2022 was the year AI talk started turning into law

Of the 127 countries analyzed, only 31 have passed a single AI-related bill since 2016, and between them they account for 123 laws in total. Annual output climbed from just one law in 2016 to 37 in 2022. In 2022 the United States led with 9 laws, followed by Spain (5) and the Philippines (4); Andorra, Belgium, Italy, Portugal, Russia and the United Kingdom passed 2 each. Measured cumulatively across 2016–22 the ordering changes: the United States leads with 22, then Portugal (13), Spain (10), Italy and Russia (9 each).

What an “AI law” actually looks like

Almost none of these are AI acts in the sense the phrase now implies. The AI Index sampled five of 2022’s laws and found the words “artificial intelligence” buried inside statutes about creative industries, national security, schools, anti-discrimination and civil-service training:

• Kyrgyz Republic — About the Creative Industries Park: sets the legal status, management and operating procedures of a park created to accelerate creative industries, artificial intelligence among them.
• Latvia — Amendments to the National Security Law: places restrictions on commercial companies, associations and foundations important to national security, including any commercial company that develops artificial intelligence.
• Philippines — Second Congressional Commission on Education (EDCOM II) Act: creates a commission to review and reform education in the face of the Fourth Industrial Revolution, characterized in part by the rapid development of artificial intelligence.
• Spain — Right to equal treatment and non-discrimination: requires that AI algorithms involved in public administrations’ decision-making take bias-minimization criteria, transparency and accountability into account whenever technically feasible.
• United States — AI Training Act: directs the Office of Management and Budget to provide an AI training program for the acquisition workforce of executive agencies, so that buyers understand AI’s capabilities and risks.

A caveat the chapter puts in a footnote

The AI Index warns that this count probably undercounts real activity, because a single large bill can contain multiple AI-related sub-bills — the CHIPS and Science Act, passed by the United States in 2022, is the example it gives. Coverage is also incomplete: publicly accessible legislative databases simply do not exist for some countries.

In 2015 exactly one AI-related federal bill was proposed. By 2021 that had reached 134, falling back to 88 in 2022. Over the same stretch the number of bills actually passed sat unmoved at 3 a year for four consecutive years — and then went to 9 in 2022. As a proportion, 2% of federal AI bills became law in 2021; 10% did in 2022.

The states convert at three times the federal rate

State legislatures show the same pattern, further along. Proposals rose from 5 in 2015 to 60 in 2022; passage rose from 1 bill (16% of that year’s proposals) to 21 (35%). California led 2022 with 5 laws, then Maryland (3), and Colorado, New Jersey and Washington with 2 each. Cumulatively from 2016 to 2022, Maryland leads with 7, ahead of California (6), Massachusetts and Washington (5 each), and Illinois, Utah and Vermont (3 each). Roughly half the states have still never passed one.

Five state laws, five completely different ideas of what AI policy is

• Alabama — Artificial Intelligence, Limit the Use of Facial Recognition, to Ensure Match Results Is Not the Only Basis for Arrest: bars state and local law enforcement from using a facial recognition match as the sole basis for an arrest or for establishing probable cause in a criminal investigation.
• California — Budget Act of 2022: an appropriations provision allocating $1,300,000 to California State University, Sacramento to improve its campus childcare center, including the development of an artificial intelligence mixed-reality classroom.
• Maryland — Conservation Finance Act: directs the Department of Natural Resources to study whether digital tools including AI and machine learning can contribute to Chesapeake Bay restoration and climate solutions.
• New Jersey — 21st Century Integrated Digital Experience Act: requires the chief technology officer to evaluate annually whether state agencies could use AI and machine learning to deliver public services.
• Vermont — An Act Relating to the Use and Oversight of Artificial Intelligence in State Government: creates a Division of Artificial Intelligence inside the Agency of Digital Services to review every AI system the state develops, employs or procures, and to propose a state code of ethics.

Across the parliamentary and governmental proceedings of 81 countries, AI mentions registered a small decrease in 2022 — from 1,547 to 1,340 — in a run that has still grown nearly 6.5 times since 2016. Only 46 of the 81 countries recorded a single mention all year. Spain topped 2022 with 273 mentions, followed by Canada (211), the United Kingdom (146) and the United States (138). Over the full 2016–22 period the United Kingdom dominates with 1,092 mentions, ahead of Spain (832), the United States (626), Japan (511) and Hong Kong (478); 62 of the 81 countries have at least one.

