorari, No. 25-449 (opens in a new tab).
- EU. European Parliament resolution of 10 March 2026 (opens in a new tab) on copyright and generative AI (non-binding).
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.
Sources
- 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
- Data Centres Metered Electricity Consumption 2025 (opens in a new tab) · Central Statistics Office, 2026
- 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
- 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
- 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
- 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
- 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
- The future of employment: How susceptible are jobs to computerisation? (opens in a new tab) · Frey, Carl Benedikt and Osborne, Michael A., 2017
- 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
- The 2026 AI Index Report (opens in a new tab) · Stanford Institute for Human-Centered Artificial Intelligence, 2026
- Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI (opens in a new tab) · Humlum, Anders and Vestergaard, Emilie, 2025
- Energy and AI (opens in a new tab) · International Energy Agency, 2025
- Key Questions on Energy and AI (opens in a new tab) · International Energy Agency, 2026
- Copyright and Related Rights Act 2000 (No. 28 of 2000), sections 2, 21 and 30 (opens in a new tab) · Ireland, 2000
- Making AI Less 'Thirsty' (opens in a new tab) · Li, Pengfei and Yang, Jianyi and Islam, Mohammad A. and Ren, Shaolei, 2025
- Power Hungry Processing: Watts Driving the Cost of AI Deployment? (opens in a new tab) · Luccioni, Sasha and Jernite, Yacine and Strubell, Emma, 2024
- Thinking carefully before adopting agentic AI (opens in a new tab) · Martin R and Kate S, 2026
- 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
- Thaler v. Perlmutter, No. 25-449: docket (opens in a new tab) · Supreme Court of the United States, 2026
- Thaler v. Perlmutter, No. 23-5233 (opens in a new tab) · United States Court of Appeals for the District of Columbia Circuit, 2025
- Copyright, Designs and Patents Act 1988 (c. 48), sections 9, 12 and 178 (opens in a new tab) · United Kingdom, 1988
- Copyright and Artificial Intelligence, Part 2: Copyrightability (opens in a new tab) · United States Copyright Office, 2025