The awakening and the principles gold rush (2013–2019)

Lesson 2 of 5 in From Asimov to the AI Act: A History of AI Governance.

Deep learning’s 2012 breakthrough (AlexNet) put learned models into products, and within four years the harms became measurable. The years 2013–2018 are the algorithmic-accountability awakening — the period when journalists and researchers proved, case by named case, that deployed systems were failing specific people in patterned ways:

  • 2013 — the Campaign to Stop Killer Robots launches; by 2014 the UN’s Convention on Certain Conventional Weapons opens expert talks on lethal autonomous weapons — the first standing intergovernmental AI-governance forum, still unresolved a decade later.
  • 2016 — ProPublica publishes “Machine Bias”: the COMPAS recidivism score used in US sentencing had roughly double the false-positive rate for Black defendants. The ensuing academic fight over which fairness definition COMPAS violated produced the impossibility theorems you met in the ethics module. The same year, Cathy O’Neil’s Weapons of Math Destruction gave the public a vocabulary.
  • 2018 — Gender Shades (Buolamwini & Gebru) showed commercial face analysis failing darkest-skinned women at up to 34.7% error versus under 1% for light-skinned men; Cambridge Analytica showed data-driven persuasion at electoral scale. Research communities organised (FAT/ML becoming the FAccT conference).
  • 2019 — San Francisco enacts the first municipal government facial-recognition ban. Regulation had begun — bottom-up, city by city.

Scandals and answers, 2013–2019

  • 2016-05-23 — ProPublica publishes “Machine Bias” (COMPAS):

    Black defendants who didn’t reoffend were flagged high-risk at nearly twice the white rate; the vendor showed equal calibration. Both were right — the fairness impossibility theorem goes mainstream.

  • 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).

  • 2018-02-01 — Gender Shades study:

    Buolamwini and Gebru show commercial face analysis erring up to 34% on darker-skinned women vs under 1% on lighter-skinned men — the case for disaggregated testing, made unignorable.

  • 2018-03-18 — Uber ATG test vehicle kills Elaine Herzberg:

    The first pedestrian death by an autonomous vehicle — a distracted safety driver behind an automated system becomes the canonical human-oversight failure case.

  • 2018-10-01 — Amazon scraps its AI recruiting tool:

    Trained on ten years of male-dominated hiring, the model learned to downgrade résumés mentioning “women’s”. Killed before deployment — the textbook label-bias parable.

  • 2019-05-14 — San Francisco bans government facial recognition:

    The first big-city ban — the beginning of the US municipal wave against biometric surveillance.

  • 2020-01-18 — Clearview AI exposed:

    Three billion faces scraped without consent for police search. The backlash writes itself into law: EU AI Act Art 5 bans untargeted face-scraping outright.

  • 2021-01-15 — Dutch government falls over the childcare-benefits scandal:

    An algorithmic fraud system with nationality as a risk factor ruined tens of thousands of families. A cabinet resigns — the starkest proof that algorithmic harm is political.

Institutions responded with the tool they could agree on fastest: principles. Between 2016 and 2019, more than a hundred AI-ethics declarations appeared — the principles era. The landmarks: the Asilomar AI Principles (2017, from the research community), the Montreal Declaration (2018, participatory and rights-based), IEEE’s Ethically Aligned Design, the EU High-Level Expert Group’s Ethics Guidelines for Trustworthy AI (2019, direct ancestor of the AI Act’s requirements), and above all the OECD AI Principles (May 2019) — the first intergovernmental AI standard, endorsed within a month by the G20, and the document whose definitions still anchor the field.

Researchers who mapped the flood (Jobin et al. counted 84 documents; Fjeld et al. at Harvard mapped 36) found striking convergence: transparency, fairness, non-maleficence, responsibility, privacy appear almost everywhere. The field had reached consensus on what matters — and possessed not one mechanism to make anyone do it. Google’s external AI ethics board, announced March 2019, dissolved in nine days. The gap between declared principle and governed practice became the defining problem of the next era.

Key terms: soft law, ethics washing, principle proliferation, incident-driven regulation

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