What legislators actually argue about

The chapter samples the debates themselves, and the range is the point — the same two words are doing very different work in different chambers:

• United Kingdom — Dame Angela Eagle asked the Commons, during the Financial Services and Markets Bill, how open regulators would be about the way AI is applied and whether it might evolve so as to embed discrimination that excludes certain people.
• Japan — Kohei Otsuka told the House of Councillors’ Commission on the Constitution that as AI and internet technology fuse, human rights guarantees need updating to cover individual scoring, discrimination, and online advertising that unfairly influences how citizens vote.
• Zambia — Minister Collins Nzovu told the National Assembly that, with UN financial support, the government is turning to AI to improve the quality and accuracy of weather forecasting.
• Australia — Industry and Science Minister Ed Husic, speaking to the National Reconstruction Fund Corporation Bill 2022, argued the country should build on its “undeniable expertise” in quantum technologies, robotics and artificial intelligence.
• Brazil — Congressman Gustavo Fruet invoked The Fourth Industrial Revolution in a debate on granting telework to parents of young children, on which professions 5G, the Internet of Things and AI would make extinct and which would demand more qualifications.

Inside the US Congress: committee reports and policy papers

Mentions of AI in the committee reports of the House and Senate rose sharply from the 115th session onward and reached 73 in the 117th (2021–22), the highest in a series that begins in 2001. Appropriations led the House with 20 mentions, ahead of Science, Space, and Technology (9) and Rules (5); in the Senate, Homeland Security and Governmental Affairs led with 8, ahead of Appropriations (3). Outside government, the AI Index tracked 55 US-based organizations — 19 think tanks and policy institutes, 14 university institutes and research programs, 9 civil society bodies, 9 industry and consultancy organizations and 4 government agencies — which published 284 AI policy papers in 2022, 3.2 times the 2018 figure. The leading primary topics were industry and regulation (107), innovation and technology (90) and government and public administration (82). Privacy, safety, and security, the single most reported topic in 2021, slipped to fourth (59).

Canada launched the world’s first national AI strategy in March 2017, alongside China and Finland that year. Since then 62 strategies have been released in total, with the peak arriving in 2019, when 26 countries published one — Argentina, Denmark, Estonia, Japan, Kenya, Qatar, Singapore, the United Arab Emirates and the United States among them. Fourteen followed in 2020 and ten in 2021. By 2022 only two countries released a new national AI strategy: Italy and Thailand. A further seven pledged to develop one in 2021 (Armenia, Bahrain, Cuba, Iceland, Morocco, New Zealand, Oman) and seven more in 2022 (Azerbaijan, Belgium, Benin, Israel, Jordan, Nigeria, Uzbekistan).

Two federal budget lines, moving in opposite directions

Nondefense US government agencies allocated $1.7 billion to AI R&D in fiscal year 2022 — a slight decline from FY21’s $1.75 billion, but a 208.9% increase on the $0.56 billion of FY18. A larger $1.84 billion was requested for FY23. The Department of Defense line went the other way and then reversed: its request for nonclassified AI-specific research, development, test and evaluation slid from $1.00 billion in FY20 to $0.93 billion in FY21 and $0.87 billion in FY22, before jumping to $1.10 billion for FY23 — a 26.4% increase. Neither figure includes classified AI R&D by defense and intelligence agencies, which the underlying report does not cover.

Procurement is where the growth is

• Total US government spending on AI rose from $2.7 billion in 2021 to $3.3 billion in 2022, and has grown roughly 2.5 times since 2017, when it stood at $1.29 billion.
• By segment in 2022: decision science took $1.19 billion (up 18% year over year) and computer vision $0.82 billion (up 55%), while autonomy reached $0.69 billion (up 33%). Machine learning slipped 5% to $0.41 billion and natural language processing fell 19% to $0.17 billion.
• By award type in FY22, prime contracts made up 62.5% of the value, grants 34.9% and other transaction authority awards 2.6% — $2.05 billion, $1.15 billion and $0.09 billion respectively. Against FY21 the contract share held steady while the grant share rose.