Where the principles came from

Lesson 1 of 5 in Core Ethics Principles and Trustworthy AI.

Between 2016 and 2019 the world produced AI ethics documents at industrial pace: the Asilomar AI Principles (2017, signed by thousands of researchers), the Montreal Declaration (2018), IEEE’s Ethically Aligned Design, dozens of corporate charters, and national strategies by the shelf-load. Researchers who catalogued them — Jobin, Ienca and Vayena counted 84 documents by 2019; Fjeld and colleagues at Harvard mapped 36 more — noticed two things at once.

First, chaos: no two documents used the same list, the same definitions, or the same emphasis. The field called it principle proliferation. Second, hiding inside the chaos, convergence: the same handful of themes kept surfacing under different names — fairness, transparency, accountability, privacy, safety, human control, societal benefit. The vocabulary differed; the underlying worries didn’t.

Two documents turned that convergence into official consensus, and you need to know both because statutes and frameworks cite them constantly.

The OECD AI Principles (2019, revised 2024) were the first intergovernmental AI standard — adopted by all OECD members, endorsed by the G20, and adhered to by dozens of governments beyond. Five values-based principles (inclusive growth and well-being; human rights and democratic values, including fairness and privacy; transparency and explainability; robustness, security and safety; accountability) plus five recommendations to governments. The 2024 revision updated them for generative AI — adding attention to mis/disinformation and clarifying safety expectations. When the EU AI Act needed a definition of "AI system", it took the OECD’s.

The UNESCO Recommendation on the Ethics of AI (2021) went wider: adopted by all 193 UNESCO member states — including China and (at the time) Russia — making it the closest thing to a global ethics baseline. It is more expansive than the OECD text: human dignity and human rights as the foundation, strong language on environment and ecosystem flourishing, gender equality as a named priority, and concrete policy tools — including a Readiness Assessment Methodology and Ethical Impact Assessment that member states actually run.

Keep the division of labour straight: these documents are soft law — they bind no one. Their power is that they set the menu: when legislatures and standards bodies later wrote binding rules, they drew the content from this consensus.

Key terms: transparency, explainability, human oversight, trustworthy AI characteristics, soft law

The principles era and what it fed into

  • 1980-09-23 — OECD Privacy Guidelines:

    The first international data-protection framework — the template for cross-border governance of information technology, four decades before the same body wrote AI principles.

  • 2017-01-05 — Asilomar AI Principles:

    23 principles signed by thousands of researchers — the starting gun for the “principle proliferation” era (80+ AI ethics codes within three years).

  • 2019-04-08 — EU HLEG Ethics Guidelines for Trustworthy AI:

    Seven requirements — human oversight, robustness, privacy, transparency, fairness, wellbeing, accountability — that will resurface, hardened, as Articles 9–15 of the AI Act.

  • 2019-05-22 — OECD AI Principles adopted:

    The first intergovernmental AI standard — 40+ adherents including the US and (via G20) China. Its AI-system definition becomes the shared vocabulary of the EU AI Act and US law.

  • 2021-11-23 — UNESCO Recommendation on the Ethics of AI:

    Adopted by 193 countries — the broadest AI ethics instrument on Earth, with readiness-assessment machinery for implementation.

  • 2024-05-03 — OECD AI Principles updated:

    The 2019 principles get a generative-AI refresh — including the revised AI-system definition that laws worldwide now cite.

Interactive checkpoint quiz (2 questions) — open this page in a browser to take it